diff --git a/.github/workflows/auto-tag-on-release-pr-merge.yml b/.github/workflows/auto-tag-on-release-pr-merge.yml index f31d4b835f..3a090b3ebd 100644 --- a/.github/workflows/auto-tag-on-release-pr-merge.yml +++ b/.github/workflows/auto-tag-on-release-pr-merge.yml @@ -45,7 +45,7 @@ jobs: github.event.pull_request.head.repo.full_name == github.repository runs-on: ubuntu-latest steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: ref: ${{ github.event.pull_request.merge_commit_sha }} fetch-depth: 0 diff --git a/.github/workflows/benchmark-harbor.yml b/.github/workflows/benchmark-harbor.yml index 31efe933c5..6024f00575 100644 --- a/.github/workflows/benchmark-harbor.yml +++ b/.github/workflows/benchmark-harbor.yml @@ -20,7 +20,7 @@ jobs: runs-on: ubuntu-latest timeout-minutes: 15 steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5 with: python-version: "3.12" diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 453a15de99..60507182d5 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -28,7 +28,7 @@ jobs: web: ${{ steps.filter.outputs.web }} mobile: ${{ steps.filter.outputs.mobile }} steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 2 - uses: dorny/paths-filter@7b450fff21473bca461d4b92ce414b9d0420d706 # v4.0.2 @@ -96,7 +96,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2 with: @@ -117,7 +117,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - uses: rui314/setup-mold@9c9c13bf4c3f1adef0cc596abc155580bcb04444 # v1 - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2 @@ -139,7 +139,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 2 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 @@ -235,7 +235,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - name: Get pnpm store directory id: pnpm-cache @@ -318,7 +318,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 # Reuse the relay binaries and backend test archive when none of their # inputs changed (desktop-only PRs hit this every time). The key covers @@ -391,7 +391,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - name: Start integration services run: | @@ -580,7 +580,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - name: Install cargo-nextest uses: taiki-e/install-action@0fd46367812ee04360509b4169d9f659d6892bb2 # v2.79.15 @@ -692,6 +692,18 @@ jobs: --run-ignored ignored-only env: DATABASE_URL: postgres://buzz:${{ env.BUZZ_TEST_POSTGRES_PASSWORD }}@localhost:5432/buzz + - name: Workspace profile (kind:9033) gate tests + # Call-site integration for the 9033 authorization gate: open relay + # rosterless/steward transitions and the closed-relay admin/owner rule, + # against real Postgres. #[ignore]d in the default suite, selected + # explicitly here — see handlers::relay_admin::tests. + run: | + cargo nextest run \ + --archive-file target/ci/backend-integration-tests.tar.zst \ + -E 'package(buzz-relay) and test(/handlers::relay_admin::tests/)' \ + --run-ignored ignored-only + env: + DATABASE_URL: postgres://buzz:${{ env.BUZZ_TEST_POSTGRES_PASSWORD }}@localhost:5432/buzz - name: NIP-ER reminder e2e # Feature e2e for NIP-ER (Event Reminders, kind:30300): write-path # validation, author-only read filtering, and scheduler delivery against @@ -732,7 +744,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2 with: @@ -773,7 +785,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 2 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 @@ -808,7 +820,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 2 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 @@ -869,7 +881,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - name: Dependency policy run: cargo-deny check @@ -881,7 +893,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - name: Check for dead API token references in client code run: | # Fail if dead API token patterns reappear in desktop, mobile, docs, or config. @@ -910,7 +922,7 @@ jobs: - x86_64-unknown-linux-musl - aarch64-unknown-linux-musl steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2 with: @@ -947,7 +959,7 @@ jobs: env: TARGET: x86_64-pc-windows-msvc steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 # MSVC needs windows.h (aws-lc-sys et al.), so this runs on a real Windows # runner — hermit, used by the Linux jobs, does not provide MSVC. The # toolchain (1.95.0 + clippy via profile = default) comes from the @@ -1028,7 +1040,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2 with: @@ -1042,6 +1054,7 @@ jobs: mkdir -p desktop/src-tauri/binaries touch "desktop/src-tauri/binaries/buzz-acp-$TARGET" touch "desktop/src-tauri/binaries/buzz-agent-$TARGET" + touch "desktop/src-tauri/binaries/buzz-backend-kubernetes-$TARGET" touch "desktop/src-tauri/binaries/buzz-dev-mcp-$TARGET" touch "desktop/src-tauri/binaries/git-credential-nostr-$TARGET" touch "desktop/src-tauri/binaries/buzz-$TARGET" diff --git a/.github/workflows/docker.yml b/.github/workflows/docker.yml index 52f21b28bc..564cd74e9d 100644 --- a/.github/workflows/docker.yml +++ b/.github/workflows/docker.yml @@ -101,7 +101,7 @@ jobs: steps: - name: Checkout - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 0 persist-credentials: false @@ -359,7 +359,7 @@ jobs: arch: arm64 steps: - name: Checkout - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 0 persist-credentials: false diff --git a/.github/workflows/helm-chart.yml b/.github/workflows/helm-chart.yml index e3d443d9f3..7118d16708 100644 --- a/.github/workflows/helm-chart.yml +++ b/.github/workflows/helm-chart.yml @@ -59,7 +59,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4 + - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1 with: # On chart-tag rescue dispatch, lint/render the tagged commit that the # publish job will package, not whatever `main` is when the dispatch @@ -119,7 +119,7 @@ jobs: permissions: contents: read steps: - - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4 + - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1 with: fetch-depth: 0 @@ -166,7 +166,7 @@ jobs: packages: write # push the chart to GHCR steps: - name: Checkout - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4 + uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1 with: # On the rescue dispatch, build the tagged commit (github.ref is # `main` there); on a tag push, the default ref is already the tag. diff --git a/.github/workflows/linux-canary.yml b/.github/workflows/linux-canary.yml index 9806443378..e1625f4ec8 100644 --- a/.github/workflows/linux-canary.yml +++ b/.github/workflows/linux-canary.yml @@ -166,11 +166,11 @@ jobs: - name: Build sidecars run: | - cargo build --release -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli + cargo build --release -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli ./scripts/bundle-sidecars.sh - name: Build Linux Tauri app - run: cd desktop && pnpm tauri build --ci --bundles deb,appimage --config src-tauri/tauri.canary.conf.json + run: cd desktop && pnpm tauri build --ci --bundles deb,appimage --features mesh-llm --config src-tauri/tauri.canary.conf.json env: CMAKE_POLICY_VERSION_MINIMUM: "3.5" diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml index 7d5f3fbf40..9da067b74e 100644 --- a/.github/workflows/release.yml +++ b/.github/workflows/release.yml @@ -36,7 +36,7 @@ jobs: exit 1 fi - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: fetch-depth: 0 persist-credentials: false @@ -64,7 +64,7 @@ jobs: env: VERSION: ${{ needs.setup.outputs.version }} steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: ref: ${{ needs.setup.outputs.source_sha }} fetch-depth: 0 @@ -91,7 +91,7 @@ jobs: - name: Build sidecars run: | - cargo build --release -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli + cargo build --release -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli ./scripts/bundle-sidecars.sh # Mesh rev derived from Cargo.lock (no lockstep edit on dep bump); cache key tracks it. @@ -278,7 +278,7 @@ jobs: VERSION: ${{ needs.setup.outputs.version }} TARGET: x86_64-apple-darwin steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: ref: ${{ needs.setup.outputs.source_sha }} fetch-depth: 0 @@ -308,7 +308,7 @@ jobs: - name: Build sidecars run: | - cargo build --release --target "$TARGET" -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli + cargo build --release --target "$TARGET" -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli ./scripts/bundle-sidecars.sh "$TARGET" - name: Build unsigned Tauri app @@ -495,7 +495,7 @@ jobs: apt-get update apt-get install -y --no-install-recommends gh - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: ref: ${{ needs.setup.outputs.source_sha }} fetch-depth: 0 @@ -563,7 +563,7 @@ jobs: - name: Build sidecars run: | - cargo build --release -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli + cargo build --release -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli ./scripts/bundle-sidecars.sh - name: Generate release config @@ -573,7 +573,7 @@ jobs: BUZZ_UPDATER_ENDPOINT: https://github.com/block/buzz/releases/download/buzz-desktop-latest/latest.json - name: Build Linux Tauri app - run: cd desktop && pnpm tauri build --verbose --ci --bundles deb,appimage --config src-tauri/tauri.release.conf.json + run: cd desktop && pnpm tauri build --verbose --ci --bundles deb,appimage --features mesh-llm --config src-tauri/tauri.release.conf.json env: BUZZ_UPDATER_PUBLIC_KEY: ${{ secrets.BUZZ_UPDATER_PUBLIC_KEY || secrets.SPROUT_UPDATER_PUBLIC_KEY }} BUZZ_UPDATER_ENDPOINT: https://github.com/block/buzz/releases/download/buzz-desktop-latest/latest.json @@ -666,7 +666,7 @@ jobs: VERSION: ${{ needs.setup.outputs.version }} TARGET: x86_64-pc-windows-msvc steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: ref: ${{ needs.setup.outputs.source_sha }} fetch-depth: 0 @@ -795,7 +795,7 @@ jobs: VERSION: ${{ needs.setup.outputs.version }} GH_TOKEN: ${{ secrets.GITHUB_TOKEN }} steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 with: ref: ${{ needs.setup.outputs.source_sha }} fetch-depth: 0 diff --git a/.github/workflows/signed-macos-canary.yml b/.github/workflows/signed-macos-canary.yml index fb0656028a..0a3a513eef 100644 --- a/.github/workflows/signed-macos-canary.yml +++ b/.github/workflows/signed-macos-canary.yml @@ -93,7 +93,7 @@ jobs: - name: Build sidecars run: | - cargo build --release -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli + cargo build --release -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p git-credential-nostr -p buzz-cli ./scripts/bundle-sidecars.sh # Mesh rev derived from Cargo.lock (no lockstep edit on dep bump); cache key tracks it. diff --git a/.github/workflows/sprig-image.yml b/.github/workflows/sprig-image.yml new file mode 100644 index 0000000000..5d5e12ae0c --- /dev/null +++ b/.github/workflows/sprig-image.yml @@ -0,0 +1,233 @@ +name: Sprig image + +# Builds and publishes the public agent container image as +# ghcr.io/block/buzz-sprig — the digest-pinned box the Kubernetes backend +# deploys agents into (see Dockerfile.sprig and docs/remote-agents.md). +# +# Strategy mirrors docker.yml (the relay image): each architecture builds on +# its native runner, pushes to GHCR by digest, then a merge job stitches the +# per-arch digests into one multi-arch manifest and attests provenance. +# No QEMU emulation. +# +# Triggers: +# - push to main (paths-filtered) → :main + :sha-<7> +# - tag sprig-v* → semver family (shared with sprig.yml's +# binary release — one tag versions both) +# - pull_request (paths-filtered) → build only, no push +# - workflow_dispatch → manual publish at the current ref +# +# NOTE: the first push creates the GHCR package PRIVATE by default. An org +# admin must flip ghcr.io/block/buzz-sprig to public once (Package settings → +# Change visibility). Subsequent pushes keep the visibility. + +on: + push: + branches: [main] + tags: ["sprig-v[0-9]*"] + paths: + - "Dockerfile.sprig" + - "scripts/sprig-entrypoint.sh" + - ".github/workflows/sprig-image.yml" + - "Cargo.toml" + - "Cargo.lock" + - "rust-toolchain.toml" + - "crates/**" + pull_request: + paths: + - "Dockerfile.sprig" + - "scripts/sprig-entrypoint.sh" + - ".github/workflows/sprig-image.yml" + workflow_dispatch: {} + +concurrency: + group: sprig-image-${{ github.ref }} + cancel-in-progress: ${{ github.ref_type == 'branch' && github.event_name == 'pull_request' }} + +permissions: {} + +env: + # Single source of truth for the image name; override with the + # GHCR_SPRIG_IMAGE repo variable (same pattern as docker.yml). + IMAGE_NAME: ${{ vars.GHCR_SPRIG_IMAGE != '' && vars.GHCR_SPRIG_IMAGE || 'ghcr.io/block/buzz-sprig' }} + +jobs: + build: + name: Build (${{ matrix.platform }}) + runs-on: ${{ matrix.runner }} + timeout-minutes: 60 + permissions: + contents: read + packages: write + id-token: write + attestations: write + strategy: + fail-fast: false + matrix: + include: + - platform: linux/amd64 + runner: ubuntu-24.04 + arch: amd64 + - platform: linux/arm64 + runner: ubuntu-24.04-arm + arch: arm64 + + steps: + - name: Checkout + uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 + with: + persist-credentials: false + + - name: Set up Docker Buildx + uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5 # v4.1.0 + with: + # Same OOM cap as docker.yml — Rust compiles blow the 7GB runner + # at buildkit's default parallelism of 4. + buildkitd-config-inline: | + [worker.oci] + max-parallelism = 2 + + - name: Log in to GHCR + if: github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository + uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee # v4.2.0 + with: + registry: ghcr.io + username: ${{ github.repository_owner }} + password: ${{ secrets.GITHUB_TOKEN }} + + - name: Extract metadata + id: meta + uses: docker/metadata-action@80c7e94dd9b9319bd5eb7a0e0fe9291e23a2a2e9 # v6.1.0 + with: + images: ${{ env.IMAGE_NAME }} + # match=^sprig-v(.*)$ strips the tag prefix for the semver parser, + # exactly as docker.yml does for relay-v. :latest comes from + # flavor.latest=auto — stable semver only, never main pushes. + tags: | + type=ref,event=branch + type=sha,prefix=sha-,format=short + type=semver,pattern={{version}},match=^sprig-v(.*)$ + type=semver,pattern={{major}}.{{minor}},match=^sprig-v(.*)$ + labels: | + org.opencontainers.image.title=Buzz Sprig + org.opencontainers.image.description=Agent runtime image for Buzz remote agents (buzz-acp multicall + git + curl) + org.opencontainers.image.licenses=Apache-2.0 + + - name: Build and push by digest + id: build + uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf # v7.2.0 + with: + context: . + file: ./Dockerfile.sprig + platforms: ${{ matrix.platform }} + labels: ${{ steps.meta.outputs.labels }} + outputs: type=image,name=${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=${{ github.event_name != 'pull_request' }} + cache-from: | + type=registry,ref=${{ env.IMAGE_NAME }}-buildcache:${{ matrix.arch }} + cache-to: | + ${{ (github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository) && format('type=registry,ref={0}-buildcache:{1},mode=max,compression=zstd', env.IMAGE_NAME, matrix.arch) || '' }} + + - name: Export digest + if: github.event_name != 'pull_request' + env: + DIGEST: ${{ steps.build.outputs.digest }} + run: | + mkdir -p /tmp/digests + touch "/tmp/digests/${DIGEST#sha256:}" + + - name: Upload digest + if: github.event_name != 'pull_request' + uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1 + with: + name: sprig-digest-${{ matrix.arch }} + path: /tmp/digests/* + if-no-files-found: error + retention-days: 1 + + merge: + name: Merge multi-arch manifest + if: github.event_name != 'pull_request' + runs-on: ubuntu-24.04 + needs: build + timeout-minutes: 15 + permissions: + contents: read + packages: write + id-token: write + attestations: write + + steps: + - name: Download per-arch digests + uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1 + with: + path: /tmp/digests + pattern: sprig-digest-* + merge-multiple: true + + - name: Set up Docker Buildx + uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5 # v4.1.0 + + - name: Log in to GHCR + uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee # v4.2.0 + with: + registry: ghcr.io + username: ${{ github.repository_owner }} + password: ${{ secrets.GITHUB_TOKEN }} + + - name: Extract metadata + id: meta + uses: docker/metadata-action@80c7e94dd9b9319bd5eb7a0e0fe9291e23a2a2e9 # v6.1.0 + with: + images: ${{ env.IMAGE_NAME }} + # Must mirror the build job's tag matrix exactly (see docker.yml). + flavor: | + latest=auto + tags: | + type=ref,event=branch + type=sha,prefix=sha-,format=short + type=semver,pattern={{version}},match=^sprig-v(.*)$ + type=semver,pattern={{major}}.{{minor}},match=^sprig-v(.*)$ + + - name: Create and push manifest list + id: manifest + working-directory: /tmp/digests + env: + IMAGE_NAME: ${{ env.IMAGE_NAME }} + META_TAGS: ${{ steps.meta.outputs.tags }} + run: | + set -euo pipefail + tags=() + while IFS= read -r tag; do + [ -n "$tag" ] && tags+=("-t" "$tag") + done <<< "$META_TAGS" + + digests=() + for digest in *; do + digests+=("${IMAGE_NAME}@sha256:${digest}") + done + + docker buildx imagetools create "${tags[@]}" "${digests[@]}" + + first_tag=$(echo "$META_TAGS" | head -n1) + merged_digest=$(docker buildx imagetools inspect "$first_tag" \ + --format '{{json .Manifest}}' | jq -r '.digest') + echo "digest=${merged_digest}" >> "$GITHUB_OUTPUT" + + - name: Attest provenance for the merged image + # Verify with: gh attestation verify oci://ghcr.io/block/buzz-sprig: --owner block + uses: actions/attest-build-provenance@0f67c3f4856b2e3261c31976d6725780e5e4c373 # v4.1.1 + with: + subject-name: ${{ env.IMAGE_NAME }} + subject-digest: ${{ steps.manifest.outputs.digest }} + push-to-registry: true + + - name: Summary + env: + IMAGE_NAME: ${{ env.IMAGE_NAME }} + DIGEST: ${{ steps.manifest.outputs.digest }} + run: | + { + echo "### Sprig image published" + echo '```' + echo "${IMAGE_NAME}@${DIGEST}" + echo '```' + } >> "$GITHUB_STEP_SUMMARY" diff --git a/.github/workflows/sprig.yml b/.github/workflows/sprig.yml index b2dab3583f..5e50808b3b 100644 --- a/.github/workflows/sprig.yml +++ b/.github/workflows/sprig.yml @@ -42,7 +42,7 @@ jobs: - x86_64-unknown-linux-musl - aarch64-unknown-linux-musl steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1 @@ -116,7 +116,7 @@ jobs: permissions: contents: write steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - name: Download all Sprig artifacts uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1 @@ -154,7 +154,7 @@ jobs: permissions: contents: write steps: - - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6 + - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3 - name: Download all Sprig artifacts uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1 diff --git a/.release/desktop-candidate.json b/.release/desktop-candidate.json index 1bf2efb66b..1ba64765ff 100644 --- a/.release/desktop-candidate.json +++ b/.release/desktop-candidate.json @@ -1,8 +1,8 @@ { "schema": 1, - "version": "0.5.3", - "base_sha": "54c8ef30a9bb9c59a4415a8a7ee84c7c5454b48a", - "previous_tag": "v0.5.2", - "tag": "desktop-v0.5.3", - "commit_count": 58 + "version": "0.5.4", + "base_sha": "6de85fe31d781122756aecf954bae7d357a56b9a", + "previous_tag": "desktop-v0.5.3", + "tag": "desktop-v0.5.4", + "commit_count": 40 } diff --git a/AGENTS.md b/AGENTS.md index 7ff0eb4d47..7cd43061b2 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -13,14 +13,14 @@ Buzz spans five repos. This one (`block/buzz`) is the OSS source for the relay, | Repo | Purpose | |------|---------| | [block/buzz](https://github.com/block/buzz) | OSS source — relay, desktop app, mobile app, CLI, agent harness | -| [squareup/sprout-releases](https://github.com/squareup/sprout-releases) | Buildkite pipeline producing Block-signed macOS + iOS builds with `-block` version suffix | +| [squareup/buzz-releases](https://github.com/squareup/buzz-releases) | Buildkite pipelines producing Block-signed macOS + iOS builds with `-block` desktop version suffix | | [squareup/sprout-oss](https://github.com/squareup/sprout-oss) | CI pipeline building the relay Docker image and pushing to internal ECR | | [squareup/block-coder-tf-stacks](https://github.com/squareup/block-coder-tf-stacks) | Terraform + ArgoCD deploying the relay to the staging Kubernetes cluster | | [squareup/sprout-backend-blox](https://github.com/squareup/sprout-backend-blox) | Desktop backend provider script connecting Blox workstation agents to the relay | ``` block/buzz (source) - ├─► sprout-releases (desktop + mobile builds → Artifactory, GitHub, Mobile Releases) + ├─► buzz-releases (desktop + mobile builds → Artifactory, GitHub, Mobile Releases) ├─► sprout-oss (relay Docker image → ECR) │ └─► block-coder-tf-stacks (Helm chart → ArgoCD → staging cluster) └─── sprout-backend-blox (Blox compute provider for Desktop agent launch) @@ -145,6 +145,10 @@ first, then implement handling in the relay. **Channel scoping**: Channels use `h` tags (NIP-29 group tag), not `e` tags. Filters and queries must scope to `h` tags when operating within a channel. +This applies to events *inside* a channel. Addressable events that describe a +channel carry its id in their `d` tag instead: kind:39000 (metadata), +kind:39001, kind:39002 (membership). `get_channels` resolves a user's channels +from the `d` tag of their kind:39002 events, not from `h`. **Agent-facing operations go in `buzz-cli`**: New agent-facing features belong in `buzz-cli` — add a subcommand there first, then wire the REST/WebSocket call in `client.rs`. `buzz-dev-mcp` (shell + file tools for `buzz-agent`) is separate. diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md index 5c8e263a2a..892082d96c 100644 --- a/ARCHITECTURE.md +++ b/ARCHITECTURE.md @@ -139,7 +139,7 @@ The `kind` integer is the only dispatch switch. The relay routes, stores, and fa | 46001–46012 | KIND_WORKFLOW_* | Workflow execution events | | 20001 | KIND_PRESENCE_UPDATE | Ephemeral presence heartbeat | -`buzz-core` defines all 81 kinds as `pub const KIND_*: u32` and exports `ALL_KINDS: &[u32]`. Kinds are `u32` (NIP-01 specifies unsigned integer; `u32` covers the full range). Buzz uses both standard Nostr kinds (e.g., kind 7 for reactions) and custom ranges (40000+). +`buzz-core` defines each event kind as a `pub const u32` and exports the full registry as `ALL_KINDS: &[u32]` (127 kinds at the time of writing); `crates/buzz-core/src/kind.rs` is the source of truth for the current list. Kinds are `u32` (NIP-01 specifies unsigned integer; `u32` covers the full range). Buzz uses both standard Nostr kinds (e.g., kind 7 for reactions) and custom ranges (40000+). Note: `KIND_AUTH` (22242) is `pub const KIND_AUTH: u32` in `buzz-core/src/kind.rs` and imported by `buzz-relay/src/handlers/event.rs`. `KIND_CANVAS` (40100) is likewise `pub const KIND_CANVAS: u32` in `buzz-core/src/kind.rs`. diff --git a/CHANGELOG.md b/CHANGELOG.md index 71a4bbd449..e30941a355 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,5 +1,55 @@ # Changelog +## v0.5.4 + +### Desktop and shared changes + +- fix: report agent usage per provider round, not once per turn ([#4545](https://github.com/block/buzz/pull/4545)) ([`09c86c56e52651c017743268fc8ce708bb83b265`](https://github.com/block/buzz/commit/09c86c56e52651c017743268fc8ce708bb83b265)) +- fix(desktop): harden Windows installs against Defender block and orphaned Node ([#4382](https://github.com/block/buzz/pull/4382)) ([`80315ac1a68024c40b61f3a062c9cb6bf7d4efb5`](https://github.com/block/buzz/commit/80315ac1a68024c40b61f3a062c9cb6bf7d4efb5)) +- feat(desktop): improve channel template discovery ([#4549](https://github.com/block/buzz/pull/4549)) ([`c1b88af8d71d1cf6aaca517e92ce9e918cd0e8bd`](https://github.com/block/buzz/commit/c1b88af8d71d1cf6aaca517e92ce9e918cd0e8bd)) +- fix(desktop): save key backups to authorized path ([#4022](https://github.com/block/buzz/pull/4022)) ([`01c80aa9b3eaa569361966877994438ad84a280a`](https://github.com/block/buzz/commit/01c80aa9b3eaa569361966877994438ad84a280a)) +- Add channel activity hover menu ([#3935](https://github.com/block/buzz/pull/3935)) ([`b0c6d6f744e63ac88a1738f0e995680c163e1d13`](https://github.com/block/buzz/commit/b0c6d6f744e63ac88a1738f0e995680c163e1d13)) +- feat(desktop): show saved Run on settings when editing an agent ([#4539](https://github.com/block/buzz/pull/4539)) ([`f865c0054b0a400657126c9321b4d4cb7d9cc746`](https://github.com/block/buzz/commit/f865c0054b0a400657126c9321b4d4cb7d9cc746)) +- fix(desktop): disambiguate provider API key labels and annotate mint key ([#4406](https://github.com/block/buzz/pull/4406)) ([`5e0efb0bb95182f588390b55cc5affa09114c87e`](https://github.com/block/buzz/commit/5e0efb0bb95182f588390b55cc5affa09114c87e)) +- fix(desktop): make OpenAI key re-enterable after first save in card mint dialog ([#4140](https://github.com/block/buzz/pull/4140)) ([`f810a2f49e213d25119f2aa75b5b577655119b74`](https://github.com/block/buzz/commit/f810a2f49e213d25119f2aa75b5b577655119b74)) +- fix(config-bridge): add harness-definition env tier and fix equal-value model override ([#3580](https://github.com/block/buzz/pull/3580)) ([`be95a8a986d02319b27e8fb57aefe59e33a1eb13`](https://github.com/block/buzz/commit/be95a8a986d02319b27e8fb57aefe59e33a1eb13)) +- fix(desktop): stop the create-agent provider config probe from erasing keystrokes ([#4411](https://github.com/block/buzz/pull/4411)) ([`2c0ac2467437b30953a95e00f419143488bcfcc7`](https://github.com/block/buzz/commit/2c0ac2467437b30953a95e00f419143488bcfcc7)) +- feat(acp): deliver system prompt via _meta.systemPrompt for claude-agent-acp ([#4395](https://github.com/block/buzz/pull/4395)) ([`7ff5fc31895efe6265a379d01637c8ee301872e5`](https://github.com/block/buzz/commit/7ff5fc31895efe6265a379d01637c8ee301872e5)) +- fix(security): bump nostr crates for RUSTSEC-2026-0225..0232 + default sprig image to published digest ([#4392](https://github.com/block/buzz/pull/4392)) ([`318fbf896ec335bc7bcb40edafde0b6ebca53428`](https://github.com/block/buzz/commit/318fbf896ec335bc7bcb40edafde0b6ebca53428)) +- fix(desktop): back/forward via keyboard chords, mouse X1/X2 buttons, and swipe gestures ([#3778](https://github.com/block/buzz/pull/3778)) ([`f86cfc7369d4471f8939ed98be6f597b0a4b0bb2`](https://github.com/block/buzz/commit/f86cfc7369d4471f8939ed98be6f597b0a4b0bb2)) +- feat(k8s): Kubernetes backend plugin + desktop deploy path ([#4289](https://github.com/block/buzz/pull/4289)) ([`6530b58a61d4602d0a371100fedf80c5998b1e34`](https://github.com/block/buzz/commit/6530b58a61d4602d0a371100fedf80c5998b1e34)) +- feat(projects): add buzz projects CLI commands (NIP-MP kind:30621) ([#4020](https://github.com/block/buzz/pull/4020)) ([`b7bb15122e8a2053b545dc2210afc167f6c7a626`](https://github.com/block/buzz/commit/b7bb15122e8a2053b545dc2210afc167f6c7a626)) +- fix(desktop): keep thread-open affordance in archived channels ([#4012](https://github.com/block/buzz/pull/4012)) ([`8e81afa431deecd172f1ad6aab6f022f31cd812c`](https://github.com/block/buzz/commit/8e81afa431deecd172f1ad6aab6f022f31cd812c)) +- fix(desktop): point Oh My Pi preset at omp.sh ([#3516](https://github.com/block/buzz/pull/3516)) ([`3ade48d5030a8f7dbb9d3693f171e5544dcd8df1`](https://github.com/block/buzz/commit/3ade48d5030a8f7dbb9d3693f171e5544dcd8df1)) +- fix(mesh): stop restarting a busy or loading shared-compute node ([#3909](https://github.com/block/buzz/pull/3909)) ([`fa1a5b1a797870724f5c7e7e26931861a60f22cb`](https://github.com/block/buzz/commit/fa1a5b1a797870724f5c7e7e26931861a60f22cb)) +- fix(desktop): preserve first huddle speech ([#3962](https://github.com/block/buzz/pull/3962)) ([`45314fc504113aec7c54ae6520cfe5e1562aae40`](https://github.com/block/buzz/commit/45314fc504113aec7c54ae6520cfe5e1562aae40)) +- feat(desktop): Agent Trading Cards — mintable agent-snapshot card PNGs with optional NIP-44 lock ([#3278](https://github.com/block/buzz/pull/3278)) ([`eb049ddf815d48195e1713afe039d28c950d7933`](https://github.com/block/buzz/commit/eb049ddf815d48195e1713afe039d28c950d7933)) +- feat(relay): accept kind:30621 multi-repo projects at ingest ([#3171](https://github.com/block/buzz/pull/3171)) ([`cb9701cd30fb344bf134585634a09007f3155bfb`](https://github.com/block/buzz/commit/cb9701cd30fb344bf134585634a09007f3155bfb)) + +### Other repository changes + +- test(mobile): assert follow boundary semantics ([#4559](https://github.com/block/buzz/pull/4559)) ([`6de85fe31d781122756aecf954bae7d357a56b9a`](https://github.com/block/buzz/commit/6de85fe31d781122756aecf954bae7d357a56b9a)) +- docs(release): align desktop handoff instructions ([#3988](https://github.com/block/buzz/pull/3988)) ([`44fa1e8e3af30d561de981a211ff7a79bfa36493`](https://github.com/block/buzz/commit/44fa1e8e3af30d561de981a211ff7a79bfa36493)) +- Polish mobile composer and messaging UI ([#3918](https://github.com/block/buzz/pull/3918)) ([`857e63c4ddfb76f95ab40bb691e00544413f6b81`](https://github.com/block/buzz/commit/857e63c4ddfb76f95ab40bb691e00544413f6b81)) +- ci(linux): enable mesh-llm feature in Linux release and canary builds ([#4524](https://github.com/block/buzz/pull/4524)) ([`83a285f1b1a0be862d55781fad9c75ec8813886d`](https://github.com/block/buzz/commit/83a285f1b1a0be862d55781fad9c75ec8813886d)) +- fix(mobile): recover and pace live subscriptions ([#3053](https://github.com/block/buzz/pull/3053)) ([`a5dbdf5e61e4c512acd99c219c79c154ddb57295`](https://github.com/block/buzz/commit/a5dbdf5e61e4c512acd99c219c79c154ddb57295)) +- fix(git): allow deleting the default branch ([#4297](https://github.com/block/buzz/pull/4297)) ([`fc598f5f8d70728d11d0712b9fa8e3acc44ea4c3`](https://github.com/block/buzz/commit/fc598f5f8d70728d11d0712b9fa8e3acc44ea4c3)) +- docs: formal spec for remote agents and their management ([#3748](https://github.com/block/buzz/pull/3748)) ([`28ae6cd2174309529305724e455c7ca082f6fe4b`](https://github.com/block/buzz/commit/28ae6cd2174309529305724e455c7ca082f6fe4b)) +- fix(nip-oa): accept raw Nostr tag form in parse_json_array ([#4203](https://github.com/block/buzz/pull/4203)) ([`89bf03c05df795a3575b7abbe648be898ef13388`](https://github.com/block/buzz/commit/89bf03c05df795a3575b7abbe648be898ef13388)) +- perf(relay): serve relay-membership checks from the read replica ([#4124](https://github.com/block/buzz/pull/4124)) ([`ac4fa13b8e4d947071d57deb6918dcf12bf74961`](https://github.com/block/buzz/commit/ac4fa13b8e4d947071d57deb6918dcf12bf74961)) +- chore(deps): bump nostr-relay-pool for RUSTSEC-2026-0224 ([#4139](https://github.com/block/buzz/pull/4139)) ([`9d6726e5b387310975f5809473ce8372f6fde0dc`](https://github.com/block/buzz/commit/9d6726e5b387310975f5809473ce8372f6fde0dc)) +- docs(nostr): document #h requirement for live reaction subscriptions ([#3487](https://github.com/block/buzz/pull/3487)) ([`756dd7f65d6f2995e9188a0ffe54294057f8ef4f`](https://github.com/block/buzz/commit/756dd7f65d6f2995e9188a0ffe54294057f8ef4f)) +- docs(chart): fix ArgoCD example for native OCI sources (full artifact repoURL + path) ([#3426](https://github.com/block/buzz/pull/3426)) ([`36cf932ff0105a4cf574fc687deb4c1cb01bc0d1`](https://github.com/block/buzz/commit/36cf932ff0105a4cf574fc687deb4c1cb01bc0d1)) +- docs(readme): clarify which release asset to download per platform ([#3481](https://github.com/block/buzz/pull/3481)) ([`8d5afb606763fcaffd3af811be2106e41cc7347d`](https://github.com/block/buzz/commit/8d5afb606763fcaffd3af811be2106e41cc7347d)) +- fix(relay): allow open relays to set their NIP-11 workspace icon (kind:9033) ([#3998](https://github.com/block/buzz/pull/3998)) ([`5765fc74b77224f0207ddd4b41736a5ff18d333d`](https://github.com/block/buzz/commit/5765fc74b77224f0207ddd4b41736a5ff18d333d)) +- docs: note that addressable channel events scope by d, not h ([#4103](https://github.com/block/buzz/pull/4103)) ([`3d7712cc36e8da563cb1c121fc58bfc505d38496`](https://github.com/block/buzz/commit/3d7712cc36e8da563cb1c121fc58bfc505d38496)) +- docs: fix stale kind count, quick-start numbering, and empty Further Reading ([#2613](https://github.com/block/buzz/pull/2613)) ([`909a3b2c318b2ec477a3438a998a3b611f5b6d6a`](https://github.com/block/buzz/commit/909a3b2c318b2ec477a3438a998a3b611f5b6d6a)) +- docs: add one-click Railway deploy for a hosted relay ([#2733](https://github.com/block/buzz/pull/2733)) ([`19d57b0d46baa55814ac737041a36d0b405c9f64`](https://github.com/block/buzz/commit/19d57b0d46baa55814ac737041a36d0b405c9f64)) +- fix(buzz-acp): thread cache-read tokens into NIP-AM kind:44200 events ([#3999](https://github.com/block/buzz/pull/3999)) ([`b1b283cd4c7f926e12eeee8ae1f38c7471922b16`](https://github.com/block/buzz/commit/b1b283cd4c7f926e12eeee8ae1f38c7471922b16)) +- fix(release): preserve main in desktop PR body ([#3979](https://github.com/block/buzz/pull/3979)) ([`e5e5bac2a932b2b2e4eb6b559d5545a992c21b96`](https://github.com/block/buzz/commit/e5e5bac2a932b2b2e4eb6b559d5545a992c21b96)) + +[Compare desktop-v0.5.3...desktop-v0.5.4](https://github.com/block/buzz/compare/desktop-v0.5.3...desktop-v0.5.4) + ## v0.5.3 ### Desktop and shared changes diff --git a/Cargo.lock b/Cargo.lock index 9a3f91671d..937ead564a 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -915,6 +915,26 @@ dependencies = [ "uuid", ] +[[package]] +name = "buzz-backend-kubernetes" +version = "0.1.0" +dependencies = [ + "chrono", + "hex", + "http", + "http-body-util", + "k8s-openapi", + "kube", + "nostr", + "rand 0.10.1", + "rustls", + "serde", + "serde_json", + "sha2 0.11.0", + "tokio", + "tower", +] + [[package]] name = "buzz-cli" version = "0.1.0" @@ -3643,6 +3663,7 @@ dependencies = [ "http", "hyper", "hyper-util", + "log", "rustls", "rustls-native-certs", "tokio", @@ -4303,6 +4324,31 @@ dependencies = [ "ucd-trie", ] +[[package]] +name = "jsonpath-rust" +version = "0.7.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0c00ae348f9f8fd2d09f82a98ca381c60df9e0820d8d79fce43e649b4dc3128b" +dependencies = [ + "pest", + "pest_derive", + "regex", + "serde_json", + "thiserror 2.0.18", +] + +[[package]] +name = "k8s-openapi" +version = "0.26.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "06d9e5e61dd037cdc51da0d7e2b2be10f497478ea7e120d85dad632adb99882b" +dependencies = [ + "base64 0.22.1", + "chrono", + "serde", + "serde_json", +] + [[package]] name = "kasuari" version = "0.4.12" @@ -4348,6 +4394,70 @@ version = "0.4.1" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "e037a2e1d8d5fdbd49b16a4ea09d5d6401c1f29eca5ff29d03d3824dba16256a" +[[package]] +name = "kube" +version = "2.0.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "48e7bb0b6a46502cc20e4575b6ff401af45cfea150b34ba272a3410b78aa014e" +dependencies = [ + "k8s-openapi", + "kube-client", + "kube-core", +] + +[[package]] +name = "kube-client" +version = "2.0.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4987d57a184d2b5294fdad3d7fc7f278899469d21a4da39a8f6ca16426567a36" +dependencies = [ + "base64 0.22.1", + "bytes", + "chrono", + "either", + "futures", + "home", + "http", + "http-body", + "http-body-util", + "hyper", + "hyper-rustls", + "hyper-timeout", + "hyper-util", + "jsonpath-rust", + "k8s-openapi", + "kube-core", + "pem", + "rustls", + "secrecy", + "serde", + "serde_json", + "serde_yaml", + "thiserror 2.0.18", + "tokio", + "tokio-util", + "tower", + "tower-http", + "tracing", +] + +[[package]] +name = "kube-core" +version = "2.0.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "914bbb770e7bb721a06e3538c0edd2babed46447d128f7c21caa68747060ee73" +dependencies = [ + "chrono", + "derive_more", + "form_urlencoded", + "http", + "k8s-openapi", + "serde", + "serde-value", + "serde_json", + "thiserror 2.0.18", +] + [[package]] name = "lab" version = "0.11.0" @@ -5699,9 +5809,9 @@ dependencies = [ [[package]] name = "nostr" -version = "0.44.6" +version = "0.44.7" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "e826dd648489de2c5b293920e20b92932ef820302007c1987c758d4d06eeb2cf" +checksum = "c7d3d987ea7078dc36947cde532637c472a229426702e4331dd7667325378bd9" dependencies = [ "base64 0.22.1", "bech32", @@ -5743,9 +5853,9 @@ dependencies = [ [[package]] name = "nostr-relay-pool" -version = "0.44.1" +version = "0.44.3" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "91b2c039df4f96c4bf7dae52a74fd5516ad6dda83a11c0c69dea91b5255a4f37" +checksum = "c85c54d6ca9aae4ae2bf19a7663ba9db5f45f783f1d24aff55f006386b8b99a1" dependencies = [ "async-utility", "async-wsocket", @@ -6265,6 +6375,15 @@ version = "0.2.0" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "04744f49eae99ab78e0d5c0b603ab218f515ea8cfe5a456d7629ad883a3b6e7d" +[[package]] +name = "ordered-float" +version = "2.10.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "68f19d67e5a2795c94e73e0bb1cc1a7edeb2e28efd39e2e1c9b7a40c1108b11c" +dependencies = [ + "num-traits", +] + [[package]] name = "ordered-float" version = "4.6.0" @@ -6445,6 +6564,16 @@ dependencies = [ "hmac 0.12.1", ] +[[package]] +name = "pem" +version = "3.0.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1d30c53c26bc5b31a98cd02d20f25a7c8567146caf63ed593a9d87b2775291be" +dependencies = [ + "base64 0.22.1", + "serde_core", +] + [[package]] name = "pem-rfc7468" version = "1.0.0" @@ -8170,6 +8299,15 @@ dependencies = [ "cc", ] +[[package]] +name = "secrecy" +version = "0.10.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "e891af845473308773346dc847b2c23ee78fe442e0472ac50e22a18a93d3ae5a" +dependencies = [ + "zeroize", +] + [[package]] name = "secret-service" version = "4.0.0" @@ -8269,6 +8407,16 @@ dependencies = [ "serde_derive", ] +[[package]] +name = "serde-value" +version = "0.7.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "f3a1a3341211875ef120e117ea7fd5228530ae7e7036a779fdc9117be6b3282c" +dependencies = [ + "ordered-float 2.10.1", + "serde", +] + [[package]] name = "serde_bytes" version = "0.11.19" @@ -9869,6 +10017,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "4cfcf7e2740e6fc6d4d688b4ef00650406bb94adf4731e43c096c3a19fe40840" dependencies = [ "async-compression", + "base64 0.22.1", "bitflags 2.13.0", "bytes", "futures-core", diff --git a/Cargo.toml b/Cargo.toml index 3268cfaf8d..cc1dd0f9df 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -27,6 +27,7 @@ members = [ "crates/buzz-relay-mesh", "crates/buzz-dev-mcp", "crates/buzz-voice", + "crates/buzz-backend-kubernetes", "examples/countdown-bot", ] exclude = ["desktop/src-tauri"] @@ -58,6 +59,13 @@ sqlx = { version = "0.9", features = [ redis = { version = "1.0", features = ["tokio-comp", "connection-manager", "tokio-rustls-comp"] } deadpool-redis = { version = "0.23", features = ["rt_tokio_1"] } +# Kubernetes (buzz-backend-kubernetes provider). No `ring` feature here: the +# process-level CryptoProvider is installed explicitly at startup, matching +# buzz-cli/buzz-acp/buzz-admin/buzz-relay/buzz-dev-mcp — see the comment on the +# crate's own rustls dependency. +kube = { version = "2.0", default-features = false, features = ["client", "rustls-tls"] } +k8s-openapi = { version = "0.26", features = ["v1_31"] } + # Nostr nostr = { version = "0.44", features = ["nip44", "nip98"] } diff --git a/Dockerfile.sprig b/Dockerfile.sprig new file mode 100644 index 0000000000..160e0b5662 --- /dev/null +++ b/Dockerfile.sprig @@ -0,0 +1,44 @@ +# syntax=docker/dockerfile:1.7 +# Multi-arch is produced by building this file on native amd64 and arm64 runners. +# Keep both bases pinned to manifest-list digests so either architecture resolves +# to immutable source bytes. +FROM rust:1.95-alpine3.22@sha256:064dfc925d68d1a63f4fd2871bd7dc6e6ea56692989a487185855d62885d90aa AS builder + +RUN apk add --no-cache \ + build-base \ + cmake \ + git \ + musl-dev \ + openssl-dev \ + openssl-libs-static \ + perl \ + pkgconf \ + protoc +WORKDIR /build +COPY . . +RUN cargo build --locked --profile sprig -p sprig \ + && strip target/sprig/sprig + +FROM alpine:3.22@sha256:14358309a308569c32bdc37e2e0e9694be33a9d99e68afb0f5ff33cc1f695dce + +RUN apk add --no-cache bash ca-certificates curl git \ + && adduser -D -h /home/agent agent \ + && install -d -o agent -g agent /workspace /home/agent \ + && git config --system gpg.format x509 \ + && git config --system gpg.x509.program /usr/local/bin/git-sign-nostr \ + && git config --system commit.gpgSign true \ + && git config --system tag.gpgSign true + +COPY --from=builder --chmod=0755 /build/target/sprig/sprig /usr/local/bin/sprig +COPY --chmod=0755 scripts/sprig-entrypoint.sh /usr/local/bin/sprig-entrypoint +RUN for name in \ + buzz-acp buzz-agent buzz-dev-mcp rg tree buzz \ + git-credential-nostr git-sign-nostr; do \ + ln -s sprig "/usr/local/bin/$name"; \ + done + +ENV HOME=/home/agent \ + PATH=/usr/local/bin:/usr/local/sbin:/usr/sbin:/usr/bin:/sbin:/bin +WORKDIR /home/agent +USER agent +ENTRYPOINT ["/usr/local/bin/sprig-entrypoint"] diff --git a/Justfile b/Justfile index 64a1f36daf..d6e86c8d09 100644 --- a/Justfile +++ b/Justfile @@ -155,7 +155,11 @@ _ensure-sidecar-stubs: set -euo pipefail TARGET=$(rustc -vV | sed -n 's|host: ||p') mkdir -p desktop/src-tauri/binaries - for bin in buzz-acp buzz-agent buzz-dev-mcp git-credential-nostr buzz; do + SIDECARS=(buzz-acp buzz-agent buzz-dev-mcp git-credential-nostr buzz) + if [[ "$TARGET" != *windows* ]]; then + SIDECARS+=(buzz-backend-kubernetes) + fi + for bin in "${SIDECARS[@]}"; do touch "desktop/src-tauri/binaries/${bin}-${TARGET}" done @@ -236,6 +240,9 @@ desktop-release-build target="aarch64-apple-darwin": mkdir -p desktop/src-tauri/binaries touch "desktop/src-tauri/binaries/buzz-acp-$TARGET" touch "desktop/src-tauri/binaries/buzz-agent-$TARGET" + if [[ "$TARGET" != *windows* ]]; then + touch "desktop/src-tauri/binaries/buzz-backend-kubernetes-$TARGET" + fi touch "desktop/src-tauri/binaries/buzz-dev-mcp-$TARGET" touch "desktop/src-tauri/binaries/git-credential-nostr-$TARGET" touch "desktop/src-tauri/binaries/buzz-$TARGET" @@ -274,6 +281,7 @@ test: # Run unit tests only (no infra needed) test-unit: #!/usr/bin/env bash + set -euo pipefail if command -v cargo-nextest &>/dev/null; then cargo nextest run -p buzz-core -p buzz-auth --lib cargo nextest run -p buzz-voice --lib @@ -293,6 +301,12 @@ test-unit: # Gateway unit and black-box HTTP tests are infra-free. Postgres-backed # contract/race tests run in the dedicated CI job below. cargo nextest run -p buzz-push-gateway + # Kubernetes backend provider: the decision layers (state machine, GC + # planner, env precedence, naming, wire) are pure functions with a fake + # substrate, so they belong in the unit job. Enumerated explicitly + # because nothing in CI runs `cargo test --workspace` — workspace + # membership alone buys clippy/check, not a single executed test. + cargo nextest run -p buzz-backend-kubernetes else ./scripts/run-tests.sh unit fi @@ -430,7 +444,7 @@ dev *ARGS: bootstrap _ensure-sidecar-stubs _ensure-migrations fi done fi - cargo build -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr -p buzz-relay + cargo build -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr -p buzz-relay if [[ -n "{{mesh}}" ]]; then export MESH_LLM_NATIVE_RUNTIME_CACHE_DIR="$(./scripts/ensure-mesh-native-runtime.sh)" fi @@ -477,10 +491,10 @@ desktop-standalone *ARGS: _ensure-sidecar-stubs #!/usr/bin/env bash set -euo pipefail export PATH="{{justfile_directory()}}/bin:$PATH" - cargo build -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr + cargo build -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr TARGET=$(rustc -vV | sed -n 's|host: ||p') TARGET_DIR=$(cargo metadata --format-version 1 --no-deps | node -p "JSON.parse(require('fs').readFileSync(0, 'utf8')).target_directory") - for bin in buzz-acp buzz-agent buzz-dev-mcp git-credential-nostr buzz; do + for bin in buzz-acp buzz-agent buzz-backend-kubernetes buzz-dev-mcp git-credential-nostr buzz; do cp "${TARGET_DIR}/debug/${bin}" "desktop/src-tauri/binaries/${bin}-${TARGET}" chmod +x "desktop/src-tauri/binaries/${bin}-${TARGET}" done @@ -506,17 +520,26 @@ staging *ARGS: bootstrap _ensure-sidecar-stubs set -euo pipefail export PATH="{{justfile_directory()}}/bin:$PATH" pnpm install # unconditional: staging must always start with a clean dep tree - cargo build --release -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr + cargo build --release -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr FEATURES=() if [[ -n "{{mesh}}" ]]; then FEATURES=(--features mesh-llm) export MESH_LLM_NATIVE_RUNTIME_CACHE_DIR="$(./scripts/ensure-mesh-native-runtime.sh)" fi - # Replace the 0-byte sidecar stub with the real CLI binary so tauri dev picks it up. + # Replace 0-byte sidecar stubs with real binaries so tauri dev picks them up. + # buzz: the CLI sidecar. buzz-backend-kubernetes: provider discovery scans the + # exe dir for executable buzz-backend-* files, so the non-executable stub that + # tauri dev copies next to the exe would hide the provider from "Run on". TARGET=$(rustc -vV | sed -n 's|host: ||p') TARGET_DIR=$(cargo metadata --format-version 1 --no-deps | node -p "JSON.parse(require('fs').readFileSync(0, 'utf8')).target_directory") - cp "${TARGET_DIR}/release/buzz" "desktop/src-tauri/binaries/buzz-${TARGET}" - chmod +x "desktop/src-tauri/binaries/buzz-${TARGET}" + STAGING_SIDECARS=(buzz) + if [[ "$TARGET" != *windows* ]]; then + STAGING_SIDECARS+=(buzz-backend-kubernetes) + fi + for bin in "${STAGING_SIDECARS[@]}"; do + cp "${TARGET_DIR}/release/${bin}" "desktop/src-tauri/binaries/${bin}-${TARGET}" + chmod +x "desktop/src-tauri/binaries/${bin}-${TARGET}" + done cd {{desktop_dir}} export BUZZ_RELAY_URL="wss://sprout-oss.stage.blox.sqprod.co" source ../scripts/instance-env.sh @@ -533,17 +556,26 @@ production *ARGS: bootstrap _ensure-sidecar-stubs set -euo pipefail export PATH="{{justfile_directory()}}/bin:$PATH" pnpm install # unconditional: production must always start with a clean dep tree - cargo build --release -p buzz-acp -p buzz-agent -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr + cargo build --release -p buzz-acp -p buzz-agent -p buzz-backend-kubernetes -p buzz-dev-mcp -p buzz-cli -p git-credential-nostr FEATURES=() if [[ -n "{{mesh}}" ]]; then FEATURES=(--features mesh-llm) export MESH_LLM_NATIVE_RUNTIME_CACHE_DIR="$(./scripts/ensure-mesh-native-runtime.sh)" fi - # Replace the 0-byte sidecar stub with the real CLI binary so tauri dev picks it up. + # Replace 0-byte sidecar stubs with real binaries so tauri dev picks them up. + # buzz: the CLI sidecar. buzz-backend-kubernetes: provider discovery scans the + # exe dir for executable buzz-backend-* files, so the non-executable stub that + # tauri dev copies next to the exe would hide the provider from "Run on". TARGET=$(rustc -vV | sed -n 's|host: ||p') TARGET_DIR=$(cargo metadata --format-version 1 --no-deps | node -p "JSON.parse(require('fs').readFileSync(0, 'utf8')).target_directory") - cp "${TARGET_DIR}/release/buzz" "desktop/src-tauri/binaries/buzz-${TARGET}" - chmod +x "desktop/src-tauri/binaries/buzz-${TARGET}" + PRODUCTION_SIDECARS=(buzz) + if [[ "$TARGET" != *windows* ]]; then + PRODUCTION_SIDECARS+=(buzz-backend-kubernetes) + fi + for bin in "${PRODUCTION_SIDECARS[@]}"; do + cp "${TARGET_DIR}/release/${bin}" "desktop/src-tauri/binaries/${bin}-${TARGET}" + chmod +x "desktop/src-tauri/binaries/${bin}-${TARGET}" + done cd {{desktop_dir}} export BUZZ_RELAY_URL="wss://buzz.block.builderlab.xyz" source ../scripts/instance-env.sh diff --git a/NOSTR.md b/NOSTR.md index 59df31b991..cce70f2f77 100644 --- a/NOSTR.md +++ b/NOSTR.md @@ -39,7 +39,7 @@ just relay & # relay on :3000 PGPASSWORD=buzz_dev psql -h localhost -U buzz -d buzz -c \ "INSERT INTO pubkey_allowlist (pubkey) VALUES (decode('<64-char-hex-pubkey>', 'hex'))" -# 5. Connect any NIP-29 + NIP-42 client to ws://localhost:3000 +# 4. Connect any NIP-29 + NIP-42 client to ws://localhost:3000 ``` ### What Works @@ -163,6 +163,10 @@ nak req -k 9 --tag "h=" --stream \ nak event -k 7 -c "+" --tag "h=" --tag "e=" \ --auth --sec ws://localhost:3000 +# Subscribe to reactions to channel messages — include #h for live delivery (see note below) +nak req -k 7 --tag "h=" --stream \ + --auth --sec ws://localhost:3000 + # Delete a message (#h optional; #e required; must be self-authored) nak event -k 5 -c "reason" --tag "h=" --tag "e=" \ --auth --sec ws://localhost:3000 @@ -185,6 +189,14 @@ nak req -k 1059 --tag "p=" \ --auth --sec ws://localhost:3000 ``` +> **Note:** The relay derives a reaction's channel from its `#e` target (client `#h` is +> ignored for channel determination). Reactions to channel-scoped events are therefore +> channel-scoped. Live fan-out keeps channel-scoped and global subscriptions strictly +> separate, which means a kinds-only subscription (`{"kinds":[7]}`) receives none of +> those reactions — subscribe with `{"kinds":[7],"#h":[""]}` instead. +> `#h` matching works whether or not the signed reaction carries an `h` tag: explicit +> `h` tags are matched directly, and tagless reactions match via their stored channel. + ### Tested Clients (Direct) | Client | Platform | Evidence | Notes | @@ -354,3 +366,7 @@ but only admins/owners can set it. Full spec: --- ## Further Reading + +- [nostr-protocol/nips](https://github.com/nostr-protocol/nips) — the upstream NIP specifications (NIP-01, NIP-29, NIP-42, and the other NIPs referenced throughout this guide). +- [`docs/nips/`](docs/nips/) — Buzz's own NIP extension documents. +- [`ARCHITECTURE.md`](ARCHITECTURE.md) — event kinds, wire protocol, and relay internals. diff --git a/README.md b/README.md index 2c58ceecad..56439f00bc 100644 --- a/README.md +++ b/README.md @@ -116,10 +116,30 @@ New to Buzz? Pick the path that matches you. ### I just want to try the app -Grab a packaged build from the [latest release](https://github.com/block/buzz/releases/latest) — macOS (`.dmg`), Linux (`.AppImage` / `.deb`), or Windows (`.exe`). Install it like any other app. +Grab a packaged build from the [latest release](https://github.com/block/buzz/releases/latest): + +| Platform | File | +|---|---| +| macOS (Apple Silicon) | `Buzz__aarch64.dmg` | +| macOS (Intel) | `Buzz__x64.dmg` | +| Linux (x86_64) | `Buzz__amd64.AppImage` or `Buzz__amd64.deb` | +| Windows (x64) | `Buzz__x64-setup_alpha-unsigned.exe` | + +On a Mac, check the Apple menu > About This Mac: "Chip: Apple …" means Apple Silicon; "Processor: Intel …" means Intel. + +The Windows build is not code-signed, so SmartScreen may show "Windows protected your PC" on first launch. If available, click **More info**, then **Run anyway**. + By default the app connects to `ws://localhost:3000`. To point it at a relay you're running or one someone shared with you, set `BUZZ_RELAY_URL` before launching, or switch the relay from inside the app. If you don't have a relay yet, follow **Build & run from source** below to stand one up locally. +### I want my own hosted relay + +To run a relay for your team without managing servers, you can deploy one to Railway in a click: + +[![Deploy on Railway](https://railway.com/button.svg)](https://railway.com/deploy/buzz-relay-block) + +See [here](https://engineering.block.xyz/blog/run-your-own-buzz-relay) for details. + ### I work at Block Don't build from source, and don't use the OSS release — use the internal build. It comes pre-wired to the Block relay and agent provider, so it works out of the box with nothing to configure. diff --git a/RELEASING.md b/RELEASING.md index e729f8b50c..23dacea2ce 100644 --- a/RELEASING.md +++ b/RELEASING.md @@ -56,14 +56,15 @@ or mobile GitHub Release. updates the PR. 2. Review the recorded base and candidate SHA, the complete changelog, and CI. The required **Desktop Release Candidate** check validates the exact head. - Authorization is either an approval on that exact head or a permitted Default - ruleset bypass at merge time. Any regeneration changes the head and requires - the checks—and, for the review path, approval—to run again. + A trusted repository member, owner, or collaborator must approve that exact + candidate head. Any regeneration or push changes the head, invalidates the + prior approval, and requires both the checks and approval to run again. 3. **Squash merge** the PR. The protected branch must still be exactly the recorded base; otherwise regenerate the candidate from current `main`. 4. `auto-tag-on-release-pr-merge` verifies the frozen parent, full-tree identity, - required checks, and one of the two authorization paths, then tags the squash - commit as `desktop-v`. + required checks, and trusted approval on the exact candidate head, then tags + the squash commit as `desktop-v`. An admin or ruleset bypass does not + authorize desktop tagging. 5. The tag triggers `release.yml`. It builds and stages Apple Silicon and Intel macOS, Windows, and Linux artifacts; publishes the versioned release only after the complete set succeeds; then updates the rolling updater manifest @@ -184,10 +185,12 @@ Buildkite pipeline accepts only an exact candidate tag. For mobile, trigger the private [Release Mobile pipeline](https://buildkite.com/runway/buzz-mobile-releases) with -an exact RC tag for the platform build being cut. For desktop, use -[Release Desktop](https://buildkite.com/runway/sprout-releases). See the +an exact RC tag for the platform build being cut. For desktop, start +[Release Desktop](https://buildkite.com/runway/sprout-releases) and enter the +exact public source tag as `desktop_ref=desktop-v`; a generic +`v` tag is intentionally rejected. See the [buzz-releases README](https://github.com/squareup/buzz-releases#cutting-a-release) -for the private pipeline contract. +for the rest of the private pipeline contract. --- @@ -269,9 +272,9 @@ actor list. Do not update the branch manually and do not weaken the ruleset. Run `just release-desktop ` again from current `main`; this regenerates the -candidate, reruns CI, and requires a fresh approval when using the review path. -The post-merge verifier refuses to tag a squash whose parent differs from the -recorded candidate base or whose tree differs from the validated PR head. +candidate, reruns CI, and requires a fresh trusted approval on the new exact +head. The post-merge verifier refuses to tag a squash whose parent differs from +the recorded candidate base or whose tree differs from the validated PR head. ### Local `just release-desktop` fails with "must be on main branch" Switch to `main` and pull latest before running the release recipe. diff --git a/VISION_REMOTE_AGENTS.md b/VISION_REMOTE_AGENTS.md index b02d1bc92d..4b187f355a 100644 --- a/VISION_REMOTE_AGENTS.md +++ b/VISION_REMOTE_AGENTS.md @@ -56,7 +56,7 @@ Remote agents solve it from the inside. Because the desktop retains no substrate **The body's state is mortal.** Files, checkouts, half-finished working trees — gone with the body unless the substrate persists them. The agent survives; its scratch space doesn't. Durable knowledge belongs on the relay, and agents are built to put it there. -**Presence can lag the truth, but not for long.** If the substrate kills a body without ceremony, the presence dot can outlive the agent — by seconds if the connection drops cleanly, by at most about ninety if it doesn't. Presence is a lease the agent renews, not a flag it sets: a dead agent stops renewing and the relay forgets it. Ninety seconds of a wrong dot, never an indefinite one. +**Presence can lag the truth, but not for long.** If the substrate kills a body without ceremony, the presence dot can outlive the agent — by seconds if the connection drops cleanly, by at most about three minutes if it doesn't. Presence is a lease the agent renews, not a flag it sets: a dead agent stops renewing and the relay forgets it. A bounded wrong dot, never an indefinite one. **A running agent finishes on the configuration it started with.** New keys, new models, new settings take effect on the next body. And an instance that never got far enough to run — a body that failed to start — is the substrate operator's residue to clear, with the substrate's own tools. Editing an agent mid-sentence was never on the menu. diff --git a/crates/buzz-acp/src/acp.rs b/crates/buzz-acp/src/acp.rs index 8a698954a0..700d5e8dcf 100644 --- a/crates/buzz-acp/src/acp.rs +++ b/crates/buzz-acp/src/acp.rs @@ -619,29 +619,46 @@ impl AcpClient { /// Send `session/new` and return the full response alongside the session ID. /// /// `cwd` must be an absolute path. `mcp_servers` may be empty. - /// `system_prompt` is included in the request when `Some` — agents that - /// support the field will use it; others ignore unknown fields per JSON-RPC. + /// + /// `system_prompt` controls how the prompt text is delivered: + /// + /// - `None` — no system-prompt field in the request (legacy framing). + /// - `Some(SystemPromptTransport::Field(text))` — bare `systemPrompt` field + /// (ACP protocol v2, buzz-agent, goose unused). + /// - `Some(SystemPromptTransport::ClaudeMeta(text))` — `_meta.systemPrompt` + /// as `{"append": text}`, keeping claude-agent-acp's native preset intact. + /// /// `session_title` rides in `_meta.sessionTitle` when `Some`; `_meta` is /// omitted entirely otherwise, since adapters may distinguish an absent - /// member from a null one. + /// member from a null one. When both `ClaudeMeta` and `session_title` are + /// present the two `_meta` members are merged into a single object. + /// /// Callers use [`extract_model_config_options`] and [`extract_model_state`] /// to pull model info from the raw result. pub async fn session_new_full( &mut self, cwd: &str, mcp_servers: Vec, - system_prompt: Option<&str>, + system_prompt: Option>, session_title: Option<&str>, ) -> Result { let mut params = serde_json::json!({ "cwd": cwd, "mcpServers": mcp_servers, }); - if let Some(sp) = system_prompt { - params["systemPrompt"] = serde_json::Value::String(sp.to_owned()); + match system_prompt { + Some(SystemPromptTransport::Field(sp)) => { + params["systemPrompt"] = serde_json::Value::String(sp.to_owned()); + } + Some(SystemPromptTransport::ClaudeMeta(sp)) => { + // Merge into _meta so sessionTitle (set below) is not clobbered. + params["_meta"]["systemPrompt"] = serde_json::json!({ "append": sp }); + } + None => {} } if let Some(title) = session_title { - params["_meta"] = serde_json::json!({ "sessionTitle": title }); + // Merge — _meta may already carry systemPrompt from ClaudeMeta above. + params["_meta"]["sessionTitle"] = serde_json::Value::String(title.to_owned()); } let result = self.send_request("session/new", params).await?; let session_id = result["sessionId"] @@ -663,7 +680,7 @@ impl AcpClient { &mut self, cwd: &str, mcp_servers: Vec, - system_prompt: Option<&str>, + system_prompt: Option>, session_title: Option<&str>, ) -> Result { Ok(self @@ -2038,6 +2055,22 @@ pub struct SessionNewResponse { pub raw: serde_json::Value, } +/// How to deliver a system prompt on `session/new`. +/// +/// The two variants match the two mechanisms supported by current adapters: +/// +/// - **`Field`** — bare `systemPrompt` field (ACP protocol v2, buzz-agent). +/// - **`ClaudeMeta`** — `_meta.systemPrompt: {"append": text}`, used by +/// `claude-agent-acp` to append to the adapter's own native system prompt +/// while keeping its tool-use preset intact. +#[derive(Debug, Clone, PartialEq)] +pub enum SystemPromptTransport<'a> { + /// Deliver as a bare top-level `systemPrompt` field. + Field(&'a str), + /// Deliver as `_meta.systemPrompt: {"append": text}`. + ClaudeMeta(&'a str), +} + /// How to switch to a particular model on a session. #[derive(Debug, Clone, PartialEq, serde::Serialize)] #[serde(tag = "type")] @@ -3271,7 +3304,12 @@ mod tests { .expect("initialize should succeed"); let resp = client - .session_new_full("/tmp", vec![], Some("Custom system prompt"), None) + .session_new_full( + "/tmp", + vec![], + Some(SystemPromptTransport::Field("Custom system prompt")), + None, + ) .await .expect("session_new_full should succeed"); @@ -3423,6 +3461,87 @@ mod tests { ); } + // ── claude-agent-acp _meta.systemPrompt transport ───────────────────── + + #[tokio::test] + async fn session_new_full_sends_claude_meta_system_prompt_when_claude_meta_transport() { + // When ClaudeMeta transport is requested, the prompt must appear as + // _meta.systemPrompt: {"append": text} — never as a bare systemPrompt field. + let script = r#" + read -t 2 _init + echo '{"jsonrpc":"2.0","id":0,"result":{"protocolVersion":1,"agentCapabilities":{}}}' + read -t 2 REQ + echo '{"jsonrpc":"2.0","id":1,"result":{"sessionId":"ses_claude","_receivedRequest":'"$REQ"'}}' + sleep 1 + "#; + let mut client = spawn_script(script).await; + client + .initialize() + .await + .expect("initialize should succeed"); + + let resp = client + .session_new_full( + "/tmp", + vec![], + Some(SystemPromptTransport::ClaudeMeta("Be concise")), + None, + ) + .await + .expect("session_new_full should succeed"); + + let received = &resp.raw["_receivedRequest"]; + assert!( + received["params"].get("systemPrompt").is_none(), + "bare systemPrompt must not be present for ClaudeMeta transport" + ); + assert_eq!( + received["params"]["_meta"]["systemPrompt"]["append"].as_str(), + Some("Be concise"), + "_meta.systemPrompt.append must carry the prompt text" + ); + } + + #[tokio::test] + async fn session_new_full_merges_claude_meta_and_session_title_into_single_meta_object() { + // Both ClaudeMeta prompt and session_title must coexist under _meta — + // the prompt must not clobber sessionTitle or vice versa. + let script = r#" + read -t 2 _init + echo '{"jsonrpc":"2.0","id":0,"result":{"protocolVersion":1,"agentCapabilities":{}}}' + read -t 2 REQ + echo '{"jsonrpc":"2.0","id":1,"result":{"sessionId":"ses_merged","_receivedRequest":'"$REQ"'}}' + sleep 1 + "#; + let mut client = spawn_script(script).await; + client + .initialize() + .await + .expect("initialize should succeed"); + + let resp = client + .session_new_full( + "/tmp", + vec![], + Some(SystemPromptTransport::ClaudeMeta("Be concise")), + Some("Fizz · #buzz-dev"), + ) + .await + .expect("session_new_full should succeed"); + + let received = &resp.raw["_receivedRequest"]; + assert_eq!( + received["params"]["_meta"]["systemPrompt"]["append"].as_str(), + Some("Be concise"), + "_meta.systemPrompt.append must be present" + ); + assert_eq!( + received["params"]["_meta"]["sessionTitle"].as_str(), + Some("Fizz · #buzz-dev"), + "_meta.sessionTitle must be present alongside systemPrompt" + ); + } + // ── Goose-native steer scaffold (PR follow-up to #1160) ────────────── /// Helper: spawn an inert `cat` subprocess so we have a real AcpClient diff --git a/crates/buzz-acp/src/config.rs b/crates/buzz-acp/src/config.rs index dab61be30a..35aaec188d 100644 --- a/crates/buzz-acp/src/config.rs +++ b/crates/buzz-acp/src/config.rs @@ -474,6 +474,11 @@ pub struct CliArgs { #[arg(long, env = "BUZZ_ACP_RELAY_OBSERVER", default_value_t = false)] pub relay_observer: bool, + /// Exit after this many seconds with no dispatched events and no turn in flight. + /// 0 disables inactivity self-termination. + #[arg(long, env = "BUZZ_ACP_EXIT_AFTER_INACTIVITY", default_value_t = 0)] + pub exit_after_inactivity: u64, + /// Connect and subscribe before starting the ACP/LLM subprocess pool. #[arg(long, env = "BUZZ_ACP_LAZY_POOL", default_value_t = false)] pub lazy_pool: bool, @@ -550,6 +555,8 @@ pub struct Config { pub has_generated_codex_config: bool, /// Whether to publish encrypted observer frames through the relay. pub relay_observer: bool, + /// Seconds without dispatched events before an idle harness exits. 0 = disabled. + pub exit_after_inactivity_secs: u64, /// Whether ACP/LLM subprocess initialization is deferred until accepted work arrives. pub lazy_pool: bool, /// Agent owner pubkey (hex). Used for `--respond-to=owner-only` gate. @@ -1098,6 +1105,7 @@ impl Config { persona_env_vars, has_generated_codex_config, relay_observer: args.relay_observer, + exit_after_inactivity_secs: args.exit_after_inactivity, lazy_pool: args.lazy_pool, agent_owner: args.agent_owner.map(|s| s.trim().to_ascii_lowercase()), no_base_prompt: args.no_base_prompt, @@ -1468,6 +1476,7 @@ mod tests { persona_env_vars: vec![], has_generated_codex_config: false, relay_observer: false, + exit_after_inactivity_secs: 0, lazy_pool: false, agent_owner: None, no_base_prompt: false, @@ -2167,6 +2176,22 @@ channels = "ALL" assert!(err.to_string().contains("turn liveness interval must be 0")); } + #[test] + fn inactivity_exit_defaults_disabled_and_accepts_cli_value() { + let key = "0".repeat(64); + let default = CliArgs::parse_from(["buzz-acp", "--private-key", &key]); + assert_eq!(default.exit_after_inactivity, 0); + + let configured = CliArgs::parse_from([ + "buzz-acp", + "--private-key", + &key, + "--exit-after-inactivity", + "120", + ]); + assert_eq!(configured.exit_after_inactivity, 120); + } + #[test] fn lazy_pool_defaults_off() { let key = "0".repeat(64); diff --git a/crates/buzz-acp/src/lib.rs b/crates/buzz-acp/src/lib.rs index 403512a322..811253e4ac 100644 --- a/crates/buzz-acp/src/lib.rs +++ b/crates/buzz-acp/src/lib.rs @@ -1228,6 +1228,59 @@ impl Drop for RespawnGuard { // sync entry point — `std::env::set_var` is only safe before tokio spawns // worker threads (Rust 2024 edition safety requirement). +fn inactivity_expired( + last_activity: tokio::time::Instant, + now: tokio::time::Instant, + bound: Duration, + turn_in_flight: bool, +) -> bool { + !bound.is_zero() && !turn_in_flight && now.duration_since(last_activity) >= bound +} + +#[cfg(test)] +mod inactivity_tests { + use super::*; + + #[test] + fn zero_disables_expiry_and_in_flight_turns_defer_it() { + let started = tokio::time::Instant::now(); + let after_bound = started + Duration::from_secs(61); + + assert!(!inactivity_expired( + started, + after_bound, + Duration::ZERO, + false + )); + assert!(!inactivity_expired( + started, + after_bound, + Duration::from_secs(60), + true + )); + assert!(inactivity_expired( + started, + after_bound, + Duration::from_secs(60), + false + )); + } + + #[test] + fn dispatched_activity_restarts_the_inactivity_bound() { + let started = tokio::time::Instant::now(); + let dispatched = started + Duration::from_secs(50); + let checked = started + Duration::from_secs(61); + + assert!(!inactivity_expired( + dispatched, + checked, + Duration::from_secs(60), + false + )); + } +} + pub fn run() -> Result<()> { config::propagate_legacy_env_vars(); tokio_main() @@ -1601,6 +1654,21 @@ async fn tokio_main() -> Result<()> { let mut typing_channels: HashMap = HashMap::new(); let mut presence_task: Option> = None; + // Independent of pool readiness: a never-mentioned lazy agent must still + // self-terminate. The watch interval is capped so small configured bounds + // remain reasonably precise without waking long-lived agents frequently. + let inactivity_bound = Duration::from_secs(config.exit_after_inactivity_secs); + let mut last_activity = tokio::time::Instant::now(); + let mut inactivity_reaper = if inactivity_bound.is_zero() { + None + } else { + let interval = inactivity_bound.min(Duration::from_secs(30)); + Some(tokio::time::interval_at( + tokio::time::Instant::now() + interval, + interval, + )) + }; + // Runs at the TOP of every loop iteration via Instant check — cannot be // starved by the biased select. Slot refill spawns background tasks so // spawn_and_init never blocks the main loop. @@ -1774,7 +1842,9 @@ async fn tokio_main() -> Result<()> { // called on relay events or pool results, neither of which // arrive when the channel is silent. if queue.has_flushable_work() { - for (channel_id, thread_tags) in dispatch_pending(&mut pool, &mut queue, &ctx) { + for (channel_id, thread_tags) in + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) + { typing_channels.insert(channel_id, thread_tags); } } @@ -1810,7 +1880,9 @@ async fn tokio_main() -> Result<()> { // this, batches requeued during crash recovery sit idle until the // next relay event arrives — which can be minutes on quiet channels. if respawn_collected { - for (channel_id, thread_tags) in dispatch_pending(&mut pool, &mut queue, &ctx) { + for (channel_id, thread_tags) in + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) + { typing_channels.insert(channel_id, thread_tags); } } @@ -2258,7 +2330,7 @@ async fn tokio_main() -> Result<()> { } if pool_ready { for (channel_id, thread_tags) in - dispatch_pending(&mut pool, &mut queue, &ctx) + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) { typing_channels.insert(channel_id, thread_tags); } @@ -2275,6 +2347,27 @@ async fn tokio_main() -> Result<()> { } None } + _ = async { + match inactivity_reaper.as_mut() { + Some(timer) => timer.tick().await, + None => std::future::pending().await, + } + } => { + let _ = result_rx; + if inactivity_expired( + last_activity, + tokio::time::Instant::now(), + inactivity_bound, + queue.has_in_flight() || heartbeat_in_flight, + ) { + tracing::info!( + inactivity_seconds = config.exit_after_inactivity_secs, + "inactivity bound reached — exiting gracefully" + ); + let _ = shutdown_tx.send(()); + } + None + } _ = async { match heartbeat.as_mut() { Some(hb) => hb.tick().await, @@ -2287,7 +2380,7 @@ async fn tokio_main() -> Result<()> { } else if queue.has_flushable_work() { tracing::debug!("heartbeat_skipped_events"); for (channel_id, thread_tags) in - dispatch_pending(&mut pool, &mut queue, &ctx) + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) { typing_channels.insert(channel_id, thread_tags); } @@ -2385,7 +2478,9 @@ async fn tokio_main() -> Result<()> { { break; } - for (channel_id, thread_tags) in dispatch_pending(&mut pool, &mut queue, &ctx) { + for (channel_id, thread_tags) in + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) + { typing_channels.insert(channel_id, thread_tags); } } @@ -2408,7 +2503,9 @@ async fn tokio_main() -> Result<()> { tracing::error!("all agents dead — exiting"); break; } - for (channel_id, thread_tags) in dispatch_pending(&mut pool, &mut queue, &ctx) { + for (channel_id, thread_tags) in + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) + { typing_channels.insert(channel_id, thread_tags); } } @@ -2550,7 +2647,9 @@ async fn tokio_main() -> Result<()> { // tear down the in-flight task; on its completion the // queue drains. We still try here in case the in-flight // task has already returned. - for (channel_id, thread_tags) in dispatch_pending(&mut pool, &mut queue, &ctx) { + for (channel_id, thread_tags) in + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) + { typing_channels.insert(channel_id, thread_tags); } } @@ -2577,7 +2676,7 @@ async fn tokio_main() -> Result<()> { None, ); for (channel_id, thread_tags) in - dispatch_pending(&mut pool, &mut queue, &ctx) + dispatch_pending(&mut pool, &mut queue, &ctx, &mut last_activity) { typing_channels.insert(channel_id, thread_tags); } @@ -2911,6 +3010,7 @@ fn dispatch_pending( pool: &mut AgentPool, queue: &mut EventQueue, ctx: &Arc, + last_activity: &mut tokio::time::Instant, ) -> Vec<(Uuid, ThreadTags)> { let mut dispatched_channels = Vec::new(); loop { @@ -2990,6 +3090,7 @@ fn dispatch_pending( }, ); dispatched_channels.push((channel_id, typing_scope)); + *last_activity = tokio::time::Instant::now(); } tracing::debug!( dispatched = dispatched_channels.len(), @@ -5031,6 +5132,7 @@ mod build_mcp_servers_tests { persona_env_vars: vec![], has_generated_codex_config: false, relay_observer: false, + exit_after_inactivity_secs: 0, lazy_pool: false, agent_owner: None, no_base_prompt: false, @@ -5252,6 +5354,7 @@ mod error_outcome_emission_tests { persona_env_vars: vec![], has_generated_codex_config: false, relay_observer: false, + exit_after_inactivity_secs: 0, lazy_pool: false, agent_owner: None, no_base_prompt: false, diff --git a/crates/buzz-acp/src/pool.rs b/crates/buzz-acp/src/pool.rs index 158477c0af..64edf68ee2 100644 --- a/crates/buzz-acp/src/pool.rs +++ b/crates/buzz-acp/src/pool.rs @@ -32,6 +32,7 @@ use uuid::Uuid; use crate::acp::{ extract_model_config_options, extract_model_state, model_in_catalog, resolve_model_switch_method, AcpClient, AcpError, McpServer, ModelSwitchMethod, StopReason, + SystemPromptTransport, }; use crate::config::{compose_session_title, DedupMode, PermissionMode}; use crate::observer; @@ -171,6 +172,13 @@ pub struct OwnedAgent { pub protocol_version: u32, } +/// Package name reported by `claude-agent-acp` in its `initialize` response. +/// Any adapter reporting this name supports `_meta.systemPrompt: {append: ...}` +/// on `session/new` — the feature landed in v0.6.0 (Oct 2025), before the +/// `@zed-industries/claude-code-acp` → `@agentclientprotocol/claude-agent-acp` +/// rename, so the new name is a reliable capability gate. +const CLAUDE_AGENT_ACP_NAME: &str = "@agentclientprotocol/claude-agent-acp"; + fn has_system_prompt_support( protocol_version: u32, agent_name: &str, @@ -178,20 +186,25 @@ fn has_system_prompt_support( ) -> bool { if agent_name == "goose" { goose_system_prompt_supported == Some(true) + } else if agent_name == CLAUDE_AGENT_ACP_NAME { + true } else { protocol_version >= 2 } } -fn session_new_system_prompt( +fn session_new_system_prompt<'a>( is_goose: bool, protocol_version: u32, - prompt: Option<&str>, -) -> Option<&str> { - if is_goose || protocol_version < 2 { + agent_name: &str, + prompt: Option<&'a str>, +) -> Option> { + if is_goose || (protocol_version < 2 && agent_name != CLAUDE_AGENT_ACP_NAME) { None + } else if agent_name == CLAUDE_AGENT_ACP_NAME { + prompt.map(SystemPromptTransport::ClaudeMeta) } else { - prompt + prompt.map(SystemPromptTransport::Field) } } @@ -907,6 +920,7 @@ async fn create_session_and_apply_model( session_new_system_prompt( is_goose, agent.protocol_version, + &agent.agent_name, combined_system_prompt.as_deref(), ), session_title.as_deref(), @@ -3624,7 +3638,11 @@ pub(crate) fn build_turn_metric_counts( // from input+output. total_tokens: usage.turn_total_tokens, cost_usd: usage.turn_cost_usd, - cache_read_tokens: None, + // Field-local: present when the cumulative counter was monotonic + // across this turn. Zero means no cache hits this turn (not absent). + cache_read_tokens: usage.turn_cache_read_tokens, + // buzz-agent does not emit a cache-write count on the wire today; + // leave None rather than deriving it from other fields. cache_write_tokens: None, }) } else { @@ -3642,7 +3660,13 @@ pub(crate) fn build_turn_metric_counts( // one. Never derived from input+output (NIP-AM MUST NOT). total_tokens: usage.cumulative_total_tokens, cost_usd: usage.cumulative_cost_usd, - cache_read_tokens: None, + // Session-cumulative cache-read tokens; None when the harness never + // reported this field (e.g. goose or older buzz-agent sessions). + // Passes through directly — do not wrap in Some() as the field already + // carries provenance (None vs Some(0) are distinct meanings). + cache_read_tokens: usage.cumulative_cache_read_tokens, + // buzz-agent does not emit a cache-write count on the wire today; + // leave None rather than deriving it from other fields. cache_write_tokens: None, }); (turn_counts, cumulative_counts) @@ -3993,18 +4017,48 @@ mod tests { assert!(has_system_prompt_support(2, "goose", Some(true))); assert!(has_system_prompt_support(1, "goose", Some(true))); assert!(has_system_prompt_support(2, "buzz-agent", None)); + // Goose never receives system prompt via session/new (uses post-hoc method). assert_eq!( - session_new_system_prompt(true, 2, Some("instructions")), + session_new_system_prompt(true, 2, "goose", Some("instructions")), None ); + // Protocol-v2 non-goose gets Field transport. assert_eq!( - session_new_system_prompt(false, 2, Some("instructions")), - Some("instructions") + session_new_system_prompt(false, 2, "buzz-agent", Some("instructions")), + Some(SystemPromptTransport::Field("instructions")) ); + // Protocol-v1 non-goose, non-claude gets None (legacy user-message framing). assert_eq!( - session_new_system_prompt(false, 1, Some("instructions")), + session_new_system_prompt(false, 1, "codex", Some("instructions")), None ); + // claude-agent-acp gets ClaudeMeta transport regardless of protocol version. + assert_eq!( + session_new_system_prompt(false, 1, CLAUDE_AGENT_ACP_NAME, Some("instructions")), + Some(SystemPromptTransport::ClaudeMeta("instructions")) + ); + assert_eq!( + session_new_system_prompt(true, 1, CLAUDE_AGENT_ACP_NAME, Some("instructions")), + None, + "goose path must never produce a transport even when agent_name matches" + ); + } + + #[test] + fn claude_agent_acp_has_system_prompt_support_regardless_of_protocol_version() { + // claude-agent-acp declares protocolVersion:1 but supports _meta.systemPrompt; + // has_system_prompt_support must return true so user-message framing is suppressed. + assert!(has_system_prompt_support(1, CLAUDE_AGENT_ACP_NAME, None)); + assert!(has_system_prompt_support(2, CLAUDE_AGENT_ACP_NAME, None)); + } + + #[test] + fn old_zed_adapter_name_falls_through_to_protocol_version_gate() { + // The renamed @zed-industries package predates the _meta.systemPrompt support, + // so it must not be treated as capable and stays on legacy user-message framing. + let old_name = "@zed-industries/claude-code-acp"; + assert!(!has_system_prompt_support(1, old_name, None)); + assert!(has_system_prompt_support(2, old_name, None)); } #[test] @@ -6022,10 +6076,12 @@ mod tests { turn_output_tokens: Some(50), turn_total_tokens: None, turn_cost_usd: None, + turn_cache_read_tokens: None, cumulative_input_tokens: 100, cumulative_output_tokens: 50, cumulative_total_tokens: None, cumulative_cost_usd: None, + cumulative_cache_read_tokens: None, model: None, }; // owner_pubkey = None → early return, no panic. @@ -6056,10 +6112,12 @@ mod tests { turn_output_tokens: Some(80), turn_total_tokens: None, turn_cost_usd: Some(0.001), + turn_cache_read_tokens: None, cumulative_input_tokens: 200, cumulative_output_tokens: 80, cumulative_total_tokens: None, cumulative_cost_usd: Some(0.001), + cumulative_cache_read_tokens: None, model: None, }; // Will try to publish and fail (no real relay) but must not panic. @@ -6091,10 +6149,12 @@ mod tests { turn_output_tokens: Some(20), turn_total_tokens: None, turn_cost_usd: None, + turn_cache_read_tokens: None, cumulative_input_tokens: 150, cumulative_output_tokens: 70, cumulative_total_tokens: None, cumulative_cost_usd: None, + cumulative_cache_read_tokens: None, model: None, }; // Must not panic; HTTP submit will fail (no real relay) — that's fine. @@ -6126,10 +6186,12 @@ mod tests { turn_output_tokens: None, turn_total_tokens: None, turn_cost_usd: None, + turn_cache_read_tokens: None, cumulative_input_tokens: 400, cumulative_output_tokens: 100, cumulative_total_tokens: None, cumulative_cost_usd: None, + cumulative_cache_read_tokens: None, model: None, }; // Will try to publish (encrypt succeeds) and fail HTTP (no relay) — must not panic. @@ -6158,10 +6220,12 @@ mod tests { turn_output_tokens: Some(30), turn_total_tokens: Some(130), // genuine per-turn total turn_cost_usd: None, + turn_cache_read_tokens: None, cumulative_input_tokens: 500, cumulative_output_tokens: 120, cumulative_total_tokens: Some(620), // genuine cumulative total cumulative_cost_usd: None, + cumulative_cache_read_tokens: None, model: None, }; @@ -6205,10 +6269,12 @@ mod tests { turn_output_tokens: Some(60), turn_total_tokens: None, // provider did not supply a total turn_cost_usd: None, + turn_cache_read_tokens: None, cumulative_input_tokens: 200, cumulative_output_tokens: 60, cumulative_total_tokens: None, // session has no total cumulative_cost_usd: None, + cumulative_cache_read_tokens: None, model: None, }; @@ -6248,6 +6314,96 @@ mod tests { ); } + /// A payload with nonzero `accumulatedCachedInputTokens` on the second turn + /// must produce a kind:44200 payload where `cumulative.cacheReadTokens` is + /// nonzero and `turn.cacheReadTokens` reflects the per-turn delta. + /// This is the acceptance-criterion test: it proves the threading is live, + /// not hardcoded to None. + #[test] + fn test_build_turn_metric_counts_cache_read_tokens_thread_through() { + // Wire-parse a buzz-agent payload with cache, run it through the tracker, + // and verify the published TokenCounts carry the cache field. + let raw1 = serde_json::json!({ + "sessionId": "cache-sess", + "update": { + "sessionUpdate": "usage_update", + "accumulatedInputTokens": 15_091, + "accumulatedOutputTokens": 156, + "accumulatedCachedInputTokens": 5_033, + } + }); + let raw2 = serde_json::json!({ + "sessionId": "cache-sess", + "update": { + "sessionUpdate": "usage_update", + "accumulatedInputTokens": 28_500, + "accumulatedOutputTokens": 310, + "accumulatedCachedInputTokens": 11_000, + } + }); + + let mut tracker = crate::usage::UsageTracker::default(); + + // Turn 1 — establish baseline (delta unreliable, but cumulative still present). + tracker.begin_turn("cache-sess"); + if let crate::usage::GooseSessionUpdateVariant::UsageUpdate(p) = + serde_json::from_value::(raw1) + .unwrap() + .update + { + tracker.record("cache-sess", &p); + } + let t1 = tracker.take().expect("turn 1"); + + // Turn 1: cumulative must carry the cache count; turn delta is None (no baseline). + let (turn1, cum1) = crate::pool::build_turn_metric_counts(&t1); + // delta_reliable = false on first turn → no turn counts. + assert!(turn1.is_none(), "first turn: no reliable turn counts"); + let cum1 = cum1.expect("cumulative always present"); + assert_eq!( + cum1.cache_read_tokens, + Some(5_033), + "cumulative.cacheReadTokens must be 5033 after turn 1" + ); + + // Turn 2 — delta reliable. + tracker.begin_turn("cache-sess"); + if let crate::usage::GooseSessionUpdateVariant::UsageUpdate(p) = + serde_json::from_value::(raw2) + .unwrap() + .update + { + tracker.record("cache-sess", &p); + } + let t2 = tracker.take().expect("turn 2"); + + let (turn2, cum2) = crate::pool::build_turn_metric_counts(&t2); + + let turn2 = turn2.expect("reliable turn counts on turn 2"); + // Per-turn cache delta: 11_000 - 5_033 = 5_967. + assert_eq!( + turn2.cache_read_tokens, + Some(5_967), + "turn.cacheReadTokens must be the per-turn delta" + ); + // cache_write_tokens is always None — buzz-agent doesn't emit it. + assert!( + turn2.cache_write_tokens.is_none(), + "cache_write_tokens must be None — not emitted by buzz-agent" + ); + + let cum2 = cum2.expect("cumulative always present"); + assert_eq!( + cum2.cache_read_tokens, + Some(11_000), + "cumulative.cacheReadTokens must be 11_000 after turn 2" + ); + assert!( + cum2.cache_write_tokens.is_none(), + "cache_write_tokens must be None on cumulative too" + ); + } + fn make_prompt_context_no_owner() -> PromptContext { let agent_keys = nostr::Keys::generate(); make_prompt_context_impl(&agent_keys, None) diff --git a/crates/buzz-acp/src/queue.rs b/crates/buzz-acp/src/queue.rs index 029bf86dbf..5c960de202 100644 --- a/crates/buzz-acp/src/queue.rs +++ b/crates/buzz-acp/src/queue.rs @@ -646,6 +646,11 @@ impl EventQueue { self.in_flight_channels.contains(&channel_id) } + /// Whether any channel currently has a turn in flight. + pub fn has_in_flight(&self) -> bool { + !self.in_flight_channels.is_empty() + } + // ── Goose-native steer withhold (side table) ────────────────────────── // // While a goose-native `_goose/unstable/session/steer` write is in flight diff --git a/crates/buzz-acp/src/usage.rs b/crates/buzz-acp/src/usage.rs index 1629eee935..56b772d12c 100644 --- a/crates/buzz-acp/src/usage.rs +++ b/crates/buzz-acp/src/usage.rs @@ -85,12 +85,16 @@ pub(crate) struct UsageUpdatePayload { pub context_limit: u64, pub accumulated_input_tokens: u64, pub accumulated_output_tokens: u64, - /// The cache-served subset of `accumulated_input_tokens`. Optional — goose - /// does not send it, and buzz-agent only reports a non-zero value when the - /// provider returned a cache split, so `0` legitimately means either "no - /// cache hits" or "provider reported none". - #[serde(default)] - pub accumulated_cached_input_tokens: u64, + /// The cache-served subset of `accumulated_input_tokens`. + /// + /// `None` when the harness did not include the field (e.g. goose, which + /// never emits it). `Some(0)` when the harness explicitly reported zero + /// cache hits. The distinction matters: `None` means "we don't know", + /// while `Some(0)` means "provider confirmed no cache was used". + /// + /// Do NOT use `#[serde(default)]` here — that would collapse the absent + /// case into `Some(0)` and destroy provenance in the append-only archive. + pub accumulated_cached_input_tokens: Option, pub accumulated_cost: Option, /// Session-cumulative genuine provider total tokens. Optional — only /// emitted by buzz-agent when every turn in the session so far supplied a @@ -125,6 +129,12 @@ struct SessionState { /// `None` when the session has never emitted a provider total (Unseen) or /// when any prior turn lacked one (poisoned). last_total: Option, + /// Cumulative cache-read input tokens at the end of the LAST PUBLISHED turn. + /// `None` when the harness has never reported this field (e.g. goose). + /// `Some(n)` when at least one payload included the field. Field-local: + /// a decrease in this counter taints only the cache-read delta, not + /// `delta_reliable` or the input/output deltas. + last_cached_input: Option, } /// Per-turn usage record exposed to `TurnCompletionGuard` for NIP-AM publishing. @@ -151,6 +161,12 @@ pub struct TurnUsage { /// Per-turn cost delta (`current − previous`); `None` when unreliable or /// either snapshot is missing. pub turn_cost_usd: Option, + /// Per-turn cache-read token delta (`current − previous`); `None` when no + /// baseline exists, either snapshot is `None` (harness did not report it), + /// or the cumulative counter decreased (field-local taint). Field-local: + /// a decrease here never flips `delta_reliable` or invalidates the + /// input/output deltas. + pub turn_cache_read_tokens: Option, /// Session-cumulative input tokens as reported by goose at end of turn. pub cumulative_input_tokens: u64, /// Session-cumulative output tokens as reported by goose at end of turn. @@ -160,6 +176,11 @@ pub struct TurnUsage { pub cumulative_total_tokens: Option, /// Session-cumulative estimated cost in USD; `None` if goose did not report it. pub cumulative_cost_usd: Option, + /// Session-cumulative cache-read input tokens as reported by buzz-agent. + /// `None` when the harness has never reported this field (e.g. goose or + /// any harness that omits `accumulatedCachedInputTokens`). + /// `Some(0)` when the harness reported zero cache hits. + pub cumulative_cache_read_tokens: Option, /// Effective model id for this turn (maps to NIP-AM `model`). `None` if the /// harness did not include the model in its usage notification. pub model: Option, @@ -239,6 +260,7 @@ impl UsageTracker { let current_output = payload.accumulated_output_tokens; let current_cost = payload.accumulated_cost; let current_total = payload.accumulated_total_tokens; + let current_cached_input = payload.accumulated_cached_input_tokens; // Determine whether this session is currently in-flight so we know // whether to set `pending`. We compute the delta regardless so that @@ -294,6 +316,21 @@ impl UsageTracker { None => None, // no baseline yet }; + // Cache-read token delta: field-local — never affects `delta_reliable` + // or the input/output deltas. Null when: no baseline exists, either + // snapshot is None (harness did not report the field), or the cumulative + // counter decreased (harness restart, overflow). + // Some(0) is a valid result when both snapshots are Some(0) — it means + // the harness confirmed zero cache hits this turn, not that data is absent. + let turn_cache_read = match self.sessions.get(session_id) { + Some(prev) => match (current_cached_input, prev.last_cached_input) { + (Some(cur), Some(p)) if cur >= p => Some(cur - p), + (Some(_), Some(_)) => None, // decrease → field-local taint + _ => None, // either snapshot absent → no delta + }, + None => None, // no baseline yet + }; + if is_in_flight { // In-flight-match: update pending with the latest cumulative values. // Baseline is NOT advanced here — it advances only on take(). @@ -305,10 +342,12 @@ impl UsageTracker { turn_output_tokens: turn_output, turn_total_tokens: turn_total, turn_cost_usd: turn_cost, + turn_cache_read_tokens: turn_cache_read, cumulative_input_tokens: current_input, cumulative_output_tokens: current_output, cumulative_total_tokens: current_total, cumulative_cost_usd: current_cost, + cumulative_cache_read_tokens: current_cached_input, model: payload.model.clone(), }); } else if self.in_flight_session.is_none() { @@ -327,6 +366,7 @@ impl UsageTracker { last_output: current_output, last_cost: current_cost, last_total: current_total, + last_cached_input: current_cached_input, }, ); } @@ -355,6 +395,7 @@ impl UsageTracker { last_output: record.cumulative_output_tokens, last_cost: record.cumulative_cost_usd, last_total: record.cumulative_total_tokens, + last_cached_input: record.cumulative_cache_read_tokens, }, ); Some(record) @@ -366,9 +407,9 @@ mod tests { use super::*; /// The camelCase key buzz-agent actually puts on the wire must land on the - /// field. A rename mismatch here would deserialize to the serde default of - /// 0, and every trial would price as if nothing had ever been cached — the - /// exact silent failure this field was added to remove. + /// field. A rename mismatch here would deserialize to None, and every trial + /// would be treated as "not reported" — the exact silent failure this field + /// was added to remove. #[test] fn cached_input_tokens_deserialize_from_the_wire_key() { let p: UsageUpdatePayload = serde_json::from_value(serde_json::json!({ @@ -379,13 +420,14 @@ mod tests { "accumulatedCachedInputTokens": 5_033, })) .expect("payload must deserialize"); - assert_eq!(p.accumulated_cached_input_tokens, 5_033); - assert!(p.accumulated_cached_input_tokens <= p.accumulated_input_tokens); + assert_eq!(p.accumulated_cached_input_tokens, Some(5_033)); + assert!(p.accumulated_cached_input_tokens.unwrap() <= p.accumulated_input_tokens); } - /// goose does not send the field; its payloads must still deserialize. + /// goose does not send the field; its payloads must deserialize with None — + /// not zero — so that "not reported" is preserved distinct from "reported zero". #[test] - fn a_payload_without_the_cache_field_defaults_to_zero() { + fn a_payload_without_the_cache_field_deserializes_as_none() { let p: UsageUpdatePayload = serde_json::from_value(serde_json::json!({ "used": 500, "contextLimit": 200_000, @@ -393,7 +435,28 @@ mod tests { "accumulatedOutputTokens": 100, })) .expect("payload must deserialize without the cache field"); - assert_eq!(p.accumulated_cached_input_tokens, 0); + assert!( + p.accumulated_cached_input_tokens.is_none(), + "absent field must be None, not Some(0)" + ); + } + + /// A harness that explicitly reports zero cache hits must produce Some(0), + /// not None — so downstream analytics can distinguish "confirmed zero" from + /// "not reported". + #[test] + fn a_payload_with_explicit_zero_cache_field_deserializes_as_some_zero() { + let p: UsageUpdatePayload = serde_json::from_value(serde_json::json!({ + "accumulatedInputTokens": 400, + "accumulatedOutputTokens": 100, + "accumulatedCachedInputTokens": 0, + })) + .expect("payload must deserialize with zero cache field"); + assert_eq!( + p.accumulated_cached_input_tokens, + Some(0), + "explicit zero must be Some(0), not None" + ); } fn payload(input: u64, output: u64, cost: Option) -> UsageUpdatePayload { @@ -402,7 +465,7 @@ mod tests { context_limit: 200_000, accumulated_input_tokens: input, accumulated_output_tokens: output, - accumulated_cached_input_tokens: 0, + accumulated_cached_input_tokens: None, accumulated_cost: cost, accumulated_total_tokens: None, model: None, @@ -415,7 +478,7 @@ mod tests { context_limit: 0, accumulated_input_tokens: input, accumulated_output_tokens: output, - accumulated_cached_input_tokens: 0, + accumulated_cached_input_tokens: None, accumulated_cost: cost, accumulated_total_tokens: None, model: None, @@ -913,7 +976,7 @@ mod tests { context_limit: 200_000, accumulated_input_tokens: input, accumulated_output_tokens: output, - accumulated_cached_input_tokens: 0, + accumulated_cached_input_tokens: None, accumulated_cost: cost, accumulated_total_tokens: None, model: model.map(str::to_string), @@ -977,7 +1040,7 @@ mod tests { context_limit: 200_000, accumulated_input_tokens: input, accumulated_output_tokens: output, - accumulated_cached_input_tokens: 0, + accumulated_cached_input_tokens: None, accumulated_cost: None, accumulated_total_tokens: total, model: None, @@ -1132,4 +1195,320 @@ mod tests { ); assert_eq!(usage.cumulative_total_tokens, Some(250)); } + + // ── cache-read token threading ────────────────────────────────────────── + + fn payload_with_cache( + input: u64, + output: u64, + cached_input: Option, + ) -> UsageUpdatePayload { + UsageUpdatePayload { + used: input + output, + context_limit: 200_000, + accumulated_input_tokens: input, + accumulated_output_tokens: output, + accumulated_cached_input_tokens: cached_input, + accumulated_cost: None, + accumulated_total_tokens: None, + model: None, + } + } + + #[test] + fn cache_read_first_turn_produces_none_turn_delta_and_passes_cumulative_through() { + // First turn has no baseline → turn cache delta must be None, but + // cumulative_cache_read_tokens must carry the reported value through. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c1"); + tracker.record("sess-c1", &payload_with_cache(1000, 200, Some(500))); + let usage = tracker.take().expect("pending"); + + assert!( + usage.turn_cache_read_tokens.is_none(), + "first turn: no baseline → cache delta must be None" + ); + assert_eq!( + usage.cumulative_cache_read_tokens, + Some(500), + "cumulative cache read passes through on first turn" + ); + assert!(!usage.delta_reliable, "first turn is unreliable"); + } + + #[test] + fn cache_read_second_turn_delta_computed_correctly() { + // Second turn: cumulative cached 500 → 1200, delta = 700. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c2"); + tracker.record("sess-c2", &payload_with_cache(1000, 200, Some(500))); + let _ = tracker.take(); + + tracker.begin_turn("sess-c2"); + tracker.record("sess-c2", &payload_with_cache(2000, 350, Some(1200))); + let usage = tracker.take().expect("pending"); + + assert!(usage.delta_reliable); + assert_eq!( + usage.turn_cache_read_tokens, + Some(700), + "cache delta = 1200 - 500 = 700" + ); + assert_eq!( + usage.cumulative_cache_read_tokens, + Some(1200), + "cumulative cache passes through" + ); + } + + #[test] + fn cache_read_decrease_nulls_turn_cache_but_leaves_delta_reliable() { + // Cache counter decrease → cache delta None (field-local taint), but + // delta_reliable and input/output deltas are NOT affected. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c3"); + tracker.record("sess-c3", &payload_with_cache(1000, 200, Some(800))); + let _ = tracker.take(); + + tracker.begin_turn("sess-c3"); + // Cache counter decreased: 800 → 50. + tracker.record("sess-c3", &payload_with_cache(1500, 300, Some(50))); + let usage = tracker.take().expect("pending"); + + assert!( + usage.delta_reliable, + "cache decrease must NOT flip delta_reliable — field-local" + ); + assert_eq!( + usage.turn_input_tokens, + Some(500), + "input/output delta unaffected by cache decrease" + ); + assert_eq!(usage.turn_output_tokens, Some(100)); + assert!( + usage.turn_cache_read_tokens.is_none(), + "cache counter decrease → turn_cache_read_tokens None (field-local taint)" + ); + assert_eq!( + usage.cumulative_cache_read_tokens, + Some(50), + "cumulative still passes through from payload even on decrease" + ); + } + + #[test] + fn cache_read_explicit_zero_payload_after_explicit_zero_baseline_produces_some_zero_delta() { + // When both baseline and current are Some(0), turn_cache_read_tokens must + // be Some(0) — confirmed zero, not absent. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c4"); + tracker.record("sess-c4", &payload_with_cache(1000, 200, Some(0))); + let _ = tracker.take(); + + tracker.begin_turn("sess-c4"); + tracker.record("sess-c4", &payload_with_cache(1500, 300, Some(0))); + let usage = tracker.take().expect("pending"); + + assert!(usage.delta_reliable); + assert_eq!( + usage.turn_cache_read_tokens, + Some(0), + "explicit zero on both sides → Some(0), not None" + ); + assert_eq!(usage.cumulative_cache_read_tokens, Some(0)); + } + + #[test] + fn cache_read_threads_through_setup_notification_baseline() { + // A setup notification (before begin_turn) with a nonzero cache count + // must update the committed baseline so the first real turn gets a + // correct delta from that starting point. + let mut tracker = UsageTracker::default(); + + // Setup notification: cumulative cache = 300. + tracker.record("sess-c5", &payload_with_cache(1000, 200, Some(300))); + + tracker.begin_turn("sess-c5"); + tracker.record("sess-c5", &payload_with_cache(1500, 350, Some(700))); + let usage = tracker.take().expect("pending"); + + assert!(usage.delta_reliable, "baseline from setup: reliable"); + assert_eq!( + usage.turn_cache_read_tokens, + Some(400), + "cache delta from setup baseline: 700 - 300 = 400" + ); + assert_eq!(usage.cumulative_cache_read_tokens, Some(700)); + } + + #[test] + fn cache_read_omitted_field_produces_none_cumulative_and_no_turn_delta() { + // A harness that omits accumulatedCachedInputTokens (e.g. goose) must + // produce None cumulative_cache_read_tokens — not Some(0) — and the + // turn delta must also be None even on the second turn. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c6"); + // payload() uses None for accumulated_cached_input_tokens. + tracker.record("sess-c6", &payload(1000, 200, None)); + let t1 = tracker.take().expect("turn 1"); + + assert!( + t1.cumulative_cache_read_tokens.is_none(), + "goose-shaped payload: cumulative must be None, not Some(0)" + ); + assert!( + t1.turn_cache_read_tokens.is_none(), + "first turn always has no turn delta" + ); + + tracker.begin_turn("sess-c6"); + tracker.record("sess-c6", &payload(1500, 300, None)); + let t2 = tracker.take().expect("turn 2"); + + assert!( + t2.cumulative_cache_read_tokens.is_none(), + "continued goose session: cumulative must remain None" + ); + assert!( + t2.turn_cache_read_tokens.is_none(), + "absent field on both sides → no turn delta invented" + ); + assert!( + t2.delta_reliable, + "input/output reliability unaffected by absent cache field" + ); + } + + #[test] + fn cache_read_baseline_absent_then_present_produces_no_delta() { + // If the first turn omits the cache field (baseline stored as None) and + // the second turn reports a value, no delta can be computed — we have no + // baseline to subtract from. The cumulative value should still pass through. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c7"); + tracker.record("sess-c7", &payload(1000, 200, None)); // no cache field + let _ = tracker.take(); + + tracker.begin_turn("sess-c7"); + tracker.record("sess-c7", &payload_with_cache(1500, 300, Some(400))); + let usage = tracker.take().expect("turn 2"); + + assert!( + usage.turn_cache_read_tokens.is_none(), + "absent baseline → no turn delta even when current has a value" + ); + assert_eq!( + usage.cumulative_cache_read_tokens, + Some(400), + "cumulative from current payload passes through" + ); + assert!(usage.delta_reliable, "input/output reliability unaffected"); + } + + #[test] + fn cache_read_baseline_present_then_absent_produces_no_delta() { + // If the first turn reports the cache field but the second omits it + // (harness switched), no delta should be produced and cumulative is None. + let mut tracker = UsageTracker::default(); + tracker.begin_turn("sess-c8"); + tracker.record("sess-c8", &payload_with_cache(1000, 200, Some(300))); + let _ = tracker.take(); + + tracker.begin_turn("sess-c8"); + tracker.record("sess-c8", &payload(1500, 300, None)); // no cache field + let usage = tracker.take().expect("turn 2"); + + assert!( + usage.turn_cache_read_tokens.is_none(), + "absent current → no turn delta" + ); + assert!( + usage.cumulative_cache_read_tokens.is_none(), + "absent field: cumulative must be None" + ); + assert!(usage.delta_reliable, "input/output reliability unaffected"); + } + + #[test] + fn pool_omitted_cache_field_publishes_no_cache_read_tokens_in_kind44200() { + // End-to-end: a buzz-agent or goose payload that omits the cache field + // must NOT publish cacheReadTokens in the kind:44200 event — neither + // in turn nor cumulative counts. + // + // This is the core acceptance test for Thufir's finding: the old code + // would publish cacheReadTokens: 0 for every harness regardless of + // whether the field was reported. + use crate::pool::build_turn_metric_counts; + + let usage = TurnUsage { + session_id: "sess-pool-none".into(), + turn_seq: 2, + delta_reliable: true, + turn_input_tokens: Some(400), + turn_output_tokens: Some(100), + turn_total_tokens: None, + turn_cost_usd: None, + turn_cache_read_tokens: None, + cumulative_input_tokens: 700, + cumulative_output_tokens: 200, + cumulative_total_tokens: None, + cumulative_cost_usd: None, + cumulative_cache_read_tokens: None, // harness did not report the field + model: None, + }; + + let (turn_counts, cumulative_counts) = build_turn_metric_counts(&usage); + + let turn = turn_counts.expect("turn counts must be present (delta reliable)"); + assert!( + turn.cache_read_tokens.is_none(), + "omitted cache field: turn cacheReadTokens must be absent from kind:44200" + ); + + let cumulative = cumulative_counts.expect("cumulative counts always present"); + assert!( + cumulative.cache_read_tokens.is_none(), + "omitted cache field: cumulative cacheReadTokens must be absent from kind:44200" + ); + } + + #[test] + fn pool_reported_cache_field_publishes_nonzero_cache_read_tokens_in_kind44200() { + // End-to-end: a buzz-agent payload with a nonzero cache count must + // publish cacheReadTokens in both turn and cumulative counts. + use crate::pool::build_turn_metric_counts; + + let usage = TurnUsage { + session_id: "sess-pool-some".into(), + turn_seq: 2, + delta_reliable: true, + turn_input_tokens: Some(400), + turn_output_tokens: Some(100), + turn_total_tokens: None, + turn_cost_usd: None, + turn_cache_read_tokens: Some(300), + cumulative_input_tokens: 700, + cumulative_output_tokens: 200, + cumulative_total_tokens: None, + cumulative_cost_usd: None, + cumulative_cache_read_tokens: Some(600), + model: None, + }; + + let (turn_counts, cumulative_counts) = build_turn_metric_counts(&usage); + + let turn = turn_counts.expect("turn counts present"); + assert_eq!( + turn.cache_read_tokens, + Some(300), + "nonzero turn cache: must appear in kind:44200 turn counts" + ); + + let cumulative = cumulative_counts.expect("cumulative counts present"); + assert_eq!( + cumulative.cache_read_tokens, + Some(600), + "nonzero cumulative cache: must appear in kind:44200 cumulative counts" + ); + } } diff --git a/crates/buzz-agent/src/agent.rs b/crates/buzz-agent/src/agent.rs index 8e14fee195..ff87a33a1a 100644 --- a/crates/buzz-agent/src/agent.rs +++ b/crates/buzz-agent/src/agent.rs @@ -13,8 +13,8 @@ use crate::mcp::McpRegistry; use crate::mcp::ResultBudget; use crate::types::{ - AgentError, ContentBlock, HistoryItem, ProviderStop, StopReason, ToolCall, ToolResult, - ToolResultContent, TurnTotalState, + AgentError, ContentBlock, HistoryItem, ProviderStop, SessionUsageBaseline, StopReason, + ToolCall, ToolResult, ToolResultContent, TurnTotalState, }; use crate::wire::{self, WireSender}; @@ -150,9 +150,40 @@ pub struct RunCtx<'a> { /// Reset to `Unseen` at turn start in `run()`. Callers must not derive a /// total by summing input+output — that is the UI display approximation only. pub turn_total_state: &'a mut TurnTotalState, + /// Session-cumulative counters as they stood when this turn began. Added to + /// the `turn_*` accumulators above to report a cumulative figure mid-turn; + /// the session's own copy is only advanced once, after the turn returns. + pub usage_baseline: SessionUsageBaseline, } impl RunCtx<'_> { + /// Send a session-cumulative `usage_update` reflecting everything observed + /// up to and including the most recent LLM response. + /// + /// The figure is the turn-start baseline plus this turn's running + /// accumulators, which is exactly what `session/prompt` will fold into the + /// session once the turn returns — so a mid-turn notification and the + /// end-of-turn one agree, and a turn that never returns has still reported + /// everything but its final in-flight request. + async fn emit_usage_update(&self) { + let base = self.usage_baseline; + let payload = wire::usage_update_payload( + base.input_tokens + .saturating_add(self.turn_input_tokens.unwrap_or(0)), + base.output_tokens + .saturating_add(self.turn_output_tokens.unwrap_or(0)), + base.cached_input_tokens + .saturating_add(self.turn_cached_input_tokens.unwrap_or(0)), + base.total_state.merge_session(*self.turn_total_state), + self.effective_model, + ); + wire::send( + self.wire, + wire::goose_session_update(self.session_id, payload), + ) + .await; + } + pub async fn run(&mut self, prompt: Vec) -> Result { let user_text = prompt_to_text(prompt)?; if user_text.len() > MAX_PROMPT_BYTES { @@ -299,6 +330,23 @@ impl RunCtx<'_> { // this gate rather than representing absent categories as zero. if response.input_tokens.is_some() || response.output_tokens.is_some() { *self.turn_total_state = self.turn_total_state.fold(response.total_tokens); + // Report what the turn has burned SO FAR, before running the + // next round. A turn is many provider round-trips over many + // minutes, and until this point the only report was the one + // `session/prompt` sends after the turn returns — so a turn + // that was cancelled, timed out, or whose process was killed + // reported nothing at all, and its tokens (already billed) + // existed only in this stack frame. Reporting per round bounds + // the loss to the single request in flight. + // + // Emitting more than one `usage_update` per turn is expected by + // the consumer: buzz-acp's UsageTracker advances its committed + // baseline only when the turn's metric is published, so every + // notification within a turn measures from the same frozen + // baseline and the last one seen is the turn's true total. + // goose behaves the same way, which is why the tracker was + // written to tolerate it. + self.emit_usage_update().await; } if !response.reasoning.is_empty() { diff --git a/crates/buzz-agent/src/lib.rs b/crates/buzz-agent/src/lib.rs index 0a06187fa6..15eb57e2df 100644 --- a/crates/buzz-agent/src/lib.rs +++ b/crates/buzz-agent/src/lib.rs @@ -660,6 +660,7 @@ async fn run_prompt(app: Arc, id: Value, params: Value, wire_tx: WireSender effective_model_override, run_id, mut steer_rx, + usage_baseline, ) = match acquire_session(&app, &p.session_id).await { Ok(v) => v, Err(reason) => { @@ -711,6 +712,7 @@ async fn run_prompt(app: Arc, id: Value, params: Value, wire_tx: WireSender turn_output_tokens: &mut turn_output_tokens, turn_cached_input_tokens: &mut turn_cached_input_tokens, turn_total_state: &mut turn_total_state, + usage_baseline, }; let result = ctx.run(p.prompt).await; if let Some(s) = app.sessions.lock().await.get_mut(&sid) { @@ -768,28 +770,16 @@ async fn run_prompt(app: Arc, id: Value, params: Value, wire_tx: WireSender if let Some((accumulated_in, accumulated_out, accumulated_cached, accumulated_total)) = accumulated { - // Build the usage_update payload. `accumulatedTotalTokens` is only - // included when the cumulative is exactly known — never when Unseen - // (no total ever observed) or Unknown (at least one turn lacked a - // total). A goose consumer that doesn't recognise the field ignores it. - let mut update = serde_json::json!({ - "sessionUpdate": "usage_update", - // used: total tokens as a context-usage proxy; - // contextLimit: 0 (buzz-agent has no context limit tracking). - "used": accumulated_in.saturating_add(accumulated_out), - "contextLimit": 0u64, - "accumulatedInputTokens": accumulated_in, - "accumulatedOutputTokens": accumulated_out, - // A subset of accumulatedInputTokens, not an addition to - // it. Extends goose's usage_update shape; a consumer that - // does not know the field ignores it and prices exactly as - // it did before. - "accumulatedCachedInputTokens": accumulated_cached, - "model": effective_model_str, - }); - if let crate::types::TurnTotalState::Exact(total) = accumulated_total { - update["accumulatedTotalTokens"] = serde_json::json!(total); - } + // Same builder the run loop uses for its per-round reports, so the + // final notification is shape-identical to the ones that preceded + // it and a consumer taking the high-water mark lands on this one. + let update = wire::usage_update_payload( + accumulated_in, + accumulated_out, + accumulated_cached, + accumulated_total, + effective_model_str, + ); wire::send(&wire_tx, goose_session_update(&sid, update)).await; } } @@ -823,6 +813,7 @@ async fn acquire_session( Option, String, mpsc::UnboundedReceiver>, + crate::types::SessionUsageBaseline, ), &'static str, > { @@ -859,6 +850,17 @@ async fn acquire_session( effective_model, run_id, steer_rx, + // Snapshot rather than a handle: the run loop reports cumulative usage + // after every LLM round, and taking the sessions lock on each of those + // would serialise concurrent sessions behind one another's provider + // round-trips. Nothing else advances these counters while this turn + // holds `busy`, so the snapshot cannot go stale under it. + crate::types::SessionUsageBaseline { + input_tokens: s.accumulated_input_tokens, + output_tokens: s.accumulated_output_tokens, + cached_input_tokens: s.accumulated_cached_input_tokens, + total_state: s.accumulated_total_state, + }, )) } diff --git a/crates/buzz-agent/src/types.rs b/crates/buzz-agent/src/types.rs index 343a75bf72..e386421981 100644 --- a/crates/buzz-agent/src/types.rs +++ b/crates/buzz-agent/src/types.rs @@ -308,6 +308,30 @@ impl TurnTotalState { } } +/// The session-cumulative usage counters as of the START of a turn. +/// +/// Copied out of the session under the lock when a turn begins and handed to +/// `RunCtx` by value, so the run loop can emit a cumulative `usage_update` +/// after every LLM round without reaching back into `App.sessions` (which it +/// holds no handle to, and which is locked by the turn's own bookkeeping at +/// both ends). +/// +/// This exists so that usage is durable *during* a turn rather than only after +/// it. The counters a turn accrues live in the prompt task's stack frame until +/// the turn returns; a process killed mid-turn takes them with it and the +/// tokens are billed by the provider but recorded nowhere. That is not +/// hypothetical — it silently under-reported a long-horizon benchmark's cost by +/// several-fold, because every phase of a `continue_until_timeout` run is +/// terminated mid-turn by design. +#[derive(Debug, Clone, Copy, Default)] +pub struct SessionUsageBaseline { + pub input_tokens: u64, + pub output_tokens: u64, + /// The cache-served subset of `input_tokens`, not an addition to it. + pub cached_input_tokens: u64, + pub total_state: TurnTotalState, +} + #[derive(Debug, Clone, Copy, PartialEq)] pub enum StopReason { EndTurn, diff --git a/crates/buzz-agent/src/wire.rs b/crates/buzz-agent/src/wire.rs index 7b50e7982a..634fca03af 100644 --- a/crates/buzz-agent/src/wire.rs +++ b/crates/buzz-agent/src/wire.rs @@ -148,6 +148,48 @@ pub fn goose_session_update(sid: &str, update: Value) -> Value { }) } +/// Build the `usage_update` payload for a `_goose/unstable/session/update`. +/// +/// Shared by the two places that report usage — after each LLM round inside a +/// turn, and once more when the turn completes — so the wire shape cannot drift +/// between them. A consumer takes the high-water mark per session, so the +/// mid-turn payloads are supersets of each other and the final one wins; a +/// divergence in field names or units between the two call sites would instead +/// show up as tokens silently vanishing, which is the failure this reporting +/// exists to prevent. +/// +/// All counts are SESSION-cumulative, matching goose, so buzz-acp's +/// `UsageTracker` can compute per-turn deltas symmetrically for both agents. +pub fn usage_update_payload( + accumulated_input_tokens: u64, + accumulated_output_tokens: u64, + accumulated_cached_input_tokens: u64, + accumulated_total: crate::types::TurnTotalState, + model: &str, +) -> Value { + let mut update = json!({ + "sessionUpdate": "usage_update", + // used: total tokens as a context-usage proxy; + // contextLimit: 0 (buzz-agent has no context limit tracking). + "used": accumulated_input_tokens.saturating_add(accumulated_output_tokens), + "contextLimit": 0u64, + "accumulatedInputTokens": accumulated_input_tokens, + "accumulatedOutputTokens": accumulated_output_tokens, + // A subset of accumulatedInputTokens, not an addition to it. Extends + // goose's usage_update shape; a consumer that does not know the field + // ignores it and prices exactly as it did before. + "accumulatedCachedInputTokens": accumulated_cached_input_tokens, + "model": model, + }); + // Only when the cumulative is exactly known — never when Unseen (no total + // ever observed) or Unknown (at least one turn lacked a total). A goose + // consumer that doesn't recognise the field ignores it. + if let Some(total) = accumulated_total.exact_value() { + update["accumulatedTotalTokens"] = json!(total); + } + update +} + /// A `session/update` notification carrying a `update._meta.goose.` field. /// Used to advertise `activeRunId` (so steer-capable clients can target the /// in-flight run) and `queuedSteer` (so they can correlate an accepted steer diff --git a/crates/buzz-agent/tests/fake_llm.rs b/crates/buzz-agent/tests/fake_llm.rs index f782a9d476..ef6f9d2d80 100644 --- a/crates/buzz-agent/tests/fake_llm.rs +++ b/crates/buzz-agent/tests/fake_llm.rs @@ -933,6 +933,135 @@ async fn no_usage_turn_emits_no_usage_notification() { h.shutdown().await; } +/// Usage must be reported after EVERY provider round, not only once the turn +/// returns. +/// +/// A turn is many provider round-trips over many minutes. While the only report +/// was the one `session/prompt` sends after the turn returns, a turn whose +/// process was killed mid-flight reported nothing at all: its counters lived in +/// the prompt task's stack frame, the provider had already billed them, and no +/// consumer ever saw them. That is not a corner case for a long-horizon +/// benchmark — every phase of a `continue_until_timeout` run is terminated +/// mid-turn by design, which under-reported one measured run's cost several-fold. +/// +/// Two rounds with distinct usage. The assertion that matters is the FIRST +/// notification: it must carry round 1's counts alone, proving it was sent +/// before round 2 had returned, so a kill between the rounds would still have +/// left round 1 on the wire. +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn usage_is_reported_after_each_round_not_only_at_turn_end() { + let url = spawn_fake_llm(vec![ + openai_tool_call_with_usage("call_round1", "fake__noop", json!({}), 15, 6), + openai_text_with_usage("done", 20, 8), + ]) + .await; + let mut h = Harness::spawn(&url).await; + let sid = init_session(&mut h).await; + + let p_id = h + .send( + "session/prompt", + json!({"sessionId": sid, "prompt": [{"type":"text","text":"go"}]}), + ) + .await; + + let (frames_before, response) = recv_until_with_drain(&mut h, |v| v["id"] == p_id).await; + assert_eq!( + response["result"]["stopReason"], "end_turn", + "turn must complete with end_turn" + ); + + let usage: Vec<&Value> = frames_before + .iter() + .filter(|v| is_usage_update(v)) + .collect(); + assert!( + usage.len() >= 2, + "expected a usage_update per round (2 rounds), got {}; frames: {frames_before:#?}", + usage.len() + ); + + // Round 1 alone — emitted while round 2 was still outstanding. + assert_eq!( + usage[0]["params"]["update"]["accumulatedInputTokens"], + json!(15u64), + "first notification must carry round 1's input tokens only" + ); + assert_eq!( + usage[0]["params"]["update"]["accumulatedOutputTokens"], + json!(6u64), + "first notification must carry round 1's output tokens only" + ); + + // The last one is the turn total and is what a high-water-mark consumer keeps. + let last = usage[usage.len() - 1]; + assert_eq!( + last["params"]["update"]["accumulatedInputTokens"], + json!(35u64), + "final notification must carry the turn total 15+20=35" + ); + assert_eq!( + last["params"]["update"]["accumulatedOutputTokens"], + json!(14u64), + "final notification must carry the turn total 6+8=14" + ); + + h.shutdown().await; +} + +/// A mid-turn report must be SESSION-cumulative, not turn-local. +/// +/// The baseline handed to the run loop is a snapshot taken when the turn began; +/// if it were dropped, a consumer taking the high-water mark per session would +/// see turn 2's first round (a small number) arrive after turn 1's total and +/// discard it, silently losing turn 2 for any turn that never completed. +#[tokio::test(flavor = "multi_thread", worker_threads = 2)] +async fn mid_turn_usage_includes_earlier_turns() { + let url = spawn_fake_llm(vec![ + openai_text_with_usage("turn one", 10, 5), + openai_tool_call_with_usage("call_t2", "fake__noop", json!({}), 20, 8), + openai_text_with_usage("turn two done", 30, 9), + ]) + .await; + let mut h = Harness::spawn(&url).await; + let sid = init_session(&mut h).await; + + let p1 = h + .send( + "session/prompt", + json!({"sessionId": sid, "prompt": [{"type":"text","text":"turn 1"}]}), + ) + .await; + let (_, _) = recv_until_with_drain(&mut h, |v| v["id"] == p1).await; + + let p2 = h + .send( + "session/prompt", + json!({"sessionId": sid, "prompt": [{"type":"text","text":"turn 2"}]}), + ) + .await; + let (frames_before, _) = recv_until_with_drain(&mut h, |v| v["id"] == p2).await; + + let first = frames_before + .iter() + .find(|v| is_usage_update(v)) + .unwrap_or_else(|| { + panic!("expected a usage_update during turn 2; frames: {frames_before:#?}") + }); + assert_eq!( + first["params"]["update"]["accumulatedInputTokens"], + json!(30u64), + "turn 2 round 1 must report 10 (turn 1) + 20 (this round), not 20" + ); + assert_eq!( + first["params"]["update"]["accumulatedOutputTokens"], + json!(13u64), + "turn 2 round 1 must report 5 (turn 1) + 8 (this round), not 8" + ); + + h.shutdown().await; +} + /// When a turn is cancelled AFTER the provider has already returned a response /// (so token counts are observed), buzz-agent must still emit the usage /// notification before the cancelled `session/prompt` response. diff --git a/crates/buzz-backend-kubernetes/Cargo.toml b/crates/buzz-backend-kubernetes/Cargo.toml new file mode 100644 index 0000000000..1cd17030db --- /dev/null +++ b/crates/buzz-backend-kubernetes/Cargo.toml @@ -0,0 +1,36 @@ +[package] +name = "buzz-backend-kubernetes" +version.workspace = true +edition.workspace = true +rust-version.workspace = true +license.workspace = true +description = "Kubernetes backend provider for Buzz remote agents (docs/remote-agents.md)" + +[[bin]] +name = "buzz-backend-kubernetes" +path = "src/main.rs" + +[dependencies] +kube = { workspace = true } +k8s-openapi = { workspace = true } +nostr = { workspace = true } +tokio = { workspace = true } +serde = { workspace = true } +serde_json = { workspace = true } +sha2 = { workspace = true } +hex = { workspace = true } +rand = { workspace = true } +chrono = { workspace = true } +http = "1" +http-body-util = "0.1" + +# Explicit rustls dep with the ring provider — required to install the +# process-level CryptoProvider at startup. Without it this binary panics on its +# first TLS connection to the apiserver: the release build compiles every +# sidecar in one cargo invocation (.github/workflows/release.yml), which unifies +# both ring and aws-lc-rs features and leaves rustls unable to auto-select a +# provider. Same dependency and reason as crates/buzz-cli/Cargo.toml. +rustls = { version = "0.23", default-features = false, features = ["ring", "std"] } + +[dev-dependencies] +tower = { workspace = true } diff --git a/crates/buzz-backend-kubernetes/src/classify.rs b/crates/buzz-backend-kubernetes/src/classify.rs new file mode 100644 index 0000000000..e9cfaba402 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/classify.rs @@ -0,0 +1,377 @@ +//! The deploy state machine (spec §Deploy State Machine), as a pure function. +//! +//! `classify` maps a verified observation plus the desired create intent to +//! one [`Action`]. It performs no I/O, so every row of the spec's table is a +//! unit test with no cluster. `reconcile` executes actions and re-enters. +//! +//! Two invariants are structural rather than remembered: +//! +//! * [`Action::Delete`] carries the [`Fence`] from the exact observation that +//! authorized it. There is no way to build a delete without one, so a later +//! helper cannot re-read and silently substitute a fresher fence. +//! * The pull-failure classifier ([`PullFailure`]) reaches only +//! [`Action::Report`] and [`Action::Observe`]. It is absent from +//! `Action::Delete`'s type, so "reason strings are never deletion +//! authority" is enforced by the compiler. + +use crate::intent::Fingerprint; + +/// The compare-and-delete fence: UID + resourceVersion from the observation +/// that authorized the deletion. A failed precondition means the object +/// changed since the read — re-enter, never retry the delete. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Fence { + pub uid: String, + pub resource_version: String, +} + +/// Why a pod that never started looks permanently broken. Reporting only. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum PullFailure { + /// Registry auth: a 403/401 `ImagePullBackOff` retries forever without + /// ever succeeding, so "the pull retries" is false for this case. + Unauthorized, + /// The digest or repository does not exist at that registry. + ManifestUnknown, + /// The image has no variant for the node's architecture. + ArchMismatch, +} + +/// The container's startup state, already decoded from pod status. Decoding +/// happens at the edge so this module stays free of API types. +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Startup { + /// `state.running` — the harness process is up. This, not pod phase, is + /// what "live" means. + Started, + /// Started once and reached a terminal phase (Succeeded/Failed). + Terminated, + /// Never started, and self-healing is plausible: unschedulable during + /// scale-from-zero, an image pull in progress, a transient + /// `CreateContainerConfigError` whose Secret exists. + NeverStartedRecoverable, + /// Never started, and the provider *verified* the cause — not a reason + /// string. Either the referenced Secret is confirmed absent by a + /// most-recent read, or the image reference is structurally invalid. + NeverStartedProvablyBroken, + /// Never started; the pull is failing in a way that will not self-heal. + /// Still recoverable in the *never delete* sense — this only changes what + /// we report and how long we wait. + NeverStartedPullFailing(PullFailure), +} + +/// A pod that passed identity and ownership verification: label-selected, +/// full-pubkey annotation equal to the derived pubkey, management marker +/// present. Constructing this type is the verification step's output, so an +/// unverified object cannot reach `classify` at all. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct VerifiedPod { + pub name: String, + pub fence: Fence, + /// Set once the apiserver accepts a delete. In Kubernetes there is no + /// `Terminating` phase — a pod being gracefully deleted stays in phase + /// `Running` for its whole grace period — so this must be checked + /// *before* startup state or the dying pod reads as the no-op row. + pub deletion_marked: bool, + pub startup: Startup, + /// The `buzz.block.xyz/create-intent` annotation as recorded at create. + /// `None` for a pod written before the annotation existed, which counts + /// as divergence. + pub recorded_intent: Option, +} + +/// What the reconciler should do next. +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum Action { + /// Create the pod, then wait for the harness container to start. + Create, + /// Compare-and-delete, poll for actual disappearance, then re-enter. + Delete { name: String, fence: Fence }, + /// Wait for a deletion already in flight, then re-enter. + AwaitDisappearance { name: String }, + /// Strict no-op: return this `agent_id`, mutate nothing. + NoOp { agent_id: String }, + /// Keep observing until started or the operation deadline expires; on + /// expiry report the latest condition. Never deletes, on this call or any + /// later one. + Observe { name: String }, + /// Surface an actionable condition immediately rather than burning the + /// deadline on a failure that will not self-heal. + Report { name: String, failure: PullFailure }, +} + +/// Apply the spec's ordered rules to one verified observation. +/// +/// `desired` is the freshly computed create intent; comparison is always +/// recorded-annotation vs freshly-computed, never a diff against the live pod +/// (admission defaulting would make every pod look divergent). +pub fn classify(observed: Option<&VerifiedPod>, desired: &Fingerprint) -> Action { + let Some(pod) = observed else { + // Row: no instance → create. First deploy, or after GC. + return Action::Create; + }; + + // Row: deletion-marked, ANY phase. Checked before startup state because + // there is no `Terminating` phase to match on — a gracefully deleting pod + // reports phase `Running` throughout its grace period, so testing startup + // first would mistake it for the live no-op row and return an id that + // evaporates. + if pod.deletion_marked { + return Action::AwaitDisappearance { + name: pod.name.clone(), + }; + } + + match &pod.startup { + // Row: live and started → strict no-op. Start must never kill a live + // agent mid-turn, whatever the fingerprint says. + Startup::Started => Action::NoOp { + agent_id: pod.name.clone(), + }, + + // Row: terminated → delete residue, then re-enter to create. This is + // the normal restart path — how a user revives a reaped agent. + Startup::Terminated => Action::Delete { + name: pod.name.clone(), + fence: pod.fence.clone(), + }, + + // Row: never started, provably non-recoverable → fenced replace. + // "Provably" means a verified absence or a structural defect, never a + // reason string. + Startup::NeverStartedProvablyBroken => Action::Delete { + name: pod.name.clone(), + fence: pod.fence.clone(), + }, + + // Inside the recoverable row: a pull that will not self-heal is + // reported immediately instead of consuming the 600s deadline. This + // changes reporting and wait behavior only — no delete authority. + Startup::NeverStartedPullFailing(failure) => Action::Report { + name: pod.name.clone(), + failure: *failure, + }, + + // Row: never started, recoverable — split on create-intent + // divergence. Divergence is evidence of a config change the user is + // waiting on, and it is the *only* thing that replaces a + // never-started pod. Pod age triggers nothing: any finite age + // threshold collides with Cluster Autoscaler's own pod-age delays, + // and delete-recreate resets exactly the age it keys on. + Startup::NeverStartedRecoverable => { + if pod.recorded_intent.as_ref() == Some(desired) { + Action::Observe { + name: pod.name.clone(), + } + } else { + Action::Delete { + name: pod.name.clone(), + fence: pod.fence.clone(), + } + } + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + + fn fp(seed: &str) -> Fingerprint { + Fingerprint::for_test(seed) + } + + fn pod(startup: Startup, intent: Option) -> VerifiedPod { + VerifiedPod { + name: "buzz-agent-abc123def456".into(), + fence: Fence { + uid: "uid-1".into(), + resource_version: "rv-1".into(), + }, + deletion_marked: false, + startup, + recorded_intent: intent, + } + } + + #[test] + fn no_instance_creates() { + assert_eq!(classify(None, &fp("a")), Action::Create); + } + + #[test] + fn started_pod_is_strict_no_op() { + let p = pod(Startup::Started, Some(fp("a"))); + assert_eq!( + classify(Some(&p), &fp("a")), + Action::NoOp { + agent_id: p.name.clone() + } + ); + } + + /// The asymmetry the spec states plainly: an edit cannot reach a started + /// pod until it exits, but it *can* reach a never-started one — the + /// never-started pod is the one the user is editing because it did not + /// start. + #[test] + fn started_pod_no_ops_even_when_intent_diverges() { + let p = pod(Startup::Started, Some(fp("old"))); + assert_eq!( + classify(Some(&p), &fp("new")), + Action::NoOp { + agent_id: p.name.clone() + } + ); + } + + /// In Kubernetes a gracefully deleting pod stays in phase `Running`. If + /// the deletion mark were checked after startup state, this pod would + /// take the no-op row and `deploy` would return an id that evaporates. + #[test] + fn deletion_mark_beats_every_startup_state() { + for startup in [ + Startup::Started, + Startup::Terminated, + Startup::NeverStartedRecoverable, + Startup::NeverStartedProvablyBroken, + Startup::NeverStartedPullFailing(PullFailure::Unauthorized), + ] { + let mut p = pod(startup.clone(), Some(fp("a"))); + p.deletion_marked = true; + assert_eq!( + classify(Some(&p), &fp("a")), + Action::AwaitDisappearance { + name: p.name.clone() + }, + "deletion mark ignored for {startup:?}" + ); + } + } + + #[test] + fn terminated_pod_is_replaced() { + let p = pod(Startup::Terminated, Some(fp("a"))); + assert_eq!( + classify(Some(&p), &fp("a")), + Action::Delete { + name: p.name.clone(), + fence: p.fence.clone() + } + ); + } + + /// A never-started winner is repairable: pod exists, Secret confirmed + /// absent, container never started. A later deploy must delete-recreate + /// rather than no-op — the test that pins started-not-phase as the no-op + /// criterion. + #[test] + fn provably_broken_never_started_pod_is_replaced() { + let p = pod(Startup::NeverStartedProvablyBroken, Some(fp("a"))); + assert_eq!( + classify(Some(&p), &fp("a")), + Action::Delete { + name: p.name.clone(), + fence: p.fence.clone() + } + ); + } + + /// The anti-livelock rule: identical desired intent means *never* delete, + /// however long the pod has been pending. Age is not an input to this + /// function at all, which is the strongest way to say so. + #[test] + fn recoverable_with_matching_intent_only_observes() { + let p = pod(Startup::NeverStartedRecoverable, Some(fp("same"))); + assert_eq!( + classify(Some(&p), &fp("same")), + Action::Observe { + name: p.name.clone() + } + ); + } + + /// Repeated identical Starts can never delete anything — the same + /// classification, arbitrarily many times. + #[test] + fn repeated_identical_starts_never_delete() { + let p = pod(Startup::NeverStartedRecoverable, Some(fp("same"))); + for _ in 0..100 { + assert!(!matches!( + classify(Some(&p), &fp("same")), + Action::Delete { .. } + )); + } + } + + /// The wedge escape: the user corrected a resource request or image, so + /// the never-started pod is built from configuration they have since + /// changed. Without this row the edit could never materialize. + #[test] + fn recoverable_with_divergent_intent_is_replaced() { + let p = pod(Startup::NeverStartedRecoverable, Some(fp("old"))); + assert_eq!( + classify(Some(&p), &fp("new")), + Action::Delete { + name: p.name.clone(), + fence: p.fence.clone() + } + ); + } + + /// A pod predating the annotation has no recorded intent — that is + /// absence, which the spec groups with divergence. + #[test] + fn missing_recorded_intent_counts_as_divergence() { + let p = pod(Startup::NeverStartedRecoverable, None); + assert_eq!( + classify(Some(&p), &fp("any")), + Action::Delete { + name: p.name.clone(), + fence: p.fence.clone() + } + ); + } + + /// Permanent-looking pull failures report immediately instead of burning + /// 600s — and, critically, never delete. + #[test] + fn pull_failures_report_and_never_delete() { + for failure in [ + PullFailure::Unauthorized, + PullFailure::ManifestUnknown, + PullFailure::ArchMismatch, + ] { + let p = pod(Startup::NeverStartedPullFailing(failure), Some(fp("a"))); + // Divergent intent too — still no delete from this arm. + for desired in [fp("a"), fp("different")] { + assert_eq!( + classify(Some(&p), &desired), + Action::Report { + name: p.name.clone(), + failure + } + ); + } + } + } + + /// Every delete carries the fence from the observation that authorized + /// it. Exhaustive over the delete-producing states, so a future arm that + /// forgets is caught here rather than in a cluster. + #[test] + fn every_delete_carries_the_authorizing_fence() { + let states = [ + (Startup::Terminated, fp("a")), + (Startup::NeverStartedProvablyBroken, fp("a")), + (Startup::NeverStartedRecoverable, fp("divergent")), + ]; + for (startup, desired) in states { + let p = pod(startup, Some(fp("a"))); + match classify(Some(&p), &desired) { + Action::Delete { fence, .. } => assert_eq!(fence, p.fence), + other => panic!("expected Delete, got {other:?}"), + } + } + } +} diff --git a/crates/buzz-backend-kubernetes/src/client.rs b/crates/buzz-backend-kubernetes/src/client.rs new file mode 100644 index 0000000000..0c3bed65b7 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/client.rs @@ -0,0 +1,182 @@ +//! Cluster auth and client construction (spec §Cluster auth, +//! `docs/remote-agents.md:985-995`). +//! +//! Standard kubeconfig resolution (`$KUBECONFIG` → `~/.kube/config`). +//! `provider_config` carries `context` and `namespace` only — credentials +//! never transit config (I2, `:196-198`). + +use kube::config::{ExecConfig, KubeConfigOptions, Kubeconfig}; +use kube::{Client, Config}; +use std::path::{Path, PathBuf}; + +/// Directories prepended to `PATH` before the client is built. +/// +/// Kubeconfigs at Block near-universally authenticate through `exec` +/// credential plugins (`aws eks get-token`, `gke-gcloud-auth-plugin`) that +/// resolve via `PATH` — and this provider inherits a Finder-launched +/// desktop's minimal `PATH`, which contains none of the places those plugins +/// install to (`:989-994`). +const PATH_PREPEND: [&str; 2] = ["/opt/homebrew/bin", "/usr/local/bin"]; + +/// Compute the new `PATH` value: plugin directories first, inherited entries +/// after, in order. Pure so the ordering can be tested without mutating the +/// process's environment. +fn prepended_path(home: Option<&Path>, existing: &std::ffi::OsStr) -> Option { + let mut dirs: Vec = PATH_PREPEND.iter().map(PathBuf::from).collect(); + if let Some(home) = home { + dirs.push(home.join(".local/bin")); + } + // An empty inherited PATH splits into one empty entry, which POSIX + // resolves as the current directory — a place a credential plugin should + // never be looked up. Drop empties rather than propagate them. + dirs.extend(std::env::split_paths(existing).filter(|p| !p.as_os_str().is_empty())); + std::env::join_paths(dirs).ok() +} + +/// Prepend the plugin directories to this process's `PATH`. +/// +/// Modifies the provider's own environment, which is sound here: one process +/// per operation, called before any client or task exists, and the child +/// processes that read it are exactly the credential plugins this exists for. +fn prepend_plugin_path() { + let home = std::env::var_os("HOME"); + let existing = std::env::var_os("PATH").unwrap_or_default(); + if let Some(joined) = prepended_path(home.as_ref().map(Path::new), &existing) { + std::env::set_var("PATH", joined); + } +} + +/// Is `command` runnable — an executable on `PATH`, or an existing path? +fn resolves_on_path(command: &str) -> bool { + if command.contains(std::path::MAIN_SEPARATOR) { + return Path::new(command).is_file(); + } + std::env::var_os("PATH") + .map(|path| std::env::split_paths(&path).any(|dir| dir.join(command).is_file())) + .unwrap_or(false) +} + +/// The exec plugin the selected context authenticates with, if any. +/// +/// Read from the kubeconfig directly rather than from `Config`, which does not +/// expose it. A read failure yields `None`: this lookup exists only to improve +/// an error message, and must never be the thing that fails a deploy. +fn exec_plugin_for(context: Option<&str>) -> Option { + let kubeconfig = Kubeconfig::read().ok()?; + let context_name = context + .map(str::to_string) + .or_else(|| kubeconfig.current_context.clone())?; + let user_name = kubeconfig + .contexts + .iter() + .find(|c| c.name == context_name) + .and_then(|c| c.context.as_ref()) + .and_then(|c| c.user.clone())?; + kubeconfig + .auth_infos + .iter() + .find(|a| a.name == user_name) + .and_then(|a| a.auth_info.as_ref()) + .and_then(|a| a.exec.clone()) +} + +/// Turn a client-construction failure into an error a user can act on. +/// +/// When the context authenticates through an exec plugin that is not on +/// `PATH`, that is almost always the cause, and the actionable fact is the +/// plugin's name — not a kube-rs error chain (`:994-995`). +fn explain(context: Option<&str>, error: &kube::Error) -> String { + if let Some(command) = exec_plugin_for(context).and_then(|e| e.command) { + if !resolves_on_path(&command) { + return format!( + "kubeconfig context {} authenticates with the credential plugin \ + {command:?}, which is not on PATH. Install it or add its \ + directory to PATH, then try again.", + context.unwrap_or("(current)") + ); + } + } + format!( + "could not connect to the cluster using kubeconfig context {}: {error}", + context.unwrap_or("(current)") + ) +} + +/// Build a client for the selected context. +pub async fn connect(context: Option<&str>) -> Result { + prepend_plugin_path(); + + let options = KubeConfigOptions { + context: context.map(str::to_string), + ..Default::default() + }; + let config = Config::from_kubeconfig(&options).await.map_err(|e| { + // A named context that does not exist is a user typo, and the + // kube-rs message for it is already specific. + format!( + "could not load kubeconfig for context {}: {e}", + context.unwrap_or("(current)") + ) + })?; + + Client::try_from(config).map_err(|e| explain(context, &e)) +} + +#[cfg(test)] +mod tests { + use super::*; + + /// The three plugin directories must end up ahead of the inherited PATH, + /// or a Finder-launched desktop never finds `aws`/`gke-gcloud-auth-plugin`. + /// Tested on the pure computation: mutating the process PATH here would + /// race every other test in the binary. + #[test] + fn plugin_directories_are_prepended_in_order() { + let joined = prepended_path( + Some(Path::new("/tmp/fake-home")), + std::ffi::OsStr::new("/inherited/bin:/usr/bin"), + ) + .unwrap(); + let dirs: Vec = std::env::split_paths(&joined).collect(); + assert_eq!( + dirs, + [ + "/opt/homebrew/bin", + "/usr/local/bin", + "/tmp/fake-home/.local/bin", + "/inherited/bin", + "/usr/bin", + ] + .map(PathBuf::from) + ); + } + + /// No `HOME` is not a failure — the two absolute directories still apply. + #[test] + fn missing_home_still_prepends_the_absolute_directories() { + let joined = prepended_path(None, std::ffi::OsStr::new("/inherited/bin")).unwrap(); + let dirs: Vec = std::env::split_paths(&joined).collect(); + assert_eq!( + dirs, + ["/opt/homebrew/bin", "/usr/local/bin", "/inherited/bin"].map(PathBuf::from) + ); + } + + /// An empty inherited PATH must not produce an empty entry, which the + /// shell and `resolves_on_path` would both read as the cwd. + #[test] + fn empty_inherited_path_yields_no_empty_entry() { + let joined = prepended_path(None, std::ffi::OsStr::new("")).unwrap(); + let dirs: Vec = std::env::split_paths(&joined).collect(); + assert_eq!( + dirs, + ["/opt/homebrew/bin", "/usr/local/bin"].map(PathBuf::from) + ); + } + + #[test] + fn resolves_absolute_paths_directly() { + assert!(resolves_on_path("/bin/sh")); + assert!(!resolves_on_path("/nonexistent/plugin-binary")); + } +} diff --git a/crates/buzz-backend-kubernetes/src/cluster.rs b/crates/buzz-backend-kubernetes/src/cluster.rs new file mode 100644 index 0000000000..755f41f6c8 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/cluster.rs @@ -0,0 +1,427 @@ +//! The real [`Substrate`]: kube-rs against a live apiserver. +//! +//! Everything that *decides* lives in `classify`/`gc`; this module only +//! performs I/O and maps apiserver responses onto the trait's vocabulary. +//! Three mappings here are normative rather than incidental: +//! +//! * **409 is discriminated on `Status.reason`, never on the code.** A create +//! 409 is `AlreadyExists`; a delete 409 from a failed precondition is +//! `Conflict`. Branching on `code == 409` conflates a lost create race with +//! a stale fence and is the trap the spec names (`:780-794`). +//! * **Reads leave `resourceVersion` unset**, which is the quorum read. `"0"` +//! is the cache read, and a confirmed absence from a cache is proof of +//! nothing (`:761-769`). +//! * **Deletes never set `grace_period_seconds`**, so the object's own 60s +//! budget applies. Passing `0` is a force-kill that discards the shutdown +//! window the pod declares (`:1185-1189`). + +use crate::classify::Fence; +use crate::reconcile::{CreateOutcome, DeleteOutcome, Substrate}; +use chrono::{DateTime, Utc}; +use k8s_openapi::api::core::v1::{Namespace, Pod, Secret}; +use kube::api::{Api, DeleteParams, GetParams, ListParams, PostParams, Preconditions}; +use kube::core::ErrorResponse; +use kube::{Client, Resource}; +use std::time::{Duration, Instant}; + +/// `Status.reason` values we branch on. Spelled once so the two 409 arms are +/// visibly the same discriminator read two ways. +/// +/// These are wire strings `apimachinery` chooses, not names this crate picks: +/// `StatusReasonAlreadyExists`, `StatusReasonConflict`, `StatusReasonNotFound`, +/// and `StatusReasonForbidden` in `k8s.io/apimachinery/pkg/apis/meta/v1/types.go`. +/// kube-core types `ErrorResponse::reason` as a bare `String`, so there is no +/// upstream constant to bind to and the spelling is pinned by test instead. +const REASON_ALREADY_EXISTS: &str = "AlreadyExists"; +const REASON_CONFLICT: &str = "Conflict"; +const REASON_NOT_FOUND: &str = "NotFound"; +const REASON_FORBIDDEN: &str = "Forbidden"; + +/// The apiserver-backed substrate for one deploy operation. +pub struct Cluster { + client: Client, + namespace: String, + /// Start of *this operation*, for the deadline. Monotonic: the 600s budget + /// must not move when the wall clock does. + started: Instant, +} + +/// The typed API error underneath a `kube::Error`, if it is one. +fn api_error(error: &kube::Error) -> Option<&ErrorResponse> { + match error { + kube::Error::Api(response) => Some(response), + _ => None, + } +} + +/// Does this error carry the given `Status.reason`? +fn reason_is(error: &kube::Error, reason: &str) -> bool { + api_error(error).is_some_and(|e| e.reason == reason) +} + +impl Cluster { + pub fn new(client: Client, namespace: &str) -> Self { + Self { + client, + namespace: namespace.to_string(), + started: Instant::now(), + } + } + + fn pods(&self) -> Api { + Api::namespaced(self.client.clone(), &self.namespace) + } + + fn secrets(&self) -> Api { + Api::namespaced(self.client.clone(), &self.namespace) + } + + /// List an object kind through the raw client so the response's HTTP + /// `Date` header is reachable. + /// + /// `Api::list` returns only the decoded body, and the apiserver's clock is + /// the *only* clock the orphan-Secret age gate may use — a desktop's local + /// clock running fast computes every in-flight Secret as expired + /// (`:1321-1335`). So the list goes through `Client::send`, which hands + /// back the whole `http::Response`. + async fn list_with_date( + &self, + selector: &str, + ) -> Result<(Vec, Option>), String> + where + K: Resource + + Clone + + serde::de::DeserializeOwned + + std::fmt::Debug, + K::DynamicType: Default, + { + let dt = K::DynamicType::default(); + let url = K::url_path(&dt, Some(&self.namespace)); + // resourceVersion deliberately unset: quorum read. + let params = ListParams { + label_selector: Some(selector.to_string()), + ..Default::default() + }; + let request = kube::core::Request::new(url) + .list(¶ms) + .map_err(|e| format!("could not build a list request: {e}"))?; + let (parts, body) = request.into_parts(); + let response = self + .client + .send(http::Request::from_parts(parts, body.into())) + .await + .map_err(|e| format!("could not list {}: {e}", K::plural(&dt)))?; + + // Parsed before the body is consumed, and independently of it: a + // missing or malformed header is not a list failure, it just means the + // orphan sweep has no clock and skips. + let server_now = response + .headers() + .get(http::header::DATE) + .and_then(|v| v.to_str().ok()) + .and_then(|v| DateTime::parse_from_rfc2822(v).ok()) + .map(|v| v.with_timezone(&Utc)); + + let bytes = http_body_util::BodyExt::collect(response.into_body()) + .await + .map_err(|e| format!("could not read the {} list body: {e}", K::plural(&dt)))? + .to_bytes(); + let list: kube::core::ObjectList = serde_json::from_slice(&bytes) + .map_err(|e| format!("could not decode the {} list: {e}", K::plural(&dt)))?; + + Ok((list.items, server_now)) + } +} + +impl Substrate for Cluster { + async fn ensure_namespace(&self, namespace: &str) -> Result<(), String> { + let api: Api = Api::all(self.client.clone()); + if api + .get_opt(namespace) + .await + .map_err(|e| format!("could not check whether namespace {namespace} exists: {e}"))? + .is_some() + { + return Ok(()); + } + + let spec = Namespace { + metadata: kube::core::ObjectMeta { + name: Some(namespace.to_string()), + ..Default::default() + }, + ..Default::default() + }; + match api.create(&PostParams::default(), &spec).await { + Ok(_) => Ok(()), + // Someone else created it between our check and our create. That + // is the desired end state, not a failure. + Err(e) if reason_is(&e, REASON_ALREADY_EXISTS) => Ok(()), + // Namespace-create is frequently denied on shared clusters. Name + // the exact command an operator runs, and never silently fall back + // to `default` — deploying an agent into someone else's namespace + // is worse than refusing (`:1002-1005`). + Err(e) if reason_is(&e, REASON_FORBIDDEN) => Err(format!( + "not authorized to create namespace {namespace}. Ask a cluster \ + administrator to run `kubectl create namespace {namespace}`, \ + then try again." + )), + Err(e) => Err(format!("could not create namespace {namespace}: {e}")), + } + } + + async fn list_pods(&self, selector: &str) -> Result<(Vec, Option>), String> { + self.list_with_date::(selector).await + } + + async fn list_secrets(&self, selector: &str) -> Result, String> { + Ok(self.list_with_date::(selector).await?.0) + } + + async fn secret_exists(&self, name: &str) -> Result { + // `GetParams::default()` leaves resourceVersion unset — the quorum + // read this check requires to be proof of anything. + match self.secrets().get_with(name, &GetParams::default()).await { + Ok(_) => Ok(true), + Err(e) if reason_is(&e, REASON_NOT_FOUND) => Ok(false), + Err(e) => Err(format!("could not check whether secret {name} exists: {e}")), + } + } + + async fn create_secret(&self, secret: &Secret) -> Result<(), String> { + let name = secret.metadata.name.clone().unwrap_or_default(); + self.secrets() + .create(&PostParams::default(), secret) + .await + .map(|_| ()) + .map_err(|e| format!("could not create secret {name}: {e}")) + } + + async fn create_pod(&self, pod: &Pod) -> Result { + let name = pod.metadata.name.clone().unwrap_or_default(); + match self.pods().create(&PostParams::default(), pod).await { + Ok(_) => Ok(CreateOutcome::Created), + // The deterministic name is taken: a concurrent attempt won the + // election. Discriminated on the reason — a 409 whose reason is + // `Conflict` is a different condition and must not be read as a + // lost race. + Err(e) if reason_is(&e, REASON_ALREADY_EXISTS) => Ok(CreateOutcome::AlreadyExists), + Err(e) => Err(format!("could not create pod {name}: {e}")), + } + } + + async fn delete_pod(&self, name: &str, fence: &Fence) -> Result { + let params = DeleteParams { + preconditions: Some(Preconditions { + uid: Some(fence.uid.clone()), + resource_version: Some(fence.resource_version.clone()), + }), + // grace_period_seconds deliberately unset: the pod's own 60s + // budget applies. + ..Default::default() + }; + match self.pods().delete(name, ¶ms).await { + Ok(_) => Ok(DeleteOutcome::Accepted), + Err(e) if reason_is(&e, REASON_NOT_FOUND) => Ok(DeleteOutcome::NotFound), + // The object changed since the observation that authorized this + // delete. Same HTTP code as the create race above, different + // reason, different meaning. + Err(e) if reason_is(&e, REASON_CONFLICT) => Ok(DeleteOutcome::PreconditionFailed), + Err(e) => Err(format!("could not delete pod {name}: {e}")), + } + } + + async fn delete_secret(&self, name: &str) -> Result<(), String> { + match self.secrets().delete(name, &DeleteParams::default()).await { + Ok(_) => Ok(()), + // Already gone is the desired end state. + Err(e) if reason_is(&e, REASON_NOT_FOUND) => Ok(()), + Err(e) => Err(format!("could not delete secret {name}: {e}")), + } + } + + async fn get_pod(&self, name: &str) -> Result, String> { + match self.pods().get_with(name, &GetParams::default()).await { + Ok(pod) => Ok(Some(pod)), + Err(e) if reason_is(&e, REASON_NOT_FOUND) => Ok(None), + Err(e) => Err(format!("could not read pod {name}: {e}")), + } + } + + async fn sleep(&self, duration: Duration) { + tokio::time::sleep(duration).await; + } + + fn elapsed(&self) -> Duration { + self.started.elapsed() + } +} + +#[cfg(test)] +mod tests { + use super::*; + use http::Response; + use kube::client::Body; + use std::sync::{Arc, Mutex}; + use tower::service_fn; + + fn list_response(date: Option<&str>) -> Response { + let mut response = Response::builder().status(200); + if let Some(date) = date { + response = response.header(http::header::DATE, date); + } + response + .body(Body::from( + serde_json::to_vec(&serde_json::json!({ + "apiVersion": "v1", + "kind": "PodList", + "metadata": {"resourceVersion": "17"}, + "items": [{ + "apiVersion": "v1", + "kind": "Pod", + "metadata": {"name": "sprig"}, + "spec": {"containers": [{"name": "agent", "image": "example.invalid/sprig@sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"}]} + }] + })) + .unwrap(), + )) + .unwrap() + } + + async fn list_through_real_request_path( + date: Option<&'static str>, + ) -> (Vec, Option>, String) { + let observed_uri = Arc::new(Mutex::new(None)); + let service_uri = Arc::clone(&observed_uri); + let service = service_fn(move |request: http::Request| { + let service_uri = Arc::clone(&service_uri); + async move { + *service_uri.lock().unwrap() = Some(request.uri().to_string()); + Ok::<_, std::convert::Infallible>(list_response(date)) + } + }); + let cluster = Cluster::new(Client::new(service, "ignored"), "owned-ns"); + let result = cluster + .list_with_date::("app.kubernetes.io/managed-by=buzz-backend-kubernetes") + .await + .unwrap(); + let uri = observed_uri.lock().unwrap().take().unwrap(); + (result.0, result.1, uri) + } + + /// Exercise the shipped `Request` + `Client::send` seam. A fake + /// reconciler would not prove that kube-rs emits a quorum list request or + /// that the apiserver's clock survives body decoding. + #[tokio::test] + async fn list_with_date_uses_a_quorum_request_and_returns_the_server_clock() { + let (pods, server_now, uri) = + list_through_real_request_path(Some("Sun, 02 Aug 2026 04:00:00 GMT")).await; + + assert_eq!(pods.len(), 1, "fixture must contain one decoded pod"); + assert_eq!(pods[0].metadata.name.as_deref(), Some("sprig")); + assert!(uri.starts_with("/api/v1/namespaces/owned-ns/pods?")); + assert!( + uri.contains("labelSelector=app.kubernetes.io%2Fmanaged-by%3Dbuzz-backend-kubernetes") + ); + assert!( + !uri.contains("resourceVersion"), + "cache read leaked into {uri}" + ); + assert_eq!( + server_now.unwrap().to_rfc3339(), + "2026-08-02T04:00:00+00:00" + ); + } + + /// Header failure is deliberately not list failure: without a trustworthy + /// apiserver clock the orphan sweep skips, but normal reconciliation still + /// receives the decoded objects. + #[tokio::test] + async fn list_with_date_keeps_items_when_the_server_clock_is_unusable() { + for date in [Some("not a date"), None] { + let (pods, server_now, _) = list_through_real_request_path(date).await; + assert_eq!(pods.len(), 1, "fixture must contain one decoded pod"); + assert!(server_now.is_none(), "unexpected clock for {date:?}"); + } + } + + /// A typed apiserver error, as kube-rs surfaces it. + fn api(reason: &str, code: u16) -> kube::Error { + kube::Error::Api(ErrorResponse { + status: "Failure".into(), + message: String::new(), + reason: reason.into(), + code, + }) + } + + /// The one discriminator the whole file rests on, and the trap the spec + /// predicts: "an implementation that branches on the code alone will + /// eventually take the adoption path on a failed delete or vice versa" + /// (`:788-790`). + /// + /// Both of these are 409. Reading the *code* makes them identical; reading + /// `Status.reason` keeps a lost create race and a stale fence apart. The + /// mutation that must fail this test is `e.reason == …` → `e.code == 409`, + /// which no other test in the crate would catch — the fakes never produce + /// a real `kube::Error`. + #[test] + fn the_two_409s_are_never_conflated() { + let already_exists = api(REASON_ALREADY_EXISTS, 409); + let conflict = api(REASON_CONFLICT, 409); + + assert!(reason_is(&already_exists, REASON_ALREADY_EXISTS)); + assert!(reason_is(&conflict, REASON_CONFLICT)); + // The cross terms are the whole point. + assert!(!reason_is(&already_exists, REASON_CONFLICT)); + assert!(!reason_is(&conflict, REASON_ALREADY_EXISTS)); + } + + /// A transport-level failure is not an apiserver verdict. It must fall + /// through to the error arm rather than being read as any reason — a + /// connection reset silently classified as `NotFound` would report a pod + /// as confirmed-absent, which the classifier treats as proof. + #[test] + fn a_non_api_error_carries_no_reason() { + let transport = kube::Error::LinesCodecMaxLineLengthExceeded; + assert!(api_error(&transport).is_none()); + for reason in [ + REASON_ALREADY_EXISTS, + REASON_CONFLICT, + REASON_NOT_FOUND, + REASON_FORBIDDEN, + ] { + assert!(!reason_is(&transport, reason), "matched {reason}"); + } + } + + /// `reason` is `#[serde(default)]` in kube-core, so an apiserver that + /// omits it yields an empty string. That must match nothing rather than + /// matching an empty pattern by accident. + #[test] + fn an_absent_reason_matches_nothing() { + let bare = api("", 409); + assert!(!reason_is(&bare, REASON_ALREADY_EXISTS)); + assert!(!reason_is(&bare, REASON_CONFLICT)); + } + + /// The consts are the apiserver's spelling, asserted against literals + /// rather than against themselves. + /// + /// Every other test here references the consts symbolically on both sides + /// — fixture *and* assertion — which is true for any pair of distinct + /// values. That tests the discriminator is self-consistent, not that it is + /// correct: swapping the two 409 values inverts `AlreadyExists` and + /// `Conflict` at a real apiserver (`:788-790`'s failure, reached by + /// editing a string instead of a branch) with every other test still + /// green. These are `apimachinery`'s wire strings and kube-core exposes no + /// constant for them, so a literal is the only external anchor available. + /// Found by Quinn's mutation matrix; M2/M3/M4 survived without it. + #[test] + fn the_reason_consts_are_the_apiservers_spelling() { + assert_eq!(REASON_ALREADY_EXISTS, "AlreadyExists"); + assert_eq!(REASON_CONFLICT, "Conflict"); + assert_eq!(REASON_NOT_FOUND, "NotFound"); + assert_eq!(REASON_FORBIDDEN, "Forbidden"); + } +} diff --git a/crates/buzz-backend-kubernetes/src/config.rs b/crates/buzz-backend-kubernetes/src/config.rs new file mode 100644 index 0000000000..4d96735b7b --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/config.rs @@ -0,0 +1,470 @@ +//! `provider_config` parsing and the `info` config schema +//! (spec §`provider_config` v1 fields, `docs/remote-agents.md:1384-1389`). +//! +//! Nine fields, all optional except `image` (required at parse time; the +//! schema offers the published sprig image as a prefill default — §Image). +//! No credential field exists, by I2: cluster auth comes from ambient +//! kubeconfig resolution and nothing else (`:196-198`). + +use crate::image::{self, ImageRef}; + +/// Resource requests and limits (§Pod shape: 1cpu/2Gi → 2cpu/4Gi, all four +/// configurable — `cargo build` in an agent workspace makes 500m/1Gi +/// unrealistic). +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Resources { + pub cpu_request: String, + pub memory_request: String, + pub cpu_limit: String, + pub memory_limit: String, +} + +impl Default for Resources { + fn default() -> Self { + Self { + cpu_request: "1".into(), + memory_request: "2Gi".into(), + cpu_limit: "2".into(), + memory_limit: "4Gi".into(), + } + } +} + +/// Default inactivity budget: the I5 opt-in (§Auto-Stop). The config field and +/// `BUZZ_ACP_EXIT_AFTER_INACTIVITY` are one knob, not two. +pub const DEFAULT_INACTIVITY_SECONDS: u64 = 7200; + +/// Default `image` schema prefill: the published sprig image, in tag+digest +/// form so the tag stays human-traceable to its git SHA while the digest does +/// the pinning (§Image — tag-only refs are rejected; `image::parse` drops the +/// tag on normalization). This is a UI prefill, not a baked fallback: `image` +/// stays required, an empty value still fails closed, and the value always +/// arrives explicitly in `provider_config`, so create-intent fingerprints are +/// unaffected by provider upgrades. +pub const DEFAULT_IMAGE: &str = "ghcr.io/block/buzz-sprig:sha-6530b58@sha256:17facfc7608d8ddb33bc056c9aaba1098f4ef6abe5655702fbfd7584d1f74d76"; + +/// Fixed nonzero UID/GID for the agent container (§Pod shape hardening). +pub const RUN_AS_UID: i64 = 10001; +pub const RUN_AS_GID: i64 = 10001; + +/// Writable workspace root; also `HOME` and the harness's cwd +/// (§Working directory). +pub const WORKSPACE_PATH: &str = "/home/agent"; + +/// `terminationGracePeriodSeconds` — a declared budget, not a derived sum +/// (§Pod shape). Kubernetes' default 30s would SIGKILL the harness mid-drain. +pub const TERMINATION_GRACE_SECONDS: i64 = 60; + +/// The only restart policy v1 ships. `OnFailure` is double-gated on the +/// harness exit-code contract *and* a crash-loop classification row the state +/// machine does not have (`:1121-1139`); until both land the provider refuses +/// the combination rather than shipping against an undefended convention. +pub const RESTART_POLICY: &str = "Never"; + +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct ProviderConfig { + /// kubeconfig context; `None` uses the current context. + pub context: Option, + pub namespace: String, + pub image: ImageRef, + pub resources: Resources, + /// `None` when `inactivity_seconds` was 0 — refused in v1, see [`parse`]. + pub inactivity_seconds: Option, + pub service_account: Option, +} + +/// Read an optional non-empty string field. Rejects non-string scalars rather +/// than stringifying them, so a mistyped field is named at the boundary. +fn optional_string(cfg: &serde_json::Value, field: &str) -> Result, String> { + match cfg.get(field) { + None | Some(serde_json::Value::Null) => Ok(None), + Some(serde_json::Value::String(s)) if s.trim().is_empty() => Ok(None), + Some(serde_json::Value::String(s)) => Ok(Some(s.trim().to_string())), + Some(other) => Err(format!( + "provider_config.{field} must be a string, got {other}" + )), + } +} + +/// Read an optional unsigned integer. The desktop's form omits blank numeric +/// fields rather than sending `""`, but a hand-crafted payload may send a +/// numeric string — accept both, refuse anything else. +fn optional_u64(cfg: &serde_json::Value, field: &str) -> Result, String> { + match cfg.get(field) { + None | Some(serde_json::Value::Null) => Ok(None), + Some(serde_json::Value::Number(n)) => n.as_u64().map(Some).ok_or_else(|| { + format!("provider_config.{field} must be a non-negative integer, got {n}") + }), + Some(serde_json::Value::String(s)) if s.trim().is_empty() => Ok(None), + Some(serde_json::Value::String(s)) => s.trim().parse::().map(Some).map_err(|_| { + format!("provider_config.{field} must be a non-negative integer, got {s:?}") + }), + Some(other) => Err(format!( + "provider_config.{field} must be a non-negative integer, got {other}" + )), + } +} + +/// A Kubernetes namespace name: RFC 1123 label, ≤63 chars. Validated here so a +/// typo fails with a named field instead of an apiserver rejection partway +/// through a deploy. +fn valid_namespace(name: &str) -> bool { + !name.is_empty() + && name.len() <= 63 + && name.starts_with(|c: char| c.is_ascii_lowercase() || c.is_ascii_digit()) + && name.ends_with(|c: char| c.is_ascii_lowercase() || c.is_ascii_digit()) + && name + .chars() + .all(|c| c.is_ascii_lowercase() || c.is_ascii_digit() || c == '-') +} + +pub fn parse(cfg: &serde_json::Value) -> Result { + if !cfg.is_object() && !cfg.is_null() { + return Err("provider_config must be a JSON object".to_string()); + } + + let namespace = optional_string(cfg, "namespace")?.ok_or_else(|| { + "provider_config.namespace is required: the info schema supplies a \ + generated default, so an empty value means the form was cleared" + .to_string() + })?; + if !valid_namespace(&namespace) { + return Err(format!( + "provider_config.namespace {namespace:?} is not a valid Kubernetes \ + namespace (lowercase alphanumerics and '-', ≤63 characters)" + )); + } + + let image = image::parse(optional_string(cfg, "image")?.unwrap_or_default().as_str())?; + + let defaults = Resources::default(); + let resources = Resources { + cpu_request: optional_string(cfg, "cpu_request")?.unwrap_or(defaults.cpu_request), + memory_request: optional_string(cfg, "memory_request")?.unwrap_or(defaults.memory_request), + cpu_limit: optional_string(cfg, "cpu_limit")?.unwrap_or(defaults.cpu_limit), + memory_limit: optional_string(cfg, "memory_limit")?.unwrap_or(defaults.memory_limit), + }; + + // `inactivity_seconds: 0` is a legal, blessed value in the spec (§Auto-Stop) + // meaning "no auto-stop" — but it selects `restartPolicy: OnFailure`, which + // §Pod shape forbids until the harness exit-code contract is pinned AND the + // state machine gains a crash-loop row. Refusing the *combination* is what + // the spec asks for; silently downgrading to `Never` would ship an + // indefinite agent that dies on its first crash. + let inactivity_seconds = match optional_u64(cfg, "inactivity_seconds")? { + None => Some(DEFAULT_INACTIVITY_SECONDS), + Some(0) => { + return Err( + "provider_config.inactivity_seconds: 0 (indefinite lifetime) is not \ + supported in this version: it requires restartPolicy OnFailure, \ + which is gated on the harness exit-code contract. Set a positive \ + number of seconds." + .to_string(), + ) + } + Some(n) => Some(n), + }; + + Ok(ProviderConfig { + context: optional_string(cfg, "context")?, + namespace, + image, + resources, + inactivity_seconds, + service_account: optional_string(cfg, "service_account")?, + }) +} + +/// A fresh `buzz-agents-` namespace default. +/// +/// Computed per `info` call, which is how "random default" is satisfied with +/// zero UI changes: the schema's `default` prefills the form (§K8s Namespace). +pub fn generated_namespace() -> String { + use rand::RngExt; + const ALPHABET: &[u8] = b"abcdefghijklmnopqrstuvwxyz0123456789"; + let mut rng = rand::rng(); + let suffix: String = (0..6) + .map(|_| ALPHABET[rng.random_range(0..ALPHABET.len())] as char) + .collect(); + format!("buzz-agents-{suffix}") +} + +/// The `config_schema` returned by `info`. Drives the UI form: +/// `properties[*].default` prefill, scalar coercion, `required` gating +/// (`:407-411`). +pub fn config_schema() -> serde_json::Value { + let defaults = Resources::default(); + serde_json::json!({ + "type": "object", + "properties": { + "context": { + "type": "string", + "title": "Kubeconfig context", + "description": "Context from your kubeconfig. Leave empty to use the current context." + }, + "namespace": { + "type": "string", + "title": "Namespace", + "description": "Created if it does not exist.", + "default": generated_namespace() + }, + "image": { + "type": "string", + "title": "Agent image", + "description": "Digest-pinned image containing the buzz-acp runtime ABI, e.g. ghcr.io/block/buzz-sprig@sha256:. Tags alone are not accepted: this pod holds the agent's private key.", + "default": DEFAULT_IMAGE + }, + "cpu_request": { + "type": "string", "title": "CPU request", "default": defaults.cpu_request + }, + "memory_request": { + "type": "string", "title": "Memory request", "default": defaults.memory_request + }, + "cpu_limit": { + "type": "string", "title": "CPU limit", "default": defaults.cpu_limit + }, + "memory_limit": { + "type": "string", "title": "Memory limit", "default": defaults.memory_limit + }, + "inactivity_seconds": { + "type": "number", + "title": "Stop after inactivity (seconds)", + "description": "The agent exits after this long with no work, and can be started again at any time.", + "default": DEFAULT_INACTIVITY_SECONDS + }, + "service_account": { + "type": "string", + "title": "Service account", + "description": "Scheduling/RBAC identity only. No API token is mounted." + } + }, + "required": ["namespace", "image"] + }) +} + +#[cfg(test)] +mod tests { + use super::*; + + fn digest_ref() -> String { + format!("ghcr.io/block/buzz-sprig@sha256:{}", "a".repeat(64)) + } + + fn minimal() -> serde_json::Value { + serde_json::json!({"namespace": "buzz-agents-abc123", "image": digest_ref()}) + } + + #[test] + fn applies_spec_defaults() { + let c = parse(&minimal()).unwrap(); + assert_eq!(c.resources, Resources::default()); + assert_eq!(c.resources.cpu_request, "1"); + assert_eq!(c.resources.memory_request, "2Gi"); + assert_eq!(c.resources.cpu_limit, "2"); + assert_eq!(c.resources.memory_limit, "4Gi"); + assert_eq!(c.inactivity_seconds, Some(DEFAULT_INACTIVITY_SECONDS)); + assert_eq!(c.context, None); + assert_eq!(c.service_account, None); + } + + #[test] + fn all_four_resources_are_configurable() { + let mut cfg = minimal(); + cfg["cpu_request"] = "500m".into(); + cfg["memory_request"] = "1Gi".into(); + cfg["cpu_limit"] = "4".into(); + cfg["memory_limit"] = "8Gi".into(); + let c = parse(&cfg).unwrap(); + assert_eq!( + c.resources, + Resources { + cpu_request: "500m".into(), + memory_request: "1Gi".into(), + cpu_limit: "4".into(), + memory_limit: "8Gi".into(), + } + ); + } + + /// The desktop's form omits blank numeric fields; a hand-crafted payload + /// may send a numeric string. Both must mean the same thing. + #[test] + fn inactivity_accepts_number_string_and_omission() { + let mut cfg = minimal(); + cfg["inactivity_seconds"] = serde_json::json!(300); + assert_eq!(parse(&cfg).unwrap().inactivity_seconds, Some(300)); + + cfg["inactivity_seconds"] = serde_json::json!("300"); + assert_eq!(parse(&cfg).unwrap().inactivity_seconds, Some(300)); + + cfg["inactivity_seconds"] = serde_json::json!(""); + assert_eq!( + parse(&cfg).unwrap().inactivity_seconds, + Some(DEFAULT_INACTIVITY_SECONDS) + ); + } + + /// Indefinite lifetime selects `OnFailure`, which is gated. Refuse rather + /// than silently downgrade — a downgraded agent dies on its first crash + /// while the user believes they asked for indefinite. + #[test] + fn refuses_indefinite_lifetime() { + let mut cfg = minimal(); + cfg["inactivity_seconds"] = serde_json::json!(0); + let err = parse(&cfg).unwrap_err(); + assert!(err.contains("inactivity_seconds"), "got: {err}"); + assert!( + err.contains("OnFailure"), + "error should name the gate: {err}" + ); + } + + #[test] + fn rejects_negative_and_non_numeric_inactivity() { + for bad in [ + serde_json::json!(-1), + serde_json::json!(1.5), + serde_json::json!("soon"), + serde_json::json!(true), + ] { + let mut cfg = minimal(); + cfg["inactivity_seconds"] = bad.clone(); + assert!(parse(&cfg).is_err(), "accepted {bad}"); + } + } + + #[test] + fn image_is_required_and_must_be_digest_pinned() { + let mut cfg = minimal(); + cfg.as_object_mut().unwrap().remove("image"); + assert!(parse(&cfg).unwrap_err().contains("provider_config.image")); + + cfg["image"] = "ghcr.io/block/buzz-sprig:latest".into(); + assert!(parse(&cfg).unwrap_err().contains("digest-pinned")); + } + + #[test] + fn rejects_invalid_namespace_names() { + for bad in [ + "", + "Buzz-Agents", + "-leading", + "trailing-", + "has_underscore", + &"n".repeat(64), + ] { + let mut cfg = minimal(); + cfg["namespace"] = bad.into(); + assert!(parse(&cfg).is_err(), "accepted namespace {bad:?}"); + } + } + + /// I2 corollary: there is no config path for cluster credentials, so a + /// caller that tries to supply one gets no effect from it. Asserting the + /// parsed struct has no such field is the closest a test can get to + /// "the type makes it impossible". + #[test] + fn credential_fields_have_no_effect() { + let mut cfg = minimal(); + cfg["token"] = "hunter2".into(); + cfg["client_key"] = "hunter2".into(); + let c = parse(&cfg).unwrap(); + let rendered = format!("{c:?}"); + assert!( + !rendered.contains("hunter2"), + "config absorbed a credential: {rendered}" + ); + } + + #[test] + fn mistyped_string_fields_are_named() { + let mut cfg = minimal(); + cfg["namespace"] = serde_json::json!(42); + assert!(parse(&cfg) + .unwrap_err() + .contains("provider_config.namespace")); + } + + #[test] + fn generated_namespaces_are_fresh_and_valid() { + let a = generated_namespace(); + let b = generated_namespace(); + assert_ne!(a, b, "namespace default is not random"); + assert!(valid_namespace(&a), "{a} is not a valid namespace"); + assert!(a.starts_with("buzz-agents-")); + assert_eq!(a.len(), "buzz-agents-".len() + 6); + } + + /// The schema's own namespace default must be a value the parser accepts — + /// otherwise the UI prefills a form that fails on submit. + #[test] + fn schema_default_namespace_round_trips_through_parse() { + let schema = config_schema(); + let default = schema["properties"]["namespace"]["default"] + .as_str() + .unwrap(); + let cfg = serde_json::json!({"namespace": default, "image": digest_ref()}); + assert_eq!(parse(&cfg).unwrap().namespace, default); + } + + /// Same guarantee for the image prefill: the schema's default must be a + /// value `image::parse` accepts, or the UI prefills a form that fails on + /// submit. Its tag+digest form normalizes to the tagless canonical form. + #[test] + fn schema_default_image_round_trips_through_parse() { + let schema = config_schema(); + let default = schema["properties"]["image"]["default"].as_str().unwrap(); + assert_eq!(default, DEFAULT_IMAGE); + let cfg = serde_json::json!({"namespace": "buzz-agents-abc123", "image": default}); + let parsed = parse(&cfg).unwrap(); + assert_eq!( + parsed.image.as_str(), + "ghcr.io/block/buzz-sprig@sha256:17facfc7608d8ddb33bc056c9aaba1098f4ef6abe5655702fbfd7584d1f74d76" + ); + } + + /// Nine fields exactly (§`provider_config` v1 fields). The cap is 20; the + /// count is pinned so a field added without a spec change is caught here. + #[test] + fn schema_declares_exactly_the_nine_v1_fields() { + let schema = config_schema(); + let props = schema["properties"].as_object().unwrap(); + let mut keys: Vec<&str> = props.keys().map(String::as_str).collect(); + keys.sort(); + assert_eq!( + keys, + [ + "context", + "cpu_limit", + "cpu_request", + "image", + "inactivity_seconds", + "memory_limit", + "memory_request", + "namespace", + "service_account" + ] + ); + assert_eq!( + schema["required"], + serde_json::json!(["namespace", "image"]) + ); + } + + /// I2's key lint rejects any field whose word-split contains + /// secret|password|token|key|credential. A schema field tripping it would + /// make every deploy fail validation desktop-side (`:185-198`). + #[test] + fn no_schema_field_trips_the_i2_key_lint() { + const BANNED: [&str; 5] = ["secret", "password", "token", "key", "credential"]; + let schema = config_schema(); + for field in schema["properties"].as_object().unwrap().keys() { + for word in field.split(['_', '-']) { + assert!( + !BANNED.contains(&word), + "field {field:?} contains I2-banned word {word:?}" + ); + } + } + } +} diff --git a/crates/buzz-backend-kubernetes/src/env.rs b/crates/buzz-backend-kubernetes/src/env.rs new file mode 100644 index 0000000000..badff621e8 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/env.rs @@ -0,0 +1,732 @@ +//! Building the pod environment (spec §Launch data, §Entrypoint mapping table). +//! +//! The three tiers are resolved *here*, before serialization, because a +//! Kubernetes Secret's `data` is a flat map with no precedence of its own: if +//! two tiers supplied the same key, whichever entry landed in the map would +//! win silently. Resolving in-provider makes later-wins explicit and testable. + +use crate::wire::{AgentPayload, LaunchBlock}; +use std::collections::BTreeMap; + +/// Keys the authoritative tier owns. +/// +/// Load-bearing, not documentation: tier 3 *clears* every key on this list +/// before writing its own values, so a key the authoritative tier has no value +/// for is **removed** rather than left holding a lower-tier value. Plain +/// overwrite is not enough — most of these are written conditionally +/// (`BUZZ_ACP_AGENT_ARGS` only when `launch.args` is non-empty, +/// `BUZZ_ACP_RESPOND_TO` only when set), and without the clear, a lower tier +/// could supply the value for exactly the cases the authoritative tier stays +/// silent on. Clearing is also what the local spawn does: the desktop strips +/// reserved keys from user env before the authoritative layer is written +/// (`env_vars.rs:54-57`), so absent-means-absent in both paths. +const AUTHORITATIVE_KEYS: &[&str] = &[ + "BUZZ_RELAY_URL", + "BUZZ_PRIVATE_KEY", + "NOSTR_PRIVATE_KEY", + "BUZZ_AUTH_TAG", + "BUZZ_ACP_AGENT_OWNER", + "BUZZ_ACP_AGENT_COMMAND", + "BUZZ_ACP_AGENT_ARGS", + "BUZZ_ACP_RESPOND_TO", + "BUZZ_ACP_RESPOND_TO_ALLOWLIST", + "BUZZ_ACP_MCP_COMMAND", + "BUZZ_ACP_EXIT_AFTER_INACTIVITY", + START_NONCE_KEY, +]; + +/// The attempt's generation, as the harness sees it. Also the Secret's name +/// suffix — one generation, one identity — so the reconciler restamps this on +/// every create attempt rather than letting the caller's value persist across +/// a retry. +pub const START_NONCE_KEY: &str = "BUZZ_MANAGED_AGENT_START_NONCE"; + +/// Presence is the only remote liveness signal (I3), so a launch that +/// suppresses it is non-conforming (L1 item 2) — and unlike a reserved-key +/// collision, there is no "authoritative value" to overwrite it with. Refuse. +const FORBIDDEN_KEY: &str = "BUZZ_ACP_NO_PRESENCE"; + +/// Kubernetes' own cap on the summed value bytes of a Secret +/// (`MaxSecretSize`, `pkg/apis/core/types.go`). Enforced here so an oversized +/// env surfaces as a named provider error rather than an apiserver rejection +/// partway through a deploy. +const MAX_SECRET_BYTES: usize = 1024 * 1024; + +/// A POSIX-shaped env var name: `[A-Za-z_][A-Za-z0-9_]*`. +/// +/// Kubernetes validates Secret *keys* as `IsConfigMapKey` +/// (`[-._a-zA-Z0-9]+`), which is looser — `foo.bar` is a legal Secret key. +/// What the kubelet then does with such a key **changed between versions**: +/// through 1.29 it filtered invalid env names out of `envFrom` and emitted an +/// `InvalidEnvironmentVariableNames` warning event +/// (`pkg/kubelet/kubelet_pods.go:646,654` at v1.29.0); from 1.30 that filter +/// is gone (KEP-4369) and the key is injected verbatim. The same manifest +/// would silently drop a variable on one cluster and set it on another, so we +/// fail closed on the provider side and get one deterministic behavior. +fn is_posix_env_key(key: &str) -> bool { + let mut chars = key.chars(); + match chars.next() { + Some(c) if c == '_' || c.is_ascii_alphabetic() => {} + _ => return false, + } + chars.all(|c| c == '_' || c.is_ascii_alphanumeric()) +} + +/// An identity component (L1 item 1) is present only if it is nonempty after +/// trimming — and the **trimmed form is what gets stored**. The validator and +/// the writer must never disagree about the value: a guard that accepts +/// `" wss://relay "` and then writes it with the padding intact has only +/// moved the failure from a loud refusal to a connect error in the harness. +fn identity_component(value: &str) -> Option<&str> { + let trimmed = value.trim(); + (!trimmed.is_empty()).then_some(trimmed) +} + +/// The harness's `allowlist` gate mode, spelled as the desktop serializes +/// `RespondTo` (kebab-case) and as `buzz-acp`'s CLI parses it. +const RESPOND_TO_ALLOWLIST: &str = "allowlist"; + +/// Every gate mode `buzz-acp` accepts, spelled as its `clap::ValueEnum` parses +/// them (`config.rs:95-101`, kebab-case via `RespondTo`'s `Display`). +/// +/// Deliberately the **harness's** four and not the desktop's three: the desktop +/// rejects `nobody` on purpose (`managed_agents/types.rs:871-880`), but the +/// harness starts fine with it. This guard exists to cover non-desktop callers, +/// so inheriting a desktop-only narrowing would refuse a launch that works. +const RESPOND_TO_MODES: [&str; 4] = ["owner-only", RESPOND_TO_ALLOWLIST, "anyone", "nobody"]; + +/// Refuse a respond-to gate the harness will reject at config parse. +/// +/// The local spawn path re-validates this before spawning — "doing it here +/// means we never spawn a doomed process" (`runtime.rs:378`) — but the deploy +/// path projects the record's fields straight through. Without this, a gate +/// the harness refuses becomes a pod that exits 1 at startup; `restartPolicy: +/// Never` turns that into `Terminated` → `Delete` → recreate, and each cycle +/// leaves a Secret the in-call path never reaps (only a later deploy's orphan +/// sweep does, at `ORPHAN_SECRET_MIN_AGE_SECS`). The user-visible ending is +/// "startup not confirmed", indistinguishable from a slow cluster. +/// +/// Mirrors `buzz-acp`'s own rules exactly (`config.rs:95-101,996-1004,629-641`), +/// deliberately including their asymmetry: the allowlist is validated **only** +/// in allowlist mode, and merely warned about otherwise. Validating it in +/// every mode would refuse a deploy whose identical local spawn succeeds — +/// and a stale list is already harmless here, since +/// `BUZZ_ACP_RESPOND_TO_ALLOWLIST` is an authoritative key that tier 3 clears. +fn validate_respond_to_gate(respond_to: &str, allowlist: Option<&[String]>) -> Result<(), String> { + // Exact, untrimmed: `clap` does not trim, so `" allowlist "` is `rc=2` at + // the harness — a parse failure even earlier than the config errors below. + if !RESPOND_TO_MODES.contains(&respond_to) { + return Err(format!( + "deploy refused: respond_to {respond_to:?} is not a mode the \ + harness accepts (expected one of {}) — the pod would fail to \ + parse its arguments, be replaced, and leave a Secret behind on \ + every attempt", + RESPOND_TO_MODES.join(", ") + )); + } + if respond_to != RESPOND_TO_ALLOWLIST { + return Ok(()); + } + let entries = allowlist.unwrap_or_default(); + if entries.is_empty() { + return Err(format!( + "deploy refused: respond_to is {RESPOND_TO_ALLOWLIST:?} but the \ + allowlist is empty — the harness refuses this at startup, so the \ + pod would fail, be replaced, and leave a Secret behind on every \ + attempt" + )); + } + for entry in entries { + let trimmed = entry.trim(); + if trimmed.len() != 64 || !trimmed.chars().all(|c| c.is_ascii_hexdigit()) { + return Err(format!( + "deploy refused: invalid pubkey in respond_to_allowlist: \ + {entry:?} (must be exactly 64 hex characters)" + )); + } + } + Ok(()) +} + +/// Inputs the provider itself supplies to the authoritative tier. +pub struct AuthoritativeInputs<'a> { + /// The attempt's generation token — also the Secret's name suffix, so the + /// lifecycle correlator and the Secret generation are one identity. + pub generation: &'a str, + /// Resolved from `provider_config.inactivity_seconds`; `None` when the + /// indefinite opt-in was chosen (which this version refuses elsewhere). + pub inactivity_seconds: Option, +} + +/// Resolve the full pod environment. +/// +/// Order is the spec's, and the function body is deliberately three writes in +/// that order — tier 1, tier 2, tier 3 — so "later wins" is visible rather +/// than argued. +pub fn build_env( + agent: &AgentPayload, + auth: AuthoritativeInputs<'_>, +) -> Result, String> { + let default_launch = LaunchBlock::default(); + let launch = agent.launch.as_ref().unwrap_or(&default_launch); + + let mut env: BTreeMap = BTreeMap::new(); + + // Tier 1 — overridable behavior defaults. + env.extend(launch.policy_env.clone()); + + // Tier 2 — user/layered env. The descriptor already merged + // global < persona < agent, so `agent.env_vars` is NOT re-merged on top + // (§Launch data tier 2) — doing so would resurrect a layer the desktop + // already resolved. When the desktop predates the `launch` block we fall + // back to the legacy field, which is the only case it is the truth. + if agent.launch.is_some() { + env.extend(launch.env.clone()); + } else { + env.extend(agent.env_vars.clone()); + } + + // Validate what the lower tiers contributed, before the authoritative + // tier overwrites any of it. A reserved-key collision is NOT fatal: the + // spec's precedence is later-wins, so tier 3 simply overwrites it, which + // is exactly what a local spawn does. Only a key that has no + // authoritative counterpart to overwrite it — presence suppression — is + // a refusal. + for key in env.keys() { + if !is_posix_env_key(key) { + return Err(format!( + "env key {key:?} is not a POSIX environment variable name \ + ([A-Za-z_][A-Za-z0-9_]*); Kubernetes would treat it \ + inconsistently across cluster versions" + )); + } + if key.eq_ignore_ascii_case(FORBIDDEN_KEY) { + return Err(format!( + "{FORBIDDEN_KEY} must not be set on a remote agent: presence \ + is the only signal that a remote agent is alive" + )); + } + } + + // Tier 3 — authoritative. Every key it owns is cleared first, then the + // values it has are written, so it wins at a key whether or not it has a + // value there (see [`AUTHORITATIVE_KEYS`]). + for key in AUTHORITATIVE_KEYS { + env.remove(*key); + } + // Identity comes from top-level payload fields, never from `env_vars` + // (§Reserved-key rule). All three components must be nonempty: an agent + // that cannot reach a relay is the identityless launch L1 item 1 exists to + // prevent, and a blank field would otherwise sail through into the Secret + // and produce a pod that starts, fails to connect, and looks like a + // network problem. + let Some(relay_url) = identity_component(&agent.relay_url) else { + return Err("deploy refused: relay_url is empty — the agent would have \ + no relay to connect to" + .to_string()); + }; + env.insert("BUZZ_RELAY_URL".into(), relay_url.to_string()); + env.insert("BUZZ_PRIVATE_KEY".into(), agent.private_key_nsec.clone()); + // The git credential/signing helpers read NOSTR_PRIVATE_KEY. + env.insert("NOSTR_PRIVATE_KEY".into(), agent.private_key_nsec.clone()); + + // Owner: at least one of these must resolve, or the harness cannot match + // `!shutdown` and §Stop describes a mechanism that does not work. + let auth_tag = agent.auth_tag.as_deref().and_then(identity_component); + let owner = launch.owner_pubkey.as_deref().and_then(identity_component); + match (auth_tag, owner) { + (None, None) => { + return Err("deploy refused: neither auth_tag nor launch.owner_pubkey \ + resolved — without an owner the agent cannot honor \ + !shutdown" + .to_string()) + } + (tag, own) => { + if let Some(t) = tag { + env.insert("BUZZ_AUTH_TAG".into(), t.to_string()); + } + if let Some(o) = own { + env.insert("BUZZ_ACP_AGENT_OWNER".into(), o.to_string()); + } + } + } + + // The harness and MCP binaries are resolved against the *image's* PATH. + // A host path forwarded from the desktop is guaranteed absent in the + // container (§Launch data, host-resolved values). + if let Some(command) = launch.command.as_deref().filter(|c| !c.is_empty()) { + env.insert("BUZZ_ACP_AGENT_COMMAND".into(), command.to_string()); + } + if !launch.args.is_empty() { + // Comma-joined because that is what the harness's CLI parser decodes, + // and what the desktop's local spawn does. An argument containing a + // comma is unrepresentable in both paths; inventing an escaping + // scheme here would produce args the harness cannot decode. + env.insert("BUZZ_ACP_AGENT_ARGS".into(), launch.args.join(",")); + } + env.insert("BUZZ_ACP_MCP_COMMAND".into(), "buzz-dev-mcp".into()); + + if let Some(respond_to) = agent.respond_to.as_deref().filter(|s| !s.is_empty()) { + validate_respond_to_gate(respond_to, agent.respond_to_allowlist.as_deref())?; + env.insert("BUZZ_ACP_RESPOND_TO".into(), respond_to.to_string()); + } + if let Some(list) = agent + .respond_to_allowlist + .as_ref() + .filter(|l| !l.is_empty()) + { + env.insert("BUZZ_ACP_RESPOND_TO_ALLOWLIST".into(), list.join(",")); + } + + if let Some(secs) = auth.inactivity_seconds { + env.insert("BUZZ_ACP_EXIT_AFTER_INACTIVITY".into(), secs.to_string()); + } + // The generation token doubles as the lifecycle-frame correlator, so pod + // logs and observer frames share one identity (§K8s Secrets). + env.insert(START_NONCE_KEY.into(), auth.generation.to_string()); + + let total: usize = env.values().map(String::len).sum(); + if total > MAX_SECRET_BYTES { + return Err(format!( + "agent environment is {total} bytes; Kubernetes caps Secret data \ + at {MAX_SECRET_BYTES}" + )); + } + + Ok(env) +} + +#[cfg(test)] +mod tests { + use super::*; + + fn payload_json(extra_agent: serde_json::Value) -> AgentPayload { + let mut agent = serde_json::json!({ + "name": "a", + "relay_url": "wss://relay.example", + "private_key_nsec": "nsec1example", + "auth_tag": "tag-1", + }); + let (serde_json::Value::Object(base), serde_json::Value::Object(extra)) = + (&mut agent, extra_agent) + else { + panic!("expected objects") + }; + base.extend(extra); + serde_json::from_value(agent).unwrap() + } + + fn build(agent: &AgentPayload) -> Result, String> { + build_env( + agent, + AuthoritativeInputs { + generation: "gen0001", + inactivity_seconds: Some(7200), + }, + ) + } + + #[test] + fn identity_comes_from_top_level_fields() { + let env = build(&payload_json(serde_json::json!({}))).unwrap(); + assert_eq!(env["BUZZ_RELAY_URL"], "wss://relay.example"); + assert_eq!(env["BUZZ_PRIVATE_KEY"], "nsec1example"); + assert_eq!(env["NOSTR_PRIVATE_KEY"], "nsec1example"); + assert_eq!(env["BUZZ_AUTH_TAG"], "tag-1"); + } + + /// Wren's amendment, and the spec's later-wins rule: a lower tier that + /// spoofs an authoritative key is *overwritten*, not refused. Refusing + /// would diverge from the local spawn, where the same env is written + /// before the authoritative layer and simply loses. + #[test] + fn lower_tiers_cannot_spoof_authoritative_values() { + let agent = payload_json(serde_json::json!({ + "launch": { + "command": "goose", + "policy_env": { + "BUZZ_PRIVATE_KEY": "nsec1attacker", + "BUZZ_MANAGED_AGENT_START_NONCE": "forged", + }, + "env": { + "BUZZ_RELAY_URL": "wss://attacker.example", + "NOSTR_PRIVATE_KEY": "nsec1attacker", + "BUZZ_AUTH_TAG": "forged-tag", + "BUZZ_ACP_AGENT_OWNER": "cafe", + "BUZZ_ACP_AGENT_COMMAND": "/bin/sh", + "BUZZ_ACP_MCP_COMMAND": "/bin/sh", + "BUZZ_ACP_EXIT_AFTER_INACTIVITY": "0", + }, + "owner_pubkey": "beef" + } + })); + let env = build(&agent).unwrap(); + assert_eq!(env["BUZZ_PRIVATE_KEY"], "nsec1example"); + assert_eq!(env["NOSTR_PRIVATE_KEY"], "nsec1example"); + assert_eq!(env["BUZZ_RELAY_URL"], "wss://relay.example"); + assert_eq!(env["BUZZ_AUTH_TAG"], "tag-1"); + assert_eq!(env["BUZZ_ACP_AGENT_OWNER"], "beef"); + assert_eq!(env["BUZZ_ACP_AGENT_COMMAND"], "goose"); + assert_eq!(env["BUZZ_ACP_MCP_COMMAND"], "buzz-dev-mcp"); + assert_eq!(env["BUZZ_ACP_EXIT_AFTER_INACTIVITY"], "7200"); + assert_eq!(env["BUZZ_MANAGED_AGENT_START_NONCE"], "gen0001"); + } + + /// Tier 1 is *overridable* — user env beats policy defaults, matching the + /// local spawn, where the user layer is written after them. Getting this + /// backwards would make remote agents ignore overrides local agents honor. + #[test] + fn user_env_overrides_policy_defaults() { + let agent = payload_json(serde_json::json!({ + "launch": { + "policy_env": {"GOOSE_MODE": "auto", "BUZZ_ACP_MODEL": "sonnet"}, + "env": {"GOOSE_MODE": "chat"}, + "owner_pubkey": "beef" + } + })); + let env = build(&agent).unwrap(); + assert_eq!(env["GOOSE_MODE"], "chat"); + assert_eq!(env["BUZZ_ACP_MODEL"], "sonnet"); + } + + /// `launch.env` already contains the merged user env, so re-merging the + /// legacy field would undo a layering the desktop already resolved. + #[test] + fn legacy_env_vars_are_not_remerged_when_launch_present() { + let agent = payload_json(serde_json::json!({ + "env_vars": {"STALE": "yes", "SHARED": "legacy"}, + "launch": {"env": {"SHARED": "resolved"}, "owner_pubkey": "beef"} + })); + let env = build(&agent).unwrap(); + assert_eq!(env["SHARED"], "resolved"); + assert!(!env.contains_key("STALE"), "legacy env_vars re-merged"); + } + + /// ...but a desktop predating the `launch` block has nothing else to + /// offer, so the legacy field is the truth in exactly that case. + #[test] + fn legacy_env_vars_used_when_launch_absent() { + let agent = payload_json(serde_json::json!({"env_vars": {"API": "v"}})); + let env = build(&agent).unwrap(); + assert_eq!(env["API"], "v"); + } + + #[test] + fn refuses_when_no_owner_resolves() { + let agent = payload_json(serde_json::json!({"auth_tag": null})); + let err = build(&agent).unwrap_err(); + assert!(err.contains("!shutdown"), "unhelpful error: {err}"); + } + + /// An empty string is not an owner. Without this the refusal is + /// bypassable by a blank field and the pod launches unable to be stopped. + #[test] + fn empty_owner_fields_count_as_absent() { + for blank in ["", " "] { + let agent = payload_json(serde_json::json!({ + "auth_tag": blank, + "launch": {"owner_pubkey": blank} + })); + assert!( + build(&agent).is_err(), + "whitespace resolved as an owner: {blank:?}" + ); + } + } + + /// The other half of every identity guard: what is *stored*. A validator + /// that trims and a writer that doesn't disagree about the value, and the + /// padding reaches the harness inside the Secret. Assert on the stored + /// string — asserting only that the deploy was accepted passes either way. + #[test] + fn identity_components_are_stored_trimmed() { + let agent = payload_json(serde_json::json!({ + "relay_url": " wss://relay.example ", + "auth_tag": " tag-1 ", + "launch": {"owner_pubkey": " beefcafe "} + })); + let env = build(&agent).unwrap(); + assert_eq!(env["BUZZ_RELAY_URL"], "wss://relay.example"); + assert_eq!(env["BUZZ_AUTH_TAG"], "tag-1"); + assert_eq!(env["BUZZ_ACP_AGENT_OWNER"], "beefcafe"); + } + + /// L1 item 1's third identity component. The nsec arm is enforced in + /// `naming.rs` and the owner arm above; without this one an agent + /// deploys with nothing to connect to — a pod that starts, fails at the + /// relay, and reads as a network fault rather than a refused launch. + #[test] + fn refuses_empty_relay_url() { + for blank in ["", " "] { + let agent = payload_json(serde_json::json!({"relay_url": blank})); + let err = build(&agent).unwrap_err(); + assert!(err.contains("relay_url"), "unhelpful error: {err}"); + } + } + + #[test] + fn owner_pubkey_alone_is_sufficient() { + let agent = payload_json(serde_json::json!({ + "auth_tag": null, + "launch": {"owner_pubkey": "beefcafe"} + })); + let env = build(&agent).unwrap(); + assert_eq!(env["BUZZ_ACP_AGENT_OWNER"], "beefcafe"); + assert!(!env.contains_key("BUZZ_AUTH_TAG")); + } + + #[test] + fn refuses_presence_suppression() { + let agent = payload_json(serde_json::json!({ + "launch": {"env": {"BUZZ_ACP_NO_PRESENCE": "1"}, "owner_pubkey": "beef"} + })); + let err = build(&agent).unwrap_err(); + assert!(err.contains("BUZZ_ACP_NO_PRESENCE"), "got: {err}"); + } + + /// `foo.bar` is a legal Secret key but not a legal env name: pre-1.30 + /// kubelets drop it, 1.30+ inject it. Refuse rather than behave + /// differently depending on the cluster. + #[test] + fn refuses_non_posix_env_keys() { + for bad in ["foo.bar", "foo-bar", "1LEADING", "", "has space"] { + let agent = payload_json(serde_json::json!({ + "launch": {"env": {bad: "v"}, "owner_pubkey": "beef"} + })); + assert!(build(&agent).is_err(), "accepted non-POSIX key {bad:?}"); + } + } + + #[test] + fn args_are_comma_joined_and_omitted_when_empty() { + let agent = payload_json(serde_json::json!({ + "launch": {"command": "goose", "args": ["run", "--no-session"], "owner_pubkey": "b"} + })); + let env = build(&agent).unwrap(); + assert_eq!(env["BUZZ_ACP_AGENT_ARGS"], "run,--no-session"); + + let agent = payload_json(serde_json::json!({ + "launch": {"command": "goose", "args": [], "owner_pubkey": "b"} + })); + assert!(!build(&agent).unwrap().contains_key("BUZZ_ACP_AGENT_ARGS")); + } + + /// The top-level `model`/`provider` fields are display inputs; their + /// environment consequence is per-runtime and arrives already resolved + /// inside `launch`. A provider-side mapping is wrong for three of the + /// four built-in runtimes. + #[test] + fn provider_never_maps_model_or_provider_itself() { + let agent = payload_json(serde_json::json!({ + "model": "claude-opus", "provider": "anthropic", + "launch": {"owner_pubkey": "beef"} + })); + let env = build(&agent).unwrap(); + for key in [ + "BUZZ_AGENT_PROVIDER", + "BUZZ_AGENT_MODEL", + "GOOSE_PROVIDER", + "GOOSE_MODEL", + ] { + assert!(!env.contains_key(key), "provider mapped {key} itself"); + } + } + + /// `turn_timeout_seconds` is deprecated and ignored upstream; the local + /// spawn does not emit it either. + #[test] + fn turn_timeout_is_not_mapped() { + let agent = payload_json(serde_json::json!({ + "turn_timeout_seconds": 30, "launch": {"owner_pubkey": "b"} + })); + let env = build(&agent).unwrap(); + assert!(!env.keys().any(|k| k.contains("TURN_TIMEOUT"))); + } + + #[test] + fn inactivity_omitted_when_unset() { + let agent = payload_json(serde_json::json!({"launch": {"owner_pubkey": "b"}})); + let env = build_env( + &agent, + AuthoritativeInputs { + generation: "g", + inactivity_seconds: None, + }, + ) + .unwrap(); + assert!(!env.contains_key("BUZZ_ACP_EXIT_AFTER_INACTIVITY")); + } + + /// Structural guard over the whole authoritative list at once: whatever + /// the lower tiers contain, no authoritative key holds a lower-tier value + /// — including the conditionally-written ones the authoritative tier has + /// nothing to say about, which must be **absent** rather than spoofed. + /// This test caught exactly that: `BUZZ_ACP_AGENT_ARGS` is only written + /// when `launch.args` is non-empty, so plain later-wins overwrite left the + /// spoofed value in place. + #[test] + fn no_authoritative_key_retains_a_lower_tier_value() { + let spoofed: serde_json::Map = AUTHORITATIVE_KEYS + .iter() + .map(|k| ((*k).to_string(), serde_json::json!("SPOOFED"))) + .collect(); + // Split across both lower tiers: policy_env and env are separate + // insertion points, and a fix that only cleared one would pass a + // single-tier test. + let agent = payload_json(serde_json::json!({ + "launch": { + "command": "goose", + "policy_env": spoofed.clone(), + "env": spoofed, + "owner_pubkey": "beef" + } + })); + let env = build(&agent).unwrap(); + for key in AUTHORITATIVE_KEYS { + assert_ne!( + env.get(*key).map(String::as_str), + Some("SPOOFED"), + "{key} kept its lower-tier value" + ); + } + // The keys the authoritative tier had no value for are gone, not + // merely different. + for absent in [ + "BUZZ_ACP_AGENT_ARGS", + "BUZZ_ACP_RESPOND_TO", + "BUZZ_ACP_RESPOND_TO_ALLOWLIST", + ] { + assert!(!env.contains_key(absent), "{absent} survived the clear"); + } + } + + /// A 64-hex pubkey, the only allowlist entry shape the harness accepts. + fn pubkey(fill: char) -> String { + std::iter::repeat_n(fill, 64).collect() + } + + /// The gate the harness refuses first (`config.rs:996-1004`). Refusing it + /// here is the difference between one error message and an unbounded + /// fail-replace loop that leaves a Secret per attempt. + #[test] + fn allowlist_mode_with_an_empty_list_is_refused() { + for empty in [serde_json::json!([]), serde_json::Value::Null] { + let agent = payload_json(serde_json::json!({ + "respond_to": "allowlist", + "respond_to_allowlist": empty, + })); + let err = build(&agent).unwrap_err(); + assert!( + err.contains("the allowlist is empty"), + "unexpected error: {err}" + ); + } + } + + /// `config.rs:629-641` — each entry must be exactly 64 hex characters. + /// The rejects are the distinct ways to miss that: too short, right length + /// but not hex, empty, and one character short of valid. + #[test] + fn an_allowlist_entry_that_is_not_64_hex_is_refused() { + for bad in ["abc1234", &"z".repeat(64), "", &pubkey('a')[..63]] { + let agent = payload_json(serde_json::json!({ + "respond_to": "allowlist", + "respond_to_allowlist": [pubkey('a'), bad], + })); + let err = build(&agent).unwrap_err(); + assert!( + err.contains("must be exactly 64 hex characters"), + "{bad:?} was accepted; error was: {err}" + ); + } + } + + /// The positive control: the guard refuses bad gates, not every gate. + /// Without this, a validator that refused unconditionally would pass both + /// tests above. + #[test] + fn a_valid_allowlist_gate_is_accepted_and_comma_joined() { + let agent = payload_json(serde_json::json!({ + "respond_to": "allowlist", + "respond_to_allowlist": [pubkey('a'), pubkey('b')], + })); + let env = build(&agent).unwrap(); + assert_eq!(env["BUZZ_ACP_RESPOND_TO"], "allowlist"); + assert_eq!( + env["BUZZ_ACP_RESPOND_TO_ALLOWLIST"], + format!("{},{}", pubkey('a'), pubkey('b')) + ); + } + + /// The harness validates the allowlist **only** in allowlist mode and + /// merely warns otherwise (`config.rs:1005-1010`). A stricter provider + /// would refuse a deploy whose identical local spawn succeeds, so this + /// pins the asymmetry rather than leaving it to look like an oversight. + #[test] + fn a_junk_allowlist_is_tolerated_outside_allowlist_mode() { + for mode in ["owner-only", "anyone"] { + let agent = payload_json(serde_json::json!({ + "respond_to": mode, + "respond_to_allowlist": ["not-a-pubkey"], + })); + let env = build(&agent) + .unwrap_or_else(|e| panic!("{mode} with a stale list must deploy: {e}")); + assert_eq!(env["BUZZ_ACP_RESPOND_TO"], mode); + } + } + + /// `respond_to` is an opaque `String` on the wire but a `clap::ValueEnum` + /// at the harness, so an unrecognized mode dies at `rc=2` — before config + /// parsing runs at all, earlier than either refusal above. Measured + /// against the built binary: `invalid value 'npub1abc' for '--respond-to'`. + /// This is the shape our own fixture carried until it was corrected. + #[test] + fn a_mode_the_harness_cannot_parse_is_refused() { + for bad in ["npub1abc", "OWNER-ONLY", "owner_only", "allowlistt", "x"] { + let agent = payload_json(serde_json::json!({ "respond_to": bad })); + let err = build(&agent).unwrap_err(); + assert!( + err.contains("is not a mode the harness accepts"), + "{bad:?} was accepted; error was: {err}" + ); + } + } + + /// `clap` does not trim its value-enum input, so a padded mode is `rc=2` + /// even though the same string trimmed is valid. Measured: `invalid value + /// ' allowlist ' for '--respond-to'`. Trimming here would accept a deploy + /// the harness refuses — the exact direction this guard exists to prevent. + #[test] + fn a_padded_mode_is_refused_because_clap_does_not_trim() { + for padded in [" allowlist ", "allowlist ", " owner-only", "\tnobody"] { + let agent = payload_json(serde_json::json!({ + "respond_to": padded, + "respond_to_allowlist": [pubkey('a')], + })); + let err = build(&agent).unwrap_err(); + assert!( + err.contains("is not a mode the harness accepts"), + "{padded:?} was accepted; error was: {err}" + ); + } + } + + /// Positive control for the mode check, and the reason it validates the + /// harness's four rather than the desktop's three: `nobody` is rejected by + /// `parse_wire` on purpose (`managed_agents/types.rs:871-880`) but starts + /// fine at the harness. A guard mirroring the desktop enum would refuse a + /// working launch from a non-desktop caller — the callers this guard is + /// for. Without this test, refusing `nobody` would pass everything above. + #[test] + fn every_mode_the_harness_accepts_is_deployable() { + for mode in ["owner-only", "allowlist", "anyone", "nobody"] { + let agent = payload_json(serde_json::json!({ + "respond_to": mode, + "respond_to_allowlist": [pubkey('a')], + })); + let env = build(&agent) + .unwrap_or_else(|e| panic!("{mode} is valid at the harness but was refused: {e}")); + assert_eq!(env["BUZZ_ACP_RESPOND_TO"], mode); + } + } +} diff --git a/crates/buzz-backend-kubernetes/src/gc.rs b/crates/buzz-backend-kubernetes/src/gc.rs new file mode 100644 index 0000000000..2e18825aa4 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/gc.rs @@ -0,0 +1,368 @@ +//! Preflight garbage collection (spec §K8s GC, `docs/remote-agents.md:1282-1335`). +//! +//! GC runs on every deploy, after identity derivation and before the state +//! transition. It deletes terminated pods (and their referenced Secrets) and +//! age-eligible orphan Secrets — every one of which must pass the full-pubkey +//! annotation check *and* carry the management marker. An unmarked object is +//! never GC'd regardless of its labels. +//! +//! The decision layer here is pure. The effectful caller supplies the observed +//! objects and the apiserver's clock; this module decides what may be deleted. + +use crate::naming::AgentIdentity; +use crate::observe::{referenced_secret, secret_is_ours}; +use chrono::{DateTime, Utc}; +use k8s_openapi::api::core::v1::{Pod, Secret}; + +/// The deploy operation deadline (spec §Deploy: `timeout: 600s`). +pub const OPERATION_DEADLINE_SECS: i64 = 600; + +/// An unreferenced Secret is GC-eligible only once it is older than **twice** +/// the deploy deadline. Rationale: Secret-create → pod-create is not atomic +/// against an independent GC pass, so without the gate a concurrent attempt's +/// preflight GC can delete a Secret whose pod has not been created yet and +/// strand that deploy. The age bound makes "unreferenced" mean "provably +/// abandoned" — any attempt that could still reference it has exceeded its own +/// deadline (`:1301-1319`). +pub const ORPHAN_SECRET_MIN_AGE_SECS: i64 = 2 * OPERATION_DEADLINE_SECS; + +/// What a GC pass decided to delete. Names only: the caller re-reads each +/// object's own fence at delete time. +#[derive(Debug, Default, PartialEq, Eq)] +pub struct GcPlan { + /// Terminated, verified, marker-bearing pods. + pub pods: Vec, + /// Age-eligible, verified, marker-bearing orphan Secrets. + pub secrets: Vec, +} + +/// Plan a GC pass. +/// +/// `now` is the apiserver's clock — the HTTP `Date` header from the very list +/// call that produced `secrets`. `None` means the header was absent or +/// unparseable, in which case **orphan-Secret GC is skipped entirely** rather +/// than falling back to local time: this provider runs on a user's desktop, +/// and a local clock fast by more than the margin does not race — it +/// deterministically computes every in-flight Secret as expired, on every +/// pass, reopening exactly the interleaving the gate exists to close +/// (`:1321-1335`). A deferred cleanup is free; a wrong deletion is not. +/// +/// Terminated-pod GC does not use the clock and is unaffected. +pub fn plan( + identity: &AgentIdentity, + pods: &[Pod], + secrets: &[Secret], + terminated: impl Fn(&Pod) -> bool, + now: Option>, +) -> GcPlan { + // Only pods that pass the full fence participate — in either direction. + // An unverified pod is neither deleted nor allowed to protect a Secret: + // it cannot be ours, so its `envFrom` cannot reference our generation. + let ours: Vec<&Pod> = pods + .iter() + .filter(|p| { + crate::observe::verify(p, identity, crate::classify::Startup::Started).is_some() + }) + .collect(); + + let doomed_pods: Vec<&&Pod> = ours.iter().filter(|p| terminated(p)).collect(); + + // A Secret referenced by ANY existing pod is protected — deliberately + // including not-yet-started pods, whose `envFrom` is exactly as + // load-bearing as a running pod's (`:1262-1264`). Pods being GC'd in this + // same pass are excluded, so their Secrets go with them. + let doomed_names: Vec<&str> = doomed_pods + .iter() + .filter_map(|p| p.metadata.name.as_deref()) + .collect(); + let protected: Vec = ours + .iter() + .filter(|p| !doomed_names.contains(&p.metadata.name.as_deref().unwrap_or_default())) + .filter_map(|p| referenced_secret(p)) + .collect(); + + let mut plan = GcPlan { + pods: doomed_names.iter().map(|n| n.to_string()).collect(), + secrets: doomed_pods + .iter() + .filter_map(|p| referenced_secret(p)) + .collect(), + }; + + // Orphan sweep: only with a server clock. + if let Some(now) = now { + for secret in secrets { + if !secret_is_ours(secret, identity) { + continue; + } + let Some(name) = secret.metadata.name.as_deref() else { + continue; + }; + if protected.contains(&name.to_string()) || plan.secrets.iter().any(|s| s == name) { + continue; + } + let Some(created) = secret.metadata.creation_timestamp.as_ref() else { + // No server-assigned timestamp means no age proof. Skip. + continue; + }; + if (now - created.0).num_seconds() >= ORPHAN_SECRET_MIN_AGE_SECS { + plan.secrets.push(name.to_string()); + } + } + } + + plan +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::naming::{ANNOTATION_PUBKEY_FULL, LABEL_MANAGED_BY}; + use k8s_openapi::api::core::v1::{Container, EnvFromSource, PodSpec, SecretEnvSource}; + use k8s_openapi::apimachinery::pkg::apis::meta::v1::{ObjectMeta, Time}; + use std::collections::BTreeMap; + + fn identity() -> AgentIdentity { + use nostr::nips::nip19::ToBech32; + let keys = nostr::Keys::generate(); + AgentIdentity::from_nsec(&keys.secret_key().to_bech32().unwrap()).unwrap() + } + + fn pod_named(id: &AgentIdentity, name: &str, secret: Option<&str>) -> Pod { + Pod { + metadata: ObjectMeta { + name: Some(name.into()), + uid: Some(format!("uid-{name}")), + resource_version: Some("1".into()), + labels: Some(id.labels()), + annotations: Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + id.pubkey_hex().to_string(), + )] + .into_iter() + .collect::>(), + ), + ..Default::default() + }, + spec: secret.map(|s| PodSpec { + containers: vec![Container { + name: "agent".into(), + env_from: Some(vec![EnvFromSource { + secret_ref: Some(SecretEnvSource { + name: s.into(), + optional: Some(false), + }), + ..Default::default() + }]), + ..Default::default() + }], + ..Default::default() + }), + ..Default::default() + } + } + + fn secret_named(id: &AgentIdentity, name: &str, age_secs: i64, now: DateTime) -> Secret { + Secret { + metadata: ObjectMeta { + name: Some(name.into()), + labels: Some(id.labels()), + annotations: Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + id.pubkey_hex().to_string(), + )] + .into_iter() + .collect::>(), + ), + creation_timestamp: Some(Time(now - chrono::Duration::seconds(age_secs))), + ..Default::default() + }, + ..Default::default() + } + } + + fn never(_: &Pod) -> bool { + false + } + fn always(_: &Pod) -> bool { + true + } + + #[test] + fn terminated_pods_and_their_secrets_are_collected_together() { + let id = identity(); + let now = Utc::now(); + let pod = pod_named(&id, "buzz-agent-dead", Some("buzz-agent-dead-gen1")); + let plan = plan(&id, &[pod], &[], always, Some(now)); + assert_eq!(plan.pods, ["buzz-agent-dead"]); + assert_eq!(plan.secrets, ["buzz-agent-dead-gen1"]); + } + + #[test] + fn live_pods_are_never_collected() { + let id = identity(); + let pod = pod_named(&id, "buzz-agent-live", Some("buzz-agent-live-gen1")); + let plan = plan(&id, &[pod], &[], never, Some(Utc::now())); + assert_eq!(plan, GcPlan::default()); + } + + /// The auto-repair fence applies to GC identically: an object that lacks + /// the marker, or carries a different pubkey, is never touched — however + /// well its labels match. + #[test] + fn unmarked_and_mismatched_objects_are_never_collected() { + let id = identity(); + let other = identity(); + let now = Utc::now(); + + let mut unmarked = pod_named(&id, "look-alike", Some("look-alike-gen1")); + let mut labels = id.labels(); + labels.remove(LABEL_MANAGED_BY); + unmarked.metadata.labels = Some(labels); + + let mut foreign = pod_named(&id, "someone-elses", Some("someone-elses-gen1")); + foreign.metadata.annotations = Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + other.pubkey_hex().to_string(), + )] + .into_iter() + .collect(), + ); + + let mut unmarked_secret = secret_named(&id, "orphan-unmarked", 100_000, now); + unmarked_secret.metadata.labels = Some(BTreeMap::new()); + let mut foreign_secret = secret_named(&id, "orphan-foreign", 100_000, now); + foreign_secret.metadata.annotations = Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + other.pubkey_hex().to_string(), + )] + .into_iter() + .collect(), + ); + + let plan = plan( + &id, + &[unmarked, foreign], + &[unmarked_secret, foreign_secret], + always, + Some(now), + ); + assert_eq!( + plan, + GcPlan::default(), + "GC touched an object it does not own" + ); + } + + /// The interleaving the age gate exists to close: attempt A creates its + /// Secret; concurrent attempt B's preflight GC runs before A creates its + /// pod. Without the gate B deletes A's Secret and strands A. + #[test] + fn young_unreferenced_secrets_are_protected() { + let id = identity(); + let now = Utc::now(); + let fresh = secret_named(&id, "buzz-agent-x-gen-inflight", 5, now); + assert_eq!( + plan(&id, &[], &[fresh], never, Some(now)), + GcPlan::default() + ); + } + + /// Past twice the deadline, any attempt that could still reference the + /// Secret has exceeded its own deadline — so it is provably abandoned. + #[test] + fn secrets_older_than_twice_the_deadline_are_collected() { + let id = identity(); + let now = Utc::now(); + let old = secret_named( + &id, + "buzz-agent-x-gen-abandoned", + ORPHAN_SECRET_MIN_AGE_SECS + 1, + now, + ); + let plan = plan(&id, &[], &[old], never, Some(now)); + assert_eq!(plan.secrets, ["buzz-agent-x-gen-abandoned"]); + } + + /// The boundary itself, both sides. `>= 1200s` is eligible. + #[test] + fn age_gate_boundary_is_exact() { + let id = identity(); + let now = Utc::now(); + let just_under = secret_named(&id, "under", ORPHAN_SECRET_MIN_AGE_SECS - 1, now); + let exactly = secret_named(&id, "exact", ORPHAN_SECRET_MIN_AGE_SECS, now); + assert!(plan(&id, &[], &[just_under], never, Some(now)) + .secrets + .is_empty()); + assert_eq!( + plan(&id, &[], &[exactly], never, Some(now)).secrets, + ["exact"] + ); + } + + /// The same-clock rule. No apiserver `Date` header → skip the orphan + /// sweep entirely. A local clock fast by more than the margin would + /// silently delete every in-flight Secret on every pass. + #[test] + fn without_a_server_clock_the_orphan_sweep_is_skipped() { + let id = identity(); + let now = Utc::now(); + let ancient = secret_named(&id, "buzz-agent-x-gen-ancient", 10_000_000, now); + let plan = plan(&id, &[], &[ancient], never, None); + assert!( + plan.secrets.is_empty(), + "orphan swept without a server clock — a fast local clock would delete live Secrets" + ); + } + + /// ...but terminated-pod GC does not consult the clock, so it still runs. + #[test] + fn terminated_pod_gc_runs_without_a_server_clock() { + let id = identity(); + let pod = pod_named(&id, "buzz-agent-dead", Some("buzz-agent-dead-gen1")); + let plan = plan(&id, &[pod], &[], always, None); + assert_eq!(plan.pods, ["buzz-agent-dead"]); + assert_eq!(plan.secrets, ["buzz-agent-dead-gen1"]); + } + + /// "Existing" includes not-yet-started pods: a Secret referenced by a pod + /// still pulling its image must not be swept, however old it is. + #[test] + fn secrets_referenced_by_a_pending_pod_are_protected() { + let id = identity(); + let now = Utc::now(); + let pending = pod_named(&id, "buzz-agent-pending", Some("buzz-agent-pending-gen1")); + let old = secret_named(&id, "buzz-agent-pending-gen1", 10_000_000, now); + let plan = plan(&id, &[pending], &[old], never, Some(now)); + assert!(plan.secrets.is_empty(), "swept a referenced Secret"); + } + + /// A Secret with no server-assigned creationTimestamp has no age proof, + /// so it is skipped rather than assumed old. + #[test] + fn secrets_without_a_creation_timestamp_are_skipped() { + let id = identity(); + let now = Utc::now(); + let mut no_timestamp = secret_named(&id, "buzz-agent-x-gen-unknown", 10_000_000, now); + no_timestamp.metadata.creation_timestamp = None; + assert!(plan(&id, &[], &[no_timestamp], never, Some(now)) + .secrets + .is_empty()); + } + + /// A Secret belonging to a pod being collected in this same pass goes with + /// it, and must not be listed twice. + #[test] + fn a_collected_pods_secret_is_listed_once() { + let id = identity(); + let now = Utc::now(); + let dead = pod_named(&id, "buzz-agent-dead", Some("buzz-agent-dead-gen1")); + let its_secret = secret_named(&id, "buzz-agent-dead-gen1", 10_000_000, now); + let plan = plan(&id, &[dead], &[its_secret], always, Some(now)); + assert_eq!(plan.secrets, ["buzz-agent-dead-gen1"]); + } +} diff --git a/crates/buzz-backend-kubernetes/src/image.rs b/crates/buzz-backend-kubernetes/src/image.rs new file mode 100644 index 0000000000..409989fda9 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/image.rs @@ -0,0 +1,184 @@ +//! Image reference validation (spec §Image). +//! +//! The object holding this reference runs with an nsec, so the reference must +//! be **immutable**. Registry tags are mutable pointers — Kubernetes itself +//! distinguishes them from digests for exactly this reason — so a tag-only +//! reference is rejected, not just `:latest`. +//! +//! There is no parse-time fallback: `image` is required, and its absence +//! fails closed with a named field. The published `ghcr.io/block/buzz-sprig` +//! digest is offered only as a schema `default` (a UI prefill the desktop +//! submits explicitly — see `config::DEFAULT_IMAGE`), so the create-intent +//! fingerprint never depends on compiled-in provider state. + +/// A validated, digest-qualified image reference. +/// +/// The inner string is always in canonical tagless form `name@sha256:`: +/// `name:tag@sha256:…` normalizes by dropping the tag, because the tag is +/// decorative once a digest pins the content, and leaving it in would make +/// two references to identical bytes produce different create-intent +/// fingerprints. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct ImageRef(String); + +impl ImageRef { + pub fn as_str(&self) -> &str { + &self.0 + } +} + +impl std::fmt::Display for ImageRef { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + f.write_str(&self.0) + } +} + +/// Parse and normalize a user-supplied image reference. +pub fn parse(raw: &str) -> Result { + let reference = raw.trim(); + if reference.is_empty() { + return Err("provider_config.image is required: no image is assumed \ + at parse time, so the digest-pinned image to run must be \ + given explicitly" + .to_string()); + } + + let mut parts = reference.split('@'); + let name_and_tag = parts.next().unwrap_or_default(); + let digest = match (parts.next(), parts.next()) { + (Some(d), None) => d, + (None, _) => { + return Err(format!( + "provider_config.image {reference:?} is not digest-pinned: a \ + tag is a mutable pointer, and this object runs with the \ + agent's private key. Use name@sha256:<64 hex chars>" + )) + } + (Some(_), Some(_)) => { + return Err(format!( + "provider_config.image {reference:?} contains more than one '@'" + )) + } + }; + + let hex = digest.strip_prefix("sha256:").ok_or_else(|| { + format!("provider_config.image digest {digest:?} must start with 'sha256:'") + })?; + // Lowercase only: OCI canonicalizes digest hex, and accepting uppercase + // would let two spellings of one digest produce two fingerprints. + if hex.len() != 64 + || !hex + .chars() + .all(|c| c.is_ascii_digit() || ('a'..='f').contains(&c)) + { + return Err(format!( + "provider_config.image digest {digest:?} must be exactly 64 \ + lowercase hex characters" + )); + } + + // Drop any tag: `name:tag@sha256:…` and `name@sha256:…` name the same + // bytes and must fingerprint identically. Only a *final* colon segment + // that isn't a port counts as a tag — `host:5000/name` has no tag. + let name = match name_and_tag.rfind(':') { + Some(colon) if !name_and_tag[colon + 1..].contains('/') => &name_and_tag[..colon], + _ => name_and_tag, + }; + if name.is_empty() { + return Err(format!( + "provider_config.image {reference:?} has no repository name" + )); + } + + Ok(ImageRef(format!("{name}@sha256:{hex}"))) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn accepts_digest_pinned_reference() { + let d = "a".repeat(64); + let r = parse(&format!("ghcr.io/block/buzz-sprig@sha256:{d}")).unwrap(); + assert_eq!(r.as_str(), format!("ghcr.io/block/buzz-sprig@sha256:{d}")); + } + + /// The normalization that keeps the fingerprint stable: two spellings of + /// the same bytes must produce one reference. + #[test] + fn strips_tag_from_tag_plus_digest_form() { + let d = "b".repeat(64); + let tagged = parse(&format!("ghcr.io/block/buzz-sprig:v1.2@sha256:{d}")).unwrap(); + let plain = parse(&format!("ghcr.io/block/buzz-sprig@sha256:{d}")).unwrap(); + assert_eq!(tagged, plain); + } + + /// A registry port is not a tag. `host:5000/name` must keep its port. + #[test] + fn registry_port_is_not_mistaken_for_a_tag() { + let d = "c".repeat(64); + let r = parse(&format!("localhost:5000/buzz-sprig@sha256:{d}")).unwrap(); + assert_eq!(r.as_str(), format!("localhost:5000/buzz-sprig@sha256:{d}")); + } + + #[test] + fn port_and_tag_together_drops_only_the_tag() { + let d = "d".repeat(64); + let r = parse(&format!("localhost:5000/buzz-sprig:dev@sha256:{d}")).unwrap(); + assert_eq!(r.as_str(), format!("localhost:5000/buzz-sprig@sha256:{d}")); + } + + /// Wren's amendment: *every* tag-only reference is rejected, not just + /// `:latest`. A `sha-` tag is traceable but still movable. + #[test] + fn rejects_every_tag_only_reference() { + for bad in [ + "ghcr.io/block/buzz-sprig:latest", + "ghcr.io/block/buzz-sprig:v1.2.3", + "ghcr.io/block/buzz-sprig:sha-abc1234", + "ghcr.io/block/buzz-sprig", + "localhost:5000/buzz-sprig", + ] { + let err = parse(bad).unwrap_err(); + assert!(err.contains("digest-pinned"), "for {bad:?} got: {err}"); + } + } + + /// Uppercase hex is a second spelling of one digest; accepting it would + /// let the same image fingerprint two ways. + #[test] + fn rejects_uppercase_digest_hex() { + let d = "A".repeat(64); + assert!(parse(&format!("img@sha256:{d}")).is_err()); + } + + #[test] + fn rejects_malformed_digests() { + let short = "a".repeat(63); + let long = "a".repeat(65); + let ok = "a".repeat(64); + for bad in [ + format!("img@sha256:{short}"), + format!("img@sha256:{long}"), + format!("img@sha512:{ok}"), + format!("img@{ok}"), + format!("img@sha256:{}", "g".repeat(64)), + format!("img@sha256:{ok}@sha256:{ok}"), + format!("@sha256:{ok}"), + ] { + assert!(parse(&bad).is_err(), "accepted {bad:?}"); + } + } + + /// Parsing has no fallback (the schema default is a UI prefill, not a + /// parse-time substitute), so an absent image is an error that names the + /// field rather than a silent fallback. + #[test] + fn empty_reference_names_the_field() { + for empty in ["", " "] { + let err = parse(empty).unwrap_err(); + assert!(err.contains("provider_config.image"), "got: {err}"); + } + } +} diff --git a/crates/buzz-backend-kubernetes/src/intent.rs b/crates/buzz-backend-kubernetes/src/intent.rs new file mode 100644 index 0000000000..73218f83bf --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/intent.rs @@ -0,0 +1,309 @@ +//! The create-intent fingerprint (spec §Deploy State Machine, create-intent +//! fingerprint, `docs/remote-agents.md:796-828`). +//! +//! The fingerprint is an unkeyed SHA-256 over a canonical serialization of the +//! provider's non-secret create-intent template. A plain hash is safe *only* +//! because of the scope rule: the input covers exactly the provider-controlled +//! fields that can affect scheduling or container creation, and **never Secret +//! data or attempt identity**. Hashing low-entropy secrets into a +//! world-readable annotation would be a dictionary oracle. +//! +//! That rule is enforced structurally rather than remembered. [`IntentTemplate`] +//! is a *pre-binding* type: it has no field that can hold Secret material or a +//! generation token, so there is no expression that hashes one. The +//! per-attempt Secret name never appears — the pod's `envFrom` is represented +//! by the fixed [`SECRET_PLACEHOLDER`], because otherwise every attempt would +//! diverge from every other by construction. +//! +//! Server- and admission-produced output (UID, `resourceVersion`, timestamps, +//! defaulted fields, the annotation itself) is excluded the same way: the +//! serializer is only ever handed this template, never a live `Pod`, so the +//! exclusion is checkable by inspection. + +use crate::image::ImageRef; +use serde::Serialize; +use sha2::{Digest, Sha256}; + +/// Stands in for the per-attempt Secret name in the `envFrom` position. +/// A real generation token here would make every attempt diverge. +const SECRET_PLACEHOLDER: &str = ""; + +/// The recorded/computed create intent: a hex SHA-256 digest. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Fingerprint(String); + +impl Fingerprint { + pub fn as_str(&self) -> &str { + &self.0 + } + + /// Read a fingerprint off a pod annotation. Any recorded string is + /// accepted verbatim: comparison is equality against a freshly computed + /// value, so a malformed annotation simply reads as divergence — which is + /// the correct outcome for a pod this provider version did not write. + pub fn from_annotation(value: &str) -> Self { + Self(value.to_string()) + } + + #[cfg(test)] + pub fn for_test(seed: &str) -> Self { + Self(format!("test-{seed}")) + } +} + +impl std::fmt::Display for Fingerprint { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + f.write_str(&self.0) + } +} + +/// The non-secret, pre-binding description of the pod this deploy would +/// create. Every field is provider-controlled and scheduling-relevant; there +/// is deliberately no field for env values, Secret data, or the generation. +#[derive(Debug, Clone, PartialEq, Eq, Serialize)] +pub struct IntentTemplate { + /// Schema version of the template itself. Bumping it re-fingerprints every + /// pod, which is the intended way to roll out a pod-shape change. + pub template_version: u32, + pub namespace: String, + /// Normalized, digest-qualified image reference. + pub image: String, + pub cpu_request: String, + pub memory_request: String, + pub cpu_limit: String, + pub memory_limit: String, + pub service_account: Option, + pub restart_policy: &'static str, + pub termination_grace_period_seconds: i64, + /// Env *keys* only, sorted. Keys are pod-shape (a renamed key changes the + /// container's contract); values are Secret material and must not be here. + pub env_keys: Vec, + /// Fixed placeholder for the per-attempt Secret in `envFrom`. + pub env_from_secret: &'static str, + pub workspace_mount_path: String, + pub run_as_user: i64, + pub run_as_group: i64, +} + +/// Current template schema version. +pub const TEMPLATE_VERSION: u32 = 1; + +impl IntentTemplate { + /// Compute the fingerprint. `serde_json` on a struct with declared field + /// order plus pre-sorted `env_keys` is a canonical serialization: the same + /// template always produces the same bytes. + pub fn fingerprint(&self) -> Fingerprint { + let canonical = serde_json::to_vec(self).expect("intent template is plain data"); + Fingerprint(hex::encode(Sha256::digest(&canonical))) + } + + /// Build from resolved pod-shape inputs. `env_keys` is sorted here rather + /// than at the call site so key ordering can never leak into the digest. + /// + /// The fixed pod-shape constants are read from [`crate::config`] rather + /// than passed in: `pod::build_pod` stamps the pod from those same + /// constants, so the fingerprint cannot describe a pod shape different + /// from the one actually created. Threading them through as arguments + /// would make that agreement a thing to test instead of a thing that holds. + pub fn new( + namespace: &str, + image: &ImageRef, + resources: &crate::config::Resources, + service_account: Option<&str>, + env_keys: impl IntoIterator, + ) -> Self { + let mut env_keys: Vec = env_keys.into_iter().collect(); + env_keys.sort(); + Self { + template_version: TEMPLATE_VERSION, + namespace: namespace.to_string(), + image: image.as_str().to_string(), + cpu_request: resources.cpu_request.clone(), + memory_request: resources.memory_request.clone(), + cpu_limit: resources.cpu_limit.clone(), + memory_limit: resources.memory_limit.clone(), + service_account: service_account.map(str::to_string), + restart_policy: crate::config::RESTART_POLICY, + termination_grace_period_seconds: crate::config::TERMINATION_GRACE_SECONDS, + env_keys, + env_from_secret: SECRET_PLACEHOLDER, + workspace_mount_path: crate::config::WORKSPACE_PATH.to_string(), + run_as_user: crate::config::RUN_AS_UID, + run_as_group: crate::config::RUN_AS_GID, + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::config::Resources; + + fn image(byte: char) -> ImageRef { + crate::image::parse(&format!( + "ghcr.io/block/buzz-sprig@sha256:{}", + byte.to_string().repeat(64) + )) + .unwrap() + } + + fn template() -> IntentTemplate { + IntentTemplate::new( + "buzz-agents", + &image('a'), + &Resources::default(), + None, + ["BUZZ_RELAY_URL".to_string(), "GOOSE_MODE".to_string()], + ) + } + + #[test] + fn fingerprint_is_deterministic() { + assert_eq!(template().fingerprint(), template().fingerprint()); + } + + #[test] + fn fingerprint_is_hex_sha256() { + let fp = template().fingerprint(); + assert_eq!(fp.as_str().len(), 64); + assert!(fp.as_str().chars().all(|c| c.is_ascii_hexdigit())); + } + + /// Key *order* must not reach the digest, or two identical environments + /// built in different orders would look like a config change. + #[test] + fn env_key_order_does_not_affect_the_digest() { + let a = IntentTemplate::new( + "ns", + &image('a'), + &Resources::default(), + None, + ["A".to_string(), "B".to_string(), "C".to_string()], + ); + let b = IntentTemplate::new( + "ns", + &image('a'), + &Resources::default(), + None, + ["C".to_string(), "A".to_string(), "B".to_string()], + ); + assert_eq!(a.fingerprint(), b.fingerprint()); + } + + /// A mutation applied to a fresh template clone, named for its assertion + /// message. + type Mutation = (&'static str, Box); + + /// Every scheduling-relevant knob must move the digest — this is the + /// wedge escape (§Deploy State Machine never-started recoverable row). + /// Exhaustive by construction: each mutation is applied to a fresh clone. + #[test] + fn every_scheduling_field_changes_the_digest() { + let base = template(); + let baseline = base.fingerprint(); + + let mutations: Vec = vec![ + ( + "template_version", + Box::new(|t: &mut IntentTemplate| t.template_version += 1), + ), + ( + "namespace", + Box::new(|t: &mut IntentTemplate| t.namespace = "other".into()), + ), + ( + "image", + Box::new(|t: &mut IntentTemplate| t.image = image('b').as_str().into()), + ), + ( + "cpu_request", + Box::new(|t: &mut IntentTemplate| t.cpu_request = "4".into()), + ), + ( + "memory_request", + Box::new(|t: &mut IntentTemplate| t.memory_request = "8Gi".into()), + ), + ( + "cpu_limit", + Box::new(|t: &mut IntentTemplate| t.cpu_limit = "8".into()), + ), + ( + "memory_limit", + Box::new(|t: &mut IntentTemplate| t.memory_limit = "16Gi".into()), + ), + ( + "service_account", + Box::new(|t: &mut IntentTemplate| t.service_account = Some("sa".into())), + ), + ( + "restart_policy", + Box::new(|t: &mut IntentTemplate| t.restart_policy = "OnFailure"), + ), + ( + "grace_period", + Box::new(|t: &mut IntentTemplate| t.termination_grace_period_seconds = 30), + ), + ( + "env_keys", + Box::new(|t: &mut IntentTemplate| t.env_keys.push("NEW_KEY".into())), + ), + ( + "workspace_mount_path", + Box::new(|t: &mut IntentTemplate| t.workspace_mount_path = "/w".into()), + ), + ( + "run_as_user", + Box::new(|t: &mut IntentTemplate| t.run_as_user = 2000), + ), + ( + "run_as_group", + Box::new(|t: &mut IntentTemplate| t.run_as_group = 2000), + ), + ]; + + for (name, mutate) in mutations { + let mut t = base.clone(); + mutate(&mut t); + assert_ne!( + t.fingerprint(), + baseline, + "{name} did not affect the digest" + ); + } + } + + /// The scope rule, asserted on the bytes: no Secret value and no + /// generation token can appear in the serialization, because the type has + /// nowhere to put them. The placeholder is what `envFrom` contributes. + #[test] + fn serialization_contains_no_secret_material_or_attempt_identity() { + let json = serde_json::to_string(&template()).unwrap(); + for forbidden in ["nsec1", "SPOOFED", "wss://", "gen0001"] { + assert!( + !json.contains(forbidden), + "template leaked {forbidden}: {json}" + ); + } + assert!(json.contains(SECRET_PLACEHOLDER)); + } + + /// Two attempts for the same agent differ only in generation, which is + /// absent from the template — so their fingerprints must be equal, or the + /// divergence discriminator would fire on every single deploy. + #[test] + fn attempts_differing_only_by_generation_do_not_diverge() { + // There is no generation input to pass; that *is* the property. The + // test states it explicitly so a future field addition breaks here. + assert_eq!(template().fingerprint(), template().fingerprint()); + let json = serde_json::to_string(&template()).unwrap(); + assert_eq!(json.matches(SECRET_PLACEHOLDER).count(), 1); + } + + /// A recorded annotation this provider version did not write reads as + /// divergence rather than an error. + #[test] + fn unrecognized_annotation_reads_as_divergence() { + let recorded = Fingerprint::from_annotation("not-a-digest"); + assert_ne!(recorded, template().fingerprint()); + } +} diff --git a/crates/buzz-backend-kubernetes/src/main.rs b/crates/buzz-backend-kubernetes/src/main.rs new file mode 100644 index 0000000000..5d521a9d37 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/main.rs @@ -0,0 +1,199 @@ +//! Kubernetes backend provider for Buzz remote agents +//! (spec `docs/remote-agents.md`). +//! +//! One process per operation: read exactly one JSON request from stdin, write +//! exactly one JSON response to stdout, exit. The exit code carries exactly +//! one bit — 0 for a response that was produced, 1 for a failure to produce +//! one. Everything a caller needs to distinguish is *inside* the response's +//! `ok` field, because a provider that encoded outcomes in exit codes would +//! have a second, redundant error channel to keep in sync (§Provider Protocol). + +mod classify; +mod client; +mod cluster; +mod config; +mod env; +mod gc; +mod image; +mod intent; +mod naming; +mod observe; +mod pod; +mod reconcile; +mod wire; + +use std::io::Read; +use wire::{Request, Response}; + +/// The provider a shared-compute agent resolves to. Refused here as the +/// spec's backstop: a mesh agent runs on the relay's compute, so deploying it +/// as a pod would create a second, contending consumer of the same agent +/// identity (`:214-219`). +const RELAY_MESH_PROVIDER: &str = "relay-mesh"; + +fn main() { + // rustls needs a process-level provider before the first TLS connection. + // The release build compiles every sidecar in one cargo invocation, which + // unifies the `ring` and `aws-lc-rs` features and leaves rustls unable to + // auto-select — so this is an explicit install, not a default. + let _ = rustls::crypto::ring::default_provider().install_default(); + + let mut input = String::new(); + if let Err(e) = std::io::stdin().read_to_string(&mut input) { + // No request means no request_id and no response contract to honor. + // This is the one path that exits nonzero. + eprintln!("could not read the request from stdin: {e}"); + std::process::exit(1); + } + + let response = respond(&input); + println!( + "{}", + serde_json::to_string(&response).unwrap_or_else(|e| { + // The response types are plain data; this cannot fail in practice, + // and a hand-built object is still a conforming response. + format!(r#"{{"ok":false,"error":"could not serialize a response: {e}"}}"#) + }) + ); +} + +/// Produce the single response for one request. Separated from `main` so the +/// whole dispatch is testable without a process. +fn respond(input: &str) -> Response { + // Parsed as raw JSON first: the relay-mesh refusal below MUST see the wire + // value, and `AgentPayload` deliberately does not carry `provider`. + let raw: serde_json::Value = match serde_json::from_str(input) { + Ok(value) => value, + Err(e) => return Response::error(format!("request is not valid JSON: {e}")), + }; + + if let Some(refusal) = refuse_relay_mesh(&raw) { + return Response::error(refusal); + } + + let request: Request = match serde_json::from_value(raw) { + Ok(request) => request, + Err(e) => return Response::error(format!("could not understand the request: {e}")), + }; + + match request { + Request::Info => Response::info(), + Request::Deploy(deploy) => { + let runtime = match tokio::runtime::Builder::new_current_thread() + .enable_all() + .build() + { + Ok(runtime) => runtime, + Err(e) => return Response::error(format!("could not start the runtime: {e}")), + }; + match runtime.block_on(deploy_agent(&deploy)) { + Ok(agent_id) => Response::deployed(agent_id), + Err(e) => Response::error(e), + } + } + } +} + +/// Refuse a shared-compute agent, reading the **raw wire value**. +/// +/// Trimmed before comparing: the desktop's own layers disagree about padding +/// (`relay_mesh.rs:17` and `effective_config/mod.rs:46` trim; the deploy guard +/// at `agents_deploy.rs:116` did not), and `non_blank` preserves surrounding +/// whitespace on a non-blank value. A backstop that shares its bypass with the +/// layer it backs is not a backstop. +fn refuse_relay_mesh(raw: &serde_json::Value) -> Option { + let provider = raw.get("agent")?.get("provider")?.as_str()?; + (provider.trim() == RELAY_MESH_PROVIDER).then(|| { + "deploy refused: this agent is configured for shared compute \ + (relay-mesh), which runs on the relay rather than in a pod. \ + Switch the agent to a local runtime before deploying it to \ + Kubernetes." + .to_string() + }) +} + +/// Run one deploy to a terminal outcome. +async fn deploy_agent(request: &wire::DeployRequest) -> Result { + let cfg = config::parse(&request.provider_config)?; + // Identity before any cluster contact: a malformed nsec is a refusal, not + // a failed connection (§Deploy State Machine step 0). + let identity = naming::AgentIdentity::from_nsec(&request.agent.private_key_nsec)?; + + // One generation for this operation's first attempt; the reconciler mints + // its own per attempt and restamps the correlator to match. + let env = env::build_env( + &request.agent, + env::AuthoritativeInputs { + generation: &naming::new_generation(), + inactivity_seconds: cfg.inactivity_seconds, + }, + )?; + + let client = client::connect(cfg.context.as_deref()).await?; + let substrate = cluster::Cluster::new(client, &cfg.namespace); + reconcile::deploy(&substrate, &identity, &cfg, env).await +} + +#[cfg(test)] +mod tests { + use super::*; + + fn error_of(response: &Response) -> String { + let json = serde_json::to_value(response).unwrap(); + assert_eq!(json["ok"], false, "expected a refusal: {json}"); + json["error"].as_str().unwrap().to_string() + } + + /// The spec's backstop for the relay-mesh MUST. The desktop refuses first + /// (`agents_deploy.rs:116`); this is the layer that owes the obligation. + #[test] + fn refuses_a_relay_mesh_agent() { + let request = r#"{"op":"deploy","agent":{ + "relay_url":"wss://r","private_key_nsec":"nsec1x","provider":"relay-mesh"}, + "provider_config":{"namespace":"ns"}}"#; + assert!(error_of(&respond(request)).contains("relay-mesh")); + } + + /// Padding must not bypass the backstop. Reachable by construction: + /// `GlobalConfig.provider` is a bare `Option` with no trim on + /// write, and `non_blank` rejects whitespace-only while preserving + /// surrounding whitespace on everything else. + #[test] + fn refuses_a_padded_relay_mesh_agent() { + let request = r#"{"op":"deploy","agent":{ + "relay_url":"wss://r","private_key_nsec":"nsec1x","provider":" relay-mesh "}, + "provider_config":{"namespace":"ns"}}"#; + assert!(error_of(&respond(request)).contains("relay-mesh")); + } + + /// The refusal must not fire on a normal agent — a guard that refuses + /// everything passes its own test and ships a provider that deploys + /// nothing. + #[test] + fn does_not_refuse_a_normal_provider() { + let raw: serde_json::Value = + serde_json::from_str(r#"{"agent":{"provider":"openai"}}"#).unwrap(); + assert!(refuse_relay_mesh(&raw).is_none()); + // …nor when the field is absent entirely, which is the common case: + // `AgentPayload` does not carry `provider`. + let bare: serde_json::Value = serde_json::from_str(r#"{"agent":{}}"#).unwrap(); + assert!(refuse_relay_mesh(&bare).is_none()); + } + + /// Malformed input still produces exactly one conforming response. + #[test] + fn malformed_input_is_an_in_band_error() { + assert!(error_of(&respond("not json")).contains("valid JSON")); + assert!(error_of(&respond(r#"{"op":"undeploy"}"#)).contains("understand")); + } + + /// `info` answers without touching a cluster — it is what the desktop + /// calls to render the config form, before any kubeconfig exists. + #[test] + fn info_answers_with_the_protocol_version_and_schema() { + let json = serde_json::to_value(respond(r#"{"op":"info"}"#)).unwrap(); + assert_eq!(json["ok"], true); + assert_eq!(json["protocol_version"], wire::PROTOCOL_VERSION); + assert!(json["config_schema"]["properties"]["namespace"].is_object()); + } +} diff --git a/crates/buzz-backend-kubernetes/src/naming.rs b/crates/buzz-backend-kubernetes/src/naming.rs new file mode 100644 index 0000000000..4b9d7ea035 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/naming.rs @@ -0,0 +1,223 @@ +//! Identity derivation and the object-naming contract (spec §Pod shape). +//! +//! Every name, label, and annotation below is derived from the pubkey the +//! provider decoded itself from `private_key_nsec` — never from a +//! caller-supplied pubkey (§Deploy State Machine step 0). + +use nostr::nips::nip19::FromBech32; + +/// `app.kubernetes.io/managed-by` value: the management marker's identity half. +pub const MANAGED_BY: &str = "buzz-backend-kubernetes"; + +/// Label key carrying [`MANAGED_BY`]. +pub const LABEL_MANAGED_BY: &str = "app.kubernetes.io/managed-by"; + +/// Label key carrying [`BINDING_VERSION`] — the marker's schema half. +pub const LABEL_BINDING_VERSION: &str = "buzz.block.xyz/binding-version"; + +/// Schema version of the object layout this provider writes. Bumped when the +/// pod/Secret shape changes in a way a older provider would mis-handle. +pub const BINDING_VERSION: &str = "1"; + +/// Label key: truncated pubkey, the reconciliation and GC selector. +pub const LABEL_AGENT_PUBKEY: &str = "buzz.block.xyz/agent-pubkey"; + +/// Annotation key: full pubkey. Load-bearing — the truncated label is +/// collision-*resistant*, this is what makes it safe (§Deploy State Machine +/// step 1). +pub const ANNOTATION_PUBKEY_FULL: &str = "buzz.block.xyz/agent-pubkey-full"; + +/// Annotation key: the recorded create-intent fingerprint. +pub const ANNOTATION_CREATE_INTENT: &str = "buzz.block.xyz/create-intent"; + +/// Annotation key: the image reference this generation actually resolved to, +/// for post-hoc attribution (§Image). +pub const ANNOTATION_IMAGE: &str = "buzz.block.xyz/image"; + +/// An agent identity the provider derived itself, plus every name it implies. +/// +/// Constructing this type is the *only* way to obtain the names — so a +/// caller-supplied pubkey cannot reach a selector by any path. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct AgentIdentity { + pubkey_hex: String, +} + +impl AgentIdentity { + /// Derive from the payload's `private_key_nsec`. + /// + /// Accepts bech32 `nsec1…`; a malformed or undecodable key is an + /// immediate error, before any substrate read or mutation + /// (§Deploy State Machine step 0). + pub fn from_nsec(nsec: &str) -> Result { + let secret = nostr::SecretKey::from_bech32(nsec.trim()) + .map_err(|_| "private_key_nsec is not a decodable nsec1 key".to_string())?; + let keys = nostr::Keys::new(secret); + Ok(Self { + pubkey_hex: keys.public_key().to_hex(), + }) + } + + /// Full 64-hex public key — the annotation value and the comparison + /// operand for candidate authentication. + pub fn pubkey_hex(&self) -> &str { + &self.pubkey_hex + } + + /// Selector label value: first 32 hex chars (128 bits). A full hex pubkey + /// is 64 chars and label values cap at 63, which is why this is truncated + /// and why the annotation check is normative rather than decorative. + pub fn label_pubkey(&self) -> &str { + &self.pubkey_hex[..32] + } + + /// Deterministic pod name, also the returned `agent_id`. + pub fn pod_name(&self) -> String { + format!("buzz-agent-{}", &self.pubkey_hex[..12]) + } + + /// Per-attempt Secret name. `generation` is a fresh random token per + /// create attempt — never reused — which is what makes payload and Secret + /// atomic at the pod-spec boundary (§K8s Secrets). + pub fn secret_name(&self, generation: &str) -> String { + format!("buzz-agent-{}-{}", &self.pubkey_hex[..12], generation) + } + + /// Label selector matching this identity's objects *and* our management + /// marker. Selecting on the marker as well as the identity means an + /// unmarked look-alike never even enters the candidate list. + pub fn selector(&self) -> String { + format!( + "{LABEL_AGENT_PUBKEY}={},{LABEL_MANAGED_BY}={MANAGED_BY}", + self.label_pubkey() + ) + } + + /// The label set stamped on every object this provider creates. + pub fn labels(&self) -> std::collections::BTreeMap { + [ + ( + LABEL_AGENT_PUBKEY.to_string(), + self.label_pubkey().to_string(), + ), + (LABEL_MANAGED_BY.to_string(), MANAGED_BY.to_string()), + ( + LABEL_BINDING_VERSION.to_string(), + BINDING_VERSION.to_string(), + ), + ] + .into_iter() + .collect() + } +} + +/// A fresh generation token: 8 lowercase hex chars from the OS RNG. +/// +/// Appears in the Secret name and as `BUZZ_MANAGED_AGENT_START_NONCE`, so the +/// Secret generation and the harness's lifecycle-frame correlator are one +/// identity (§Launch data tier 3). +pub fn new_generation() -> String { + use rand::RngExt; + let n: u32 = rand::rng().random(); + format!("{n:08x}") +} + +#[cfg(test)] +mod tests { + use super::*; + + /// A fixed test key. Deriving the pubkey (rather than hardcoding both + /// halves) is the point: the test exercises the same derivation the + /// reconciler depends on. + fn identity() -> AgentIdentity { + let keys = nostr::Keys::generate(); + let nsec = { + use nostr::nips::nip19::ToBech32; + keys.secret_key().to_bech32().unwrap() + }; + let id = AgentIdentity::from_nsec(&nsec).unwrap(); + assert_eq!(id.pubkey_hex(), keys.public_key().to_hex()); + id + } + + #[test] + fn rejects_malformed_nsec() { + for bad in ["", "nsec1", "not-a-key", "npub1abc"] { + assert!( + AgentIdentity::from_nsec(bad).is_err(), + "accepted malformed key {bad:?}" + ); + } + } + + #[test] + fn tolerates_surrounding_whitespace() { + let keys = nostr::Keys::generate(); + use nostr::nips::nip19::ToBech32; + let nsec = keys.secret_key().to_bech32().unwrap(); + let padded = format!(" {nsec}\n"); + assert_eq!( + AgentIdentity::from_nsec(&padded).unwrap().pubkey_hex(), + keys.public_key().to_hex() + ); + } + + /// Kubernetes label *values* cap at 63 chars; a full hex pubkey is 64, + /// one over. That one-char overflow is the whole reason the selector is + /// truncated, so it gets an explicit test. + #[test] + fn label_value_fits_kubernetes_limit() { + let id = identity(); + assert_eq!(id.pubkey_hex().len(), 64); + assert_eq!(id.label_pubkey().len(), 32); + assert!(id.label_pubkey().len() <= 63); + } + + #[test] + fn pod_name_is_deterministic_and_dns_safe() { + let id = identity(); + assert_eq!(id.pod_name(), id.pod_name()); + assert_eq!( + id.pod_name(), + format!("buzz-agent-{}", &id.pubkey_hex()[..12]) + ); + assert!(id.pod_name().len() <= 253); + assert!(id + .pod_name() + .chars() + .all(|c| c.is_ascii_lowercase() || c.is_ascii_digit() || c == '-')); + } + + /// Two attempts must never share a Secret name — that uniqueness is what + /// stops a losing contender from overwriting the winner's identity. + #[test] + fn secret_names_are_per_attempt() { + let id = identity(); + let a = id.secret_name(&new_generation()); + let b = id.secret_name(&new_generation()); + assert_ne!(a, b); + assert!(a.starts_with(&id.pod_name())); + assert!(a.len() <= 253); + } + + #[test] + fn selector_requires_the_management_marker() { + let id = identity(); + let sel = id.selector(); + assert!(sel.contains(&format!("{LABEL_AGENT_PUBKEY}={}", id.label_pubkey()))); + assert!(sel.contains(&format!("{LABEL_MANAGED_BY}={MANAGED_BY}"))); + } + + #[test] + fn every_created_object_carries_the_marker() { + let labels = identity().labels(); + assert_eq!( + labels.get(LABEL_MANAGED_BY).map(String::as_str), + Some(MANAGED_BY) + ); + assert_eq!( + labels.get(LABEL_BINDING_VERSION).map(String::as_str), + Some(BINDING_VERSION) + ); + } +} diff --git a/crates/buzz-backend-kubernetes/src/observe.rs b/crates/buzz-backend-kubernetes/src/observe.rs new file mode 100644 index 0000000000..1c6c883563 --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/observe.rs @@ -0,0 +1,592 @@ +//! Decoding API objects into verified observations (spec §Deploy State +//! Machine step 1). +//! +//! Pure: `Pod` in, [`VerifiedPod`] out. Keeping the decode here means the +//! conformance tests drive the *shipped* decoder with real API types rather +//! than a test-only stand-in, and it keeps `classify.rs` free of API types. +//! +//! Verification is the gate, not a filter: [`verify`] returns `None` for any +//! object whose full-pubkey annotation does not equal the derived pubkey or +//! that lacks the management marker, so an unverified object cannot reach +//! classification, deletion, or the returned `agent_id`. + +use crate::classify::{Fence, PullFailure, Startup, VerifiedPod}; +use crate::intent::Fingerprint; +use crate::naming::{ + AgentIdentity, ANNOTATION_CREATE_INTENT, ANNOTATION_PUBKEY_FULL, BINDING_VERSION, + LABEL_BINDING_VERSION, LABEL_MANAGED_BY, MANAGED_BY, +}; +use k8s_openapi::api::core::v1::{Pod, Secret}; + +/// Container name the provider creates; status is read from this container. +pub const CONTAINER_NAME: &str = "agent"; + +/// The startup state, or a state that cannot be settled without one more read. +#[derive(Debug, Clone, PartialEq, Eq)] +pub enum StartupObservation { + Resolved(Startup), + /// `CreateContainerConfigError` — recoverable *unless* the referenced + /// Secret is confirmed absent by a most-recent read. The kubelet's reason + /// string is a hint; the provider verifies before treating it as fatal + /// (§Deploy State Machine: "provably" means a verified absence, never a + /// reason string). + ConfigErrorPendingSecretCheck { + secret_name: String, + }, +} + +/// Does this object carry the management marker (§Pod shape)? +/// +/// Identity labels prove identity; the marker asserts protocol ownership. +/// Without it an object that merely matches our schema fails closed. +fn has_marker(labels: Option<&std::collections::BTreeMap>) -> bool { + let Some(labels) = labels else { return false }; + labels.get(LABEL_MANAGED_BY).map(String::as_str) == Some(MANAGED_BY) + && labels.get(LABEL_BINDING_VERSION).map(String::as_str) == Some(BINDING_VERSION) +} + +/// Does the full-pubkey annotation equal the derived pubkey? +/// +/// The 32-hex label is collision-*resistant*, not collision-free, which is +/// why this check is normative rather than decorative (`:1152-1166`). +fn annotation_matches( + annotations: Option<&std::collections::BTreeMap>, + identity: &AgentIdentity, +) -> bool { + annotations + .and_then(|a| a.get(ANNOTATION_PUBKEY_FULL)) + .map(|v| v == identity.pubkey_hex()) + .unwrap_or(false) +} + +/// Is this Secret ours and this identity's? The same fence GC applies before +/// deleting anything. +pub fn secret_is_ours(secret: &Secret, identity: &AgentIdentity) -> bool { + has_marker(secret.metadata.labels.as_ref()) + && annotation_matches(secret.metadata.annotations.as_ref(), identity) +} + +/// Decode a pod's startup state from its status. +/// +/// "Started" means `state.running` on our container — not pod phase. A pod can +/// sit in phase `Running` with a container that never started, and a pod being +/// gracefully deleted stays in phase `Running` for its whole grace period. +pub fn decode_startup(pod: &Pod) -> StartupObservation { + use StartupObservation::Resolved; + + let status = pod.status.as_ref(); + let phase = status.and_then(|s| s.phase.as_deref()); + + let container = status + .and_then(|s| s.container_statuses.as_ref()) + .and_then(|cs| cs.iter().find(|c| c.name == CONTAINER_NAME)); + + if let Some(state) = container.and_then(|c| c.state.as_ref()) { + if state.running.is_some() { + return Resolved(Startup::Started); + } + if state.terminated.is_some() { + return Resolved(Startup::Terminated); + } + if let Some(waiting) = state.waiting.as_ref() { + let reason = waiting.reason.as_deref().unwrap_or_default(); + let message = waiting.message.as_deref().unwrap_or_default(); + return match reason { + // Structurally invalid reference: no retry can fix it. + "InvalidImageName" => Resolved(Startup::NeverStartedProvablyBroken), + "ErrImagePull" | "ImagePullBackOff" => match classify_pull_failure(message) { + Some(failure) => Resolved(Startup::NeverStartedPullFailing(failure)), + None => Resolved(Startup::NeverStartedRecoverable), + }, + "CreateContainerConfigError" => match referenced_secret(pod) { + Some(secret_name) => { + StartupObservation::ConfigErrorPendingSecretCheck { secret_name } + } + None => Resolved(Startup::NeverStartedRecoverable), + }, + _ => Resolved(Startup::NeverStartedRecoverable), + }; + } + } + + // No container status yet (unscheduled, image pulling before the kubelet + // reports, quota-blocked). A terminal phase without container status still + // means the pod is done. + match phase { + Some("Succeeded") | Some("Failed") => Resolved(Startup::Terminated), + _ => Resolved(Startup::NeverStartedRecoverable), + } +} + +/// The Secret name this pod's `envFrom` references, if any. +pub fn referenced_secret(pod: &Pod) -> Option { + pod.spec + .as_ref()? + .containers + .iter() + .flat_map(|c| c.env_from.iter().flatten()) + .find_map(|source| source.secret_ref.as_ref().map(|r| r.name.clone())) +} + +/// Classify a pull failure from the kubelet's message. +/// +/// Reporting only — [`PullFailure`] is structurally excluded from +/// `Action::Delete`, so a wrong guess here can delay a report but can never +/// destroy anything. `None` means "no permanent cause recognized", which +/// leaves the pod on the ordinary observational path. +fn classify_pull_failure(message: &str) -> Option { + let m = message.to_ascii_lowercase(); + if m.contains("401") + || m.contains("unauthorized") + || m.contains("403") + || m.contains("denied") + || m.contains("authentication required") + { + return Some(PullFailure::Unauthorized); + } + if m.contains("manifest unknown") + || m.contains("not found") + || m.contains("manifest_unknown") + || m.contains("repository does not exist") + { + return Some(PullFailure::ManifestUnknown); + } + if m.contains("no match for platform") || m.contains("no matching manifest") { + return Some(PullFailure::ArchMismatch); + } + None +} + +/// The redacted, actionable condition text for a pull failure. +/// +/// Names the registry and the immutable reference — never credentials, and +/// never the kubelet's raw message, which can echo a registry token. +pub fn pull_failure_message(failure: PullFailure, image: &str) -> String { + let registry = image.split('/').next().unwrap_or(image); + match failure { + PullFailure::Unauthorized => format!( + "the cluster is not authorized to pull {image} from {registry}. \ + This pull retries indefinitely and will not succeed on its own: \ + grant the cluster's nodes access to that registry." + ), + PullFailure::ManifestUnknown => { + format!("{registry} has no image at {image}. Check the digest and repository.") + } + PullFailure::ArchMismatch => format!( + "{image} has no variant for the architecture of the nodes it was \ + scheduled on." + ), + } +} + +/// The latest actionable condition for a pod that has not started, redacted. +/// +/// Two sources, deliberately treated differently: +/// +/// * The container's waiting **reason** is included; its **message** is not. +/// Waiting messages are kubelet-composed and echo the thing that failed — +/// for a pull that is the registry request, which can carry credential +/// material. The reason token alone (`ImagePullBackOff`, +/// `CreateContainerConfigError`) is the diagnostic; the message adds +/// exposure, not information the user can act on. +/// * Pod-condition messages **are** included. They are scheduler- and +/// kubelet-composed from the pod's own spec and cluster capacity +/// ("0/3 nodes are available: Insufficient memory"), which is precisely the +/// actionable half and contains nothing derived from Secret data. +pub fn condition(pod: &Pod) -> Option { + let status = pod.status.as_ref()?; + + if let Some(state) = status + .container_statuses + .as_ref() + .and_then(|cs| cs.iter().find(|c| c.name == CONTAINER_NAME)) + .and_then(|c| c.state.as_ref()) + { + if let Some(waiting) = state.waiting.as_ref() { + if let Some(reason) = waiting.reason.as_deref() { + return Some(format!("the container is waiting, reason {reason}")); + } + } + // Exit code and reason only — the terminated `message` is + // process-composed output and falls under the same redaction rule as + // waiting messages. + if let Some(terminated) = state.terminated.as_ref() { + return Some(match terminated.reason.as_deref() { + Some(reason) => format!( + "the container exited with code {} ({reason})", + terminated.exit_code + ), + None => format!("the container exited with code {}", terminated.exit_code), + }); + } + } + + if let Some((type_, reason, message)) = status.conditions.as_ref().and_then(|cs| { + cs.iter().find(|c| c.status == "False").map(|c| { + ( + c.type_.clone(), + c.reason.clone().unwrap_or_default(), + c.message.clone().unwrap_or_default(), + ) + }) + }) { + let detail = [reason, message] + .into_iter() + .filter(|s| !s.is_empty()) + .collect::>() + .join(": "); + return Some(if detail.is_empty() { + format!("pod condition {type_} is false") + } else { + format!("pod condition {type_} is false: {detail}") + }); + } + + status.phase.as_deref().map(|p| format!("the pod is {p}")) +} + +/// Verify a label-selected pod and decode it, or reject it. +/// +/// `startup` is supplied by the caller because settling +/// `CreateContainerConfigError` needs a most-recent Secret read the pure layer +/// must not perform. +pub fn verify(pod: &Pod, identity: &AgentIdentity, startup: Startup) -> Option { + if !has_marker(pod.metadata.labels.as_ref()) { + return None; + } + if !annotation_matches(pod.metadata.annotations.as_ref(), identity) { + return None; + } + Some(VerifiedPod { + name: pod.metadata.name.clone()?, + fence: Fence { + uid: pod.metadata.uid.clone()?, + resource_version: pod.metadata.resource_version.clone()?, + }, + deletion_marked: pod.metadata.deletion_timestamp.is_some(), + startup, + recorded_intent: pod + .metadata + .annotations + .as_ref() + .and_then(|a| a.get(ANNOTATION_CREATE_INTENT)) + .map(|v| Fingerprint::from_annotation(v)), + }) +} + +#[cfg(test)] +mod tests { + use super::*; + use k8s_openapi::api::core::v1::{ + ContainerState, ContainerStateRunning, ContainerStateTerminated, ContainerStateWaiting, + ContainerStatus, PodStatus, + }; + use k8s_openapi::apimachinery::pkg::apis::meta::v1::{ObjectMeta, Time}; + use std::collections::BTreeMap; + + fn identity() -> AgentIdentity { + use nostr::nips::nip19::ToBech32; + let keys = nostr::Keys::generate(); + AgentIdentity::from_nsec(&keys.secret_key().to_bech32().unwrap()).unwrap() + } + + fn base_pod(id: &AgentIdentity) -> Pod { + Pod { + metadata: ObjectMeta { + name: Some(id.pod_name()), + uid: Some("uid-1".into()), + resource_version: Some("100".into()), + labels: Some(id.labels()), + annotations: Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + id.pubkey_hex().to_string(), + )] + .into_iter() + .collect::>(), + ), + ..Default::default() + }, + ..Default::default() + } + } + + fn with_container_state(mut pod: Pod, state: ContainerState) -> Pod { + pod.status = Some(PodStatus { + phase: Some("Running".into()), + container_statuses: Some(vec![ContainerStatus { + name: CONTAINER_NAME.into(), + state: Some(state), + ..Default::default() + }]), + ..Default::default() + }); + pod + } + + fn waiting(reason: &str, message: &str) -> ContainerState { + ContainerState { + waiting: Some(ContainerStateWaiting { + reason: Some(reason.into()), + message: Some(message.into()), + }), + ..Default::default() + } + } + + #[test] + fn running_container_is_started() { + let id = identity(); + let pod = with_container_state( + base_pod(&id), + ContainerState { + running: Some(ContainerStateRunning::default()), + ..Default::default() + }, + ); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::Started) + ); + } + + /// Pod phase is not the criterion. A pod in phase `Running` whose + /// container never started must NOT read as started, or the reconciler + /// no-ops on a pod that will never serve. + #[test] + fn phase_running_with_waiting_container_is_not_started() { + let id = identity(); + let pod = with_container_state(base_pod(&id), waiting("ContainerCreating", "")); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::NeverStartedRecoverable) + ); + } + + #[test] + fn terminated_container_is_terminated() { + let id = identity(); + let pod = with_container_state( + base_pod(&id), + ContainerState { + terminated: Some(ContainerStateTerminated { + exit_code: 0, + ..Default::default() + }), + ..Default::default() + }, + ); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::Terminated) + ); + } + + /// A terminal phase with no container status (evicted before the kubelet + /// reported) is still terminated — otherwise the residue is never GC'd. + #[test] + fn terminal_phase_without_container_status_is_terminated() { + let id = identity(); + for phase in ["Succeeded", "Failed"] { + let mut pod = base_pod(&id); + pod.status = Some(PodStatus { + phase: Some(phase.into()), + ..Default::default() + }); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::Terminated), + "phase {phase}" + ); + } + } + + #[test] + fn invalid_image_name_is_provably_broken() { + let id = identity(); + let pod = with_container_state(base_pod(&id), waiting("InvalidImageName", "bad ref")); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::NeverStartedProvablyBroken) + ); + } + + /// Permanent pull failures are recognized from the message; anything + /// unrecognized stays on the ordinary observational path rather than + /// being guessed at. + #[test] + fn permanent_pull_failures_are_classified() { + let id = identity(); + let cases = [ + ("401 Unauthorized", PullFailure::Unauthorized), + ("pull access denied", PullFailure::Unauthorized), + ("manifest unknown", PullFailure::ManifestUnknown), + ( + "no match for platform in manifest", + PullFailure::ArchMismatch, + ), + ]; + for (message, expected) in cases { + let pod = with_container_state(base_pod(&id), waiting("ErrImagePull", message)); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::NeverStartedPullFailing(expected)), + "message {message:?}" + ); + } + + let pod = with_container_state( + base_pod(&id), + waiting("ImagePullBackOff", "dial tcp: i/o timeout"), + ); + assert_eq!( + decode_startup(&pod), + StartupObservation::Resolved(Startup::NeverStartedRecoverable), + "a transient network failure must not be reported as permanent" + ); + } + + /// The kubelet's reason string is a hint, not proof: a config error defers + /// to a most-recent Secret read before anything is called broken. + #[test] + fn config_error_defers_to_a_secret_read() { + let id = identity(); + let mut pod = + with_container_state(base_pod(&id), waiting("CreateContainerConfigError", "")); + pod.spec = Some(k8s_openapi::api::core::v1::PodSpec { + containers: vec![k8s_openapi::api::core::v1::Container { + name: CONTAINER_NAME.into(), + env_from: Some(vec![k8s_openapi::api::core::v1::EnvFromSource { + secret_ref: Some(k8s_openapi::api::core::v1::SecretEnvSource { + name: "buzz-agent-abc-gen1".into(), + optional: Some(false), + }), + ..Default::default() + }]), + ..Default::default() + }], + ..Default::default() + }); + assert_eq!( + decode_startup(&pod), + StartupObservation::ConfigErrorPendingSecretCheck { + secret_name: "buzz-agent-abc-gen1".into() + } + ); + } + + /// The auto-repair fence: an object that matches our schema but lacks the + /// marker, or carries someone else's pubkey, is never verified — so it can + /// never be no-op'd against, deleted, or returned as an `agent_id`. + #[test] + fn unmarked_or_mismatched_objects_fail_verification() { + let id = identity(); + let other = identity(); + + let mut unmarked = base_pod(&id); + unmarked.metadata.labels = Some(BTreeMap::new()); + assert!( + verify(&unmarked, &id, Startup::Started).is_none(), + "unmarked pod verified" + ); + + let mut wrong_version = base_pod(&id); + let mut labels = id.labels(); + labels.insert(LABEL_BINDING_VERSION.to_string(), "999".to_string()); + wrong_version.metadata.labels = Some(labels); + assert!(verify(&wrong_version, &id, Startup::Started).is_none()); + + let mut mismatched = base_pod(&id); + mismatched.metadata.annotations = Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + other.pubkey_hex().to_string(), + )] + .into_iter() + .collect(), + ); + assert!( + verify(&mismatched, &id, Startup::Started).is_none(), + "collision verified" + ); + + let mut missing = base_pod(&id); + missing.metadata.annotations = Some(BTreeMap::new()); + assert!(verify(&missing, &id, Startup::Started).is_none()); + + assert!( + verify(&base_pod(&id), &id, Startup::Started).is_some(), + "own pod rejected" + ); + } + + /// The fence must come from the observed object, and the deletion mark + /// must be read even though the phase says `Running`. + #[test] + fn verified_pod_carries_the_fence_and_deletion_mark() { + let id = identity(); + let mut pod = base_pod(&id); + pod.metadata.deletion_timestamp = Some(Time(chrono::Utc::now())); + let verified = verify(&pod, &id, Startup::Started).unwrap(); + assert_eq!(verified.fence.uid, "uid-1"); + assert_eq!(verified.fence.resource_version, "100"); + assert!(verified.deletion_marked); + } + + /// A pod with no recorded intent reads as `None`, which the classifier + /// groups with divergence. + #[test] + fn missing_intent_annotation_decodes_as_none() { + let id = identity(); + assert!(verify(&base_pod(&id), &id, Startup::Started) + .unwrap() + .recorded_intent + .is_none()); + } + + /// A pull-failure report must name the registry and the immutable + /// reference and nothing else — never the kubelet's raw message, which + /// can echo a registry token. + #[test] + fn pull_failure_messages_are_actionable_and_redacted() { + let image = format!("ghcr.io/block/buzz-sprig@sha256:{}", "a".repeat(64)); + for failure in [ + PullFailure::Unauthorized, + PullFailure::ManifestUnknown, + PullFailure::ArchMismatch, + ] { + let msg = pull_failure_message(failure, &image); + assert!(msg.contains("ghcr.io"), "{msg}"); + assert!(msg.contains(&image), "{msg}"); + for secret in ["Bearer", "password", "nsec1", "token"] { + assert!(!msg.contains(secret), "leaked {secret}: {msg}"); + } + } + } + + #[test] + fn secret_ownership_requires_marker_and_annotation() { + let id = identity(); + let other = identity(); + let ours = Secret { + metadata: ObjectMeta { + labels: Some(id.labels()), + annotations: Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + id.pubkey_hex().to_string(), + )] + .into(), + ), + ..Default::default() + }, + ..Default::default() + }; + assert!(secret_is_ours(&ours, &id)); + assert!(!secret_is_ours(&ours, &other)); + + let mut unmarked = ours.clone(); + unmarked.metadata.labels = Some(BTreeMap::new()); + assert!(!secret_is_ours(&unmarked, &id)); + } +} diff --git a/crates/buzz-backend-kubernetes/src/pod.rs b/crates/buzz-backend-kubernetes/src/pod.rs new file mode 100644 index 0000000000..725f98fc7d --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/pod.rs @@ -0,0 +1,446 @@ +//! Pod and Secret construction (spec §Pod shape, §K8s Secrets). +//! +//! The builder is pure: it turns resolved inputs into API objects and performs +//! no I/O, so every normative field is a unit assertion. + +use crate::config::{ + ProviderConfig, RESTART_POLICY, RUN_AS_GID, RUN_AS_UID, TERMINATION_GRACE_SECONDS, + WORKSPACE_PATH, +}; +use crate::intent::{Fingerprint, IntentTemplate}; +use crate::naming::{ + AgentIdentity, ANNOTATION_CREATE_INTENT, ANNOTATION_IMAGE, ANNOTATION_PUBKEY_FULL, +}; +use k8s_openapi::api::core::v1::{ + Capabilities, Container, EmptyDirVolumeSource, EnvFromSource, Pod, PodSecurityContext, PodSpec, + ResourceRequirements, SeccompProfile, Secret, SecretEnvSource, SecurityContext, Volume, + VolumeMount, +}; +use k8s_openapi::apimachinery::pkg::api::resource::Quantity; +use k8s_openapi::apimachinery::pkg::apis::meta::v1::ObjectMeta; +use std::collections::BTreeMap; + +/// Volume name for the agent's writable workspace. +const WORKSPACE_VOLUME: &str = "workspace"; + +/// The container name. Fixed: log and exec tooling addresses it by name. +const CONTAINER_NAME: &str = "agent"; + +/// Build the per-attempt Secret holding the resolved environment. +/// +/// `immutable: true` — the Secret is written once per attempt and never +/// updated, which is what lets the pod's `envFrom` reference be treated as an +/// atomic binding to this exact payload (§K8s Secrets). +pub fn build_secret( + identity: &AgentIdentity, + namespace: &str, + generation: &str, + env: BTreeMap, +) -> Secret { + Secret { + metadata: ObjectMeta { + name: Some(identity.secret_name(generation)), + namespace: Some(namespace.to_string()), + labels: Some(identity.labels()), + annotations: Some( + [( + ANNOTATION_PUBKEY_FULL.to_string(), + identity.pubkey_hex().to_string(), + )] + .into_iter() + .collect(), + ), + ..Default::default() + }, + string_data: Some(env), + immutable: Some(true), + ..Default::default() + } +} + +/// Build the pod for one create attempt. +/// +/// The `fingerprint` is computed from [`intent_template`] over a type that +/// cannot contain the generation or any Secret value, so it is stable across +/// attempts of the same configuration. +pub fn build_pod( + identity: &AgentIdentity, + cfg: &ProviderConfig, + generation: &str, + fingerprint: &Fingerprint, +) -> Pod { + let annotations: BTreeMap = [ + ( + ANNOTATION_PUBKEY_FULL.to_string(), + identity.pubkey_hex().to_string(), + ), + ( + ANNOTATION_CREATE_INTENT.to_string(), + fingerprint.as_str().to_string(), + ), + (ANNOTATION_IMAGE.to_string(), cfg.image.as_str().to_string()), + ] + .into_iter() + .collect(); + + let requests: BTreeMap = [ + ( + "cpu".to_string(), + Quantity(cfg.resources.cpu_request.clone()), + ), + ( + "memory".to_string(), + Quantity(cfg.resources.memory_request.clone()), + ), + ] + .into_iter() + .collect(); + let limits: BTreeMap = [ + ("cpu".to_string(), Quantity(cfg.resources.cpu_limit.clone())), + ( + "memory".to_string(), + Quantity(cfg.resources.memory_limit.clone()), + ), + ] + .into_iter() + .collect(); + + let container = Container { + name: CONTAINER_NAME.to_string(), + image: Some(cfg.image.as_str().to_string()), + // No `command`/`args`: the image's entrypoint execs the harness as + // PID 1 (§Entrypoint). Overriding it here would be how a provider + // accidentally puts a shell in front of the signal receiver. + env_from: Some(vec![EnvFromSource { + secret_ref: Some(SecretEnvSource { + name: identity.secret_name(generation), + optional: Some(false), + }), + ..Default::default() + }]), + resources: Some(ResourceRequirements { + requests: Some(requests), + limits: Some(limits), + ..Default::default() + }), + volume_mounts: Some(vec![VolumeMount { + name: WORKSPACE_VOLUME.to_string(), + mount_path: WORKSPACE_PATH.to_string(), + ..Default::default() + }]), + security_context: Some(SecurityContext { + allow_privilege_escalation: Some(false), + capabilities: Some(Capabilities { + drop: Some(vec!["ALL".to_string()]), + ..Default::default() + }), + // `readOnlyRootFilesystem` is deliberately unset: the sprig + // toolchain writes outside the workspace mount (§Pod shape). + ..Default::default() + }), + ..Default::default() + }; + + Pod { + metadata: ObjectMeta { + name: Some(identity.pod_name()), + namespace: Some(cfg.namespace.clone()), + labels: Some(identity.labels()), + annotations: Some(annotations), + ..Default::default() + }, + spec: Some(PodSpec { + containers: vec![container], + restart_policy: Some(RESTART_POLICY.to_string()), + termination_grace_period_seconds: Some(TERMINATION_GRACE_SECONDS), + // The agent runs prompted, untrusted code while holding an nsec; + // an ambient ServiceAccount token would be an API-stealable + // credential it never needs (§Pod shape hardening). Naming a + // service account selects a scheduling/RBAC identity and MUST NOT + // re-enable token mounting. + automount_service_account_token: Some(false), + service_account_name: cfg.service_account.clone(), + security_context: Some(PodSecurityContext { + run_as_non_root: Some(true), + run_as_user: Some(RUN_AS_UID), + run_as_group: Some(RUN_AS_GID), + fs_group: Some(RUN_AS_GID), + seccomp_profile: Some(SeccompProfile { + type_: "RuntimeDefault".to_string(), + ..Default::default() + }), + ..Default::default() + }), + volumes: Some(vec![Volume { + name: WORKSPACE_VOLUME.to_string(), + empty_dir: Some(EmptyDirVolumeSource::default()), + ..Default::default() + }]), + ..Default::default() + }), + ..Default::default() + } +} + +/// The create-intent template for this configuration (§Deploy State Machine). +pub fn intent_template( + cfg: &ProviderConfig, + env_keys: impl IntoIterator, +) -> IntentTemplate { + IntentTemplate::new( + &cfg.namespace, + &cfg.image, + &cfg.resources, + cfg.service_account.as_deref(), + env_keys, + ) +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::config; + + fn identity() -> AgentIdentity { + use nostr::nips::nip19::ToBech32; + let keys = nostr::Keys::generate(); + AgentIdentity::from_nsec(&keys.secret_key().to_bech32().unwrap()).unwrap() + } + + fn provider_config() -> ProviderConfig { + config::parse(&serde_json::json!({ + "namespace": "buzz-agents-test", + "image": format!("ghcr.io/block/buzz-sprig@sha256:{}", "a".repeat(64)), + })) + .unwrap() + } + + fn pod() -> Pod { + let cfg = provider_config(); + build_pod( + &identity(), + &cfg, + "gen00001", + &intent_template(&cfg, ["BUZZ_RELAY_URL".to_string()]).fingerprint(), + ) + } + + fn spec(pod: &Pod) -> &PodSpec { + pod.spec.as_ref().unwrap() + } + + /// Every hardening default from §Pod shape, asserted individually so a + /// dropped one names itself. + #[test] + fn hardening_defaults_are_all_present() { + let pod = pod(); + let spec = spec(&pod); + assert_eq!(spec.automount_service_account_token, Some(false)); + + let sc = spec + .security_context + .as_ref() + .expect("pod security context"); + assert_eq!(sc.run_as_non_root, Some(true)); + assert_eq!(sc.run_as_user, Some(RUN_AS_UID)); + assert_ne!(sc.run_as_user, Some(0), "root UID"); + assert_eq!(sc.run_as_group, Some(RUN_AS_GID)); + assert_eq!( + sc.seccomp_profile.as_ref().map(|p| p.type_.as_str()), + Some("RuntimeDefault") + ); + + let csc = spec.containers[0] + .security_context + .as_ref() + .expect("container sc"); + assert_eq!(csc.allow_privilege_escalation, Some(false)); + assert_eq!( + csc.capabilities.as_ref().and_then(|c| c.drop.clone()), + Some(vec!["ALL".to_string()]) + ); + assert_ne!(csc.privileged, Some(true)); + } + + /// The forbidden host-namespace and hostPath escapes, asserted as absence. + #[test] + fn never_uses_host_namespaces_or_host_paths() { + let pod = pod(); + let spec = spec(&pod); + assert!(spec.host_pid.is_none() || spec.host_pid == Some(false)); + assert!(spec.host_network.is_none() || spec.host_network == Some(false)); + assert!(spec.host_ipc.is_none() || spec.host_ipc == Some(false)); + for volume in spec.volumes.as_ref().unwrap() { + assert!( + volume.host_path.is_none(), + "hostPath volume {}", + volume.name + ); + } + } + + /// `Never` only. `OnFailure` is gated on the harness exit-code contract + /// *and* a crash-loop classification row (`:1121-1139`); the config layer + /// refuses `inactivity_seconds: 0` so this arm is unreachable, and the + /// assertion keeps it that way. + #[test] + fn restart_policy_is_never() { + assert_eq!(spec(&pod()).restart_policy.as_deref(), Some("Never")); + } + + /// 60s, not Kubernetes' default 30s — which would SIGKILL the harness + /// mid-drain and leave presence stale-online (§Pod shape). + #[test] + fn declares_the_sixty_second_grace_budget() { + assert_eq!(spec(&pod()).termination_grace_period_seconds, Some(60)); + } + + /// The pod must not override the image's entrypoint: the image execs the + /// harness as PID 1, and a `command` here is how a shell ends up in front + /// of the signal receiver (§Entrypoint). + #[test] + fn does_not_override_the_image_entrypoint() { + let pod = pod(); + let container = &spec(&pod).containers[0]; + assert!(container.command.is_none(), "overrode the entrypoint"); + assert!(container.args.is_none()); + } + + #[test] + fn workspace_is_an_emptydir_mounted_at_home() { + let pod = pod(); + let spec = spec(&pod); + let volume = &spec.volumes.as_ref().unwrap()[0]; + assert!(volume.empty_dir.is_some()); + assert!(volume.persistent_volume_claim.is_none()); + let mount = &spec.containers[0].volume_mounts.as_ref().unwrap()[0]; + assert_eq!(mount.name, volume.name); + assert_eq!(mount.mount_path, WORKSPACE_PATH); + } + + #[test] + fn resources_carry_the_configured_requests_and_limits() { + let mut cfg = provider_config(); + cfg.resources.cpu_limit = "4".into(); + let pod = build_pod(&identity(), &cfg, "g", &Fingerprint::from_annotation("f")); + let r = spec(&pod).containers[0].resources.as_ref().unwrap(); + assert_eq!(r.requests.as_ref().unwrap()["cpu"], Quantity("1".into())); + assert_eq!( + r.requests.as_ref().unwrap()["memory"], + Quantity("2Gi".into()) + ); + assert_eq!(r.limits.as_ref().unwrap()["cpu"], Quantity("4".into())); + assert_eq!(r.limits.as_ref().unwrap()["memory"], Quantity("4Gi".into())); + } + + /// `envFrom` must point at this attempt's Secret and must NOT be optional: + /// an optional reference starts the container with no identity at all, + /// turning a missing-Secret bug into an agent that silently cannot + /// authenticate. + #[test] + fn env_from_references_this_attempts_secret_and_is_required() { + let id = identity(); + let cfg = provider_config(); + let pod = build_pod(&id, &cfg, "gen00042", &Fingerprint::from_annotation("f")); + let source = &spec(&pod).containers[0].env_from.as_ref().unwrap()[0]; + let secret_ref = source.secret_ref.as_ref().unwrap(); + assert_eq!(secret_ref.name, id.secret_name("gen00042")); + assert_eq!(secret_ref.optional, Some(false)); + assert!(source.config_map_ref.is_none()); + } + + /// Identity, ownership marker, and the recorded intent all travel on the + /// pod — the GC and reconciliation fences read exactly these. + #[test] + fn pod_carries_identity_marker_and_recorded_intent() { + let id = identity(); + let cfg = provider_config(); + let fp = intent_template(&cfg, ["A".to_string()]).fingerprint(); + let pod = build_pod(&id, &cfg, "g", &fp); + let meta = &pod.metadata; + assert_eq!(meta.name.as_deref(), Some(id.pod_name().as_str())); + assert_eq!(meta.namespace.as_deref(), Some("buzz-agents-test")); + assert_eq!(meta.labels.as_ref().unwrap(), &id.labels()); + let ann = meta.annotations.as_ref().unwrap(); + assert_eq!(ann[ANNOTATION_PUBKEY_FULL], id.pubkey_hex()); + assert_eq!(ann[ANNOTATION_CREATE_INTENT], fp.as_str()); + assert_eq!(ann[ANNOTATION_IMAGE], cfg.image.as_str()); + } + + /// The Secret is immutable and marker-bearing: immutability is what makes + /// the pod's `envFrom` an atomic binding, and the marker is what GC + /// requires before it will delete anything. + #[test] + fn secret_is_immutable_marked_and_holds_the_env() { + let id = identity(); + let env: BTreeMap = + [("BUZZ_RELAY_URL".to_string(), "wss://r".to_string())].into(); + let secret = build_secret(&id, "ns", "gen1", env.clone()); + assert_eq!(secret.immutable, Some(true)); + assert_eq!(secret.string_data.as_ref().unwrap(), &env); + assert_eq!( + secret.metadata.name.as_deref(), + Some(id.secret_name("gen1").as_str()) + ); + assert_eq!(secret.metadata.labels.as_ref().unwrap(), &id.labels()); + assert_eq!( + secret.metadata.annotations.as_ref().unwrap()[ANNOTATION_PUBKEY_FULL], + id.pubkey_hex() + ); + // `data` must stay unset — setting both is an apiserver rejection. + assert!(secret.data.is_none()); + } + + /// Naming a service account selects a scheduling identity; it must not + /// re-enable token mounting (§Pod shape hardening, `:1221-1225`). + #[test] + fn service_account_does_not_re_enable_token_mounting() { + let mut cfg = provider_config(); + cfg.service_account = Some("agent-sa".into()); + let pod = build_pod(&identity(), &cfg, "g", &Fingerprint::from_annotation("f")); + assert_eq!(spec(&pod).service_account_name.as_deref(), Some("agent-sa")); + assert_eq!(spec(&pod).automount_service_account_token, Some(false)); + } + + /// The fingerprint recorded on the pod is the one the classifier will + /// recompute — pinned end-to-end so a builder change that forgets to feed + /// the template a field cannot pass silently. + #[test] + fn recorded_fingerprint_matches_a_fresh_computation() { + let cfg = provider_config(); + let keys = ["BUZZ_RELAY_URL".to_string(), "GOOSE_MODE".to_string()]; + let fp = intent_template(&cfg, keys.clone()).fingerprint(); + let pod = build_pod(&identity(), &cfg, "gen-a", &fp); + let recorded = Fingerprint::from_annotation( + &pod.metadata.annotations.as_ref().unwrap()[ANNOTATION_CREATE_INTENT], + ); + assert_eq!(recorded, intent_template(&cfg, keys).fingerprint()); + } + + /// Two attempts differing only in generation must record the *same* + /// fingerprint, or the divergence discriminator fires on every deploy and + /// the never-started row deletes healthy pending pods. + #[test] + fn generation_does_not_change_the_recorded_fingerprint() { + let cfg = provider_config(); + let keys = ["BUZZ_RELAY_URL".to_string()]; + let a = intent_template(&cfg, keys.clone()).fingerprint(); + let b = intent_template(&cfg, keys).fingerprint(); + let id = identity(); + let pod_a = build_pod(&id, &cfg, "gen-1", &a); + let pod_b = build_pod(&id, &cfg, "gen-2", &b); + let read = + |p: &Pod| p.metadata.annotations.as_ref().unwrap()[ANNOTATION_CREATE_INTENT].clone(); + assert_eq!(read(&pod_a), read(&pod_b)); + // ...while the Secret they reference differs. + let secret_of = |p: &Pod| { + spec(p).containers[0].env_from.as_ref().unwrap()[0] + .secret_ref + .as_ref() + .unwrap() + .name + .clone() + }; + assert_ne!(secret_of(&pod_a), secret_of(&pod_b)); + } +} diff --git a/crates/buzz-backend-kubernetes/src/reconcile.rs b/crates/buzz-backend-kubernetes/src/reconcile.rs new file mode 100644 index 0000000000..df2f99789f --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/reconcile.rs @@ -0,0 +1,1585 @@ +//! The deploy loop: executes [`crate::classify`]'s actions against a substrate +//! and re-enters (spec §Deploy State Machine). +//! +//! The substrate is a trait so the conformance tests drive this exact +//! reconciler — the shipped code path, not a test-only reimplementation — with +//! a fake cluster and a fake clock. +//! +//! Two shapes are worth naming up front, because they are what keep the loop +//! terminating: +//! +//! * **Success means the harness container started** (`:696-699`). There is no +//! "deployed but not confirmed" success: `deploy` returns an `agent_id` or an +//! in-band error carrying the latest condition. The wire has no third form. +//! * **A create-conflict loser never repairs.** It verifies the winner, drops +//! its own Secret, and *observes* until the winner starts. Applying the +//! divergence row to the pod that just beat it is exactly the ping-pong the +//! spec forbids (`:845-850`), and the escape it names is the *next* deploy, +//! not this one. + +use crate::classify::{self, Action, Fence, Startup, VerifiedPod}; +use crate::config::ProviderConfig; +use crate::gc; +use crate::naming::AgentIdentity; +use crate::observe::{self, StartupObservation}; +use chrono::{DateTime, Utc}; +use k8s_openapi::api::core::v1::{Pod, Secret}; +use std::collections::BTreeMap; +use std::time::Duration; + +/// Outcome of a create against the deterministic pod name. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum CreateOutcome { + Created, + /// 409 with `Status.reason: AlreadyExists` — a concurrent attempt won. + /// Discriminated on the typed reason, never on the HTTP code alone: 409 is + /// also `Conflict`, which means a failed precondition (`:780-794`). + AlreadyExists, +} + +/// Outcome of a fenced delete. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum DeleteOutcome { + Accepted, + /// Already gone. Delete-not-found is success (`:842`). + NotFound, + /// 409 with `Status.reason: Conflict` — the object changed since the + /// observation that authorized this delete. Neither an error nor + /// permission to retry: re-enter and classify what exists now + /// (`:775-777`). + PreconditionFailed, +} + +/// The cluster operations the reconciler needs. Everything here is I/O; +/// everything that decides is pure and lives in `classify`/`gc`. +#[allow(async_fn_in_trait)] +pub trait Substrate { + /// Create the namespace if absent. On RBAC denial the error MUST name the + /// literal `kubectl create namespace ` command and MUST NOT fall + /// back to `default` (`:1002-1005`). + async fn ensure_namespace(&self, namespace: &str) -> Result<(), String>; + + /// Most-recent read of the pods matching this identity's selector, plus + /// the apiserver's clock from the same call's HTTP `Date` header. `None` + /// clock means the header was absent or unparseable, which makes the + /// orphan-Secret sweep skip (`:1321-1335`). + async fn list_pods(&self, selector: &str) -> Result<(Vec, Option>), String>; + + async fn list_secrets(&self, selector: &str) -> Result, String>; + + /// Most-recent existence check (`resourceVersion` explicitly unset, not + /// `"0"`): the classifier treats a confirmed absence as proof, so a + /// possibly-stale cache read would be proof of nothing (`:761-769`). + async fn secret_exists(&self, name: &str) -> Result; + + async fn create_secret(&self, secret: &Secret) -> Result<(), String>; + + async fn create_pod(&self, pod: &Pod) -> Result; + + /// Compare-and-delete against the fence from the authorizing observation. + /// Uses the object's own grace period — never `grace_period_seconds: 0`, + /// which is a force-kill that discards the declared 60s shutdown budget + /// (`:1185-1189`). + async fn delete_pod(&self, name: &str, fence: &Fence) -> Result; + + /// Best-effort: the Secret may already be gone, which is success. + async fn delete_secret(&self, name: &str) -> Result<(), String>; + + /// Read one pod by name, most-recent. `None` is a confirmed absence. + async fn get_pod(&self, name: &str) -> Result, String>; + + async fn sleep(&self, duration: Duration); + + /// Monotonic elapsed time since the operation began. Fake-clock driven in + /// tests; the deadline must not depend on wall-clock adjustments. + fn elapsed(&self) -> Duration; +} + +/// Interval between reconciler polls. Short enough that a fast start is +/// reported promptly, long enough not to hammer the apiserver for 600s. +const POLL_INTERVAL: Duration = Duration::from_secs(2); + +/// The deploy operation deadline (spec §Deploy: `timeout: 600s`). +const DEADLINE: Duration = Duration::from_secs(gc::OPERATION_DEADLINE_SECS as u64); + +/// Settle a pod's startup state, performing the one most-recent Secret read +/// that `CreateContainerConfigError` requires. +async fn settle(substrate: &impl Substrate, pod: &Pod) -> Result { + Ok(match observe::decode_startup(pod) { + StartupObservation::Resolved(startup) => startup, + StartupObservation::ConfigErrorPendingSecretCheck { secret_name } => { + // A confirmed absence is proof; anything else stays recoverable, + // because the kubelet's reason string alone is not evidence. + if substrate.secret_exists(&secret_name).await? { + Startup::NeverStartedRecoverable + } else { + Startup::NeverStartedProvablyBroken + } + } + }) +} + +/// Observe the single pod owned by this identity, verified. +/// +/// `Ok(None)` conflates "absent" with "present but not ours" on purpose *here* +/// — both mean the classifier has nothing it may act on. The create path +/// separates them, because only there is the difference actionable. +async fn observe_pod( + substrate: &impl Substrate, + identity: &AgentIdentity, +) -> Result, String> { + let Some(pod) = substrate.get_pod(&identity.pod_name()).await? else { + return Ok(None); + }; + let startup = settle(substrate, &pod).await?; + Ok(observe::verify(&pod, identity, startup)) +} + +/// The latest condition to report, read fresh at the moment of reporting. +/// +/// Reading it here rather than threading it through every loop iteration is +/// what "the *latest* redacted condition" (`:689`) asks for, and it costs a +/// read only on the paths that are already failing. +async fn latest_condition(substrate: &impl Substrate, identity: &AgentIdentity) -> String { + match substrate.get_pod(&identity.pod_name()).await { + Ok(Some(pod)) => observe::condition(&pod) + .unwrap_or_else(|| "no condition reported by the cluster".to_string()), + Ok(None) => "the pod no longer exists".to_string(), + Err(e) => format!("the pod's condition could not be read: {e}"), + } +} + +/// Preflight GC (§K8s GC). Failures are logged and swallowed: GC is hygiene, +/// and a deploy must not fail because a stale object could not be listed or +/// removed. A denial that actually blocks this deploy resurfaces at create, +/// where the message names the operation the user was denied. +async fn preflight_gc(substrate: &impl Substrate, identity: &AgentIdentity) { + if let Err(e) = try_preflight_gc(substrate, identity).await { + eprintln!("gc: preflight pass skipped: {e}"); + } +} + +async fn try_preflight_gc( + substrate: &impl Substrate, + identity: &AgentIdentity, +) -> Result<(), String> { + let selector = identity.selector(); + let (pods, server_now) = substrate.list_pods(&selector).await?; + let secrets = substrate.list_secrets(&selector).await?; + + let mut terminated: Vec = Vec::new(); + for pod in &pods { + if matches!(settle(substrate, pod).await?, Startup::Terminated) { + if let Some(name) = pod.metadata.name.clone() { + terminated.push(name); + } + } + } + + let plan = gc::plan( + identity, + &pods, + &secrets, + |pod| { + pod.metadata + .name + .as_deref() + .map(|n| terminated.iter().any(|t| t == n)) + .unwrap_or(false) + }, + server_now, + ); + + for name in &plan.pods { + // Re-read to fence the delete against the object we just observed; a + // pod that changed since the list is simply skipped this pass. + let Some(pod) = substrate.get_pod(name).await? else { + continue; + }; + let (Some(uid), Some(rv)) = ( + pod.metadata.uid.clone(), + pod.metadata.resource_version.clone(), + ) else { + continue; + }; + if !matches!(settle(substrate, &pod).await?, Startup::Terminated) { + continue; + } + let fence = Fence { + uid, + resource_version: rv, + }; + if let Err(e) = substrate.delete_pod(name, &fence).await { + eprintln!("gc: could not delete terminated pod {name}: {e}"); + } + } + for name in &plan.secrets { + if let Err(e) = substrate.delete_secret(name).await { + eprintln!("gc: could not delete secret {name}: {e}"); + } + } + Ok(()) +} + +/// Wait for a pod to actually disappear. +/// +/// Mandatory before recreating: `DELETE` returns success while the object +/// still exists, and the deterministic name stays taken for the whole grace +/// period (`:1177-1195`). +async fn await_disappearance(substrate: &impl Substrate, name: &str) -> Result<(), String> { + while substrate.elapsed() < DEADLINE { + if substrate.get_pod(name).await?.is_none() { + return Ok(()); + } + substrate.sleep(POLL_INTERVAL).await; + } + Err(format!( + "timed out after {}s waiting for {name} to finish terminating", + DEADLINE.as_secs() + )) +} + +/// Is `secret` referenced by any pod that currently exists under this +/// identity's selector? +/// +/// Protection deliberately spans *all* our pods, not just the winner: an +/// `envFrom` reference from a pod still pulling its image is exactly as +/// load-bearing as one from a running pod (`:1261-1264`). +async fn secret_is_referenced( + substrate: &impl Substrate, + identity: &AgentIdentity, + secret: &str, +) -> Result { + let (pods, _) = substrate.list_pods(&identity.selector()).await?; + Ok(pods + .iter() + .filter_map(observe::referenced_secret) + .any(|name| name == secret)) +} + +/// Drop this attempt's own Secret once nothing references it. +/// +/// Only ever called with a name this process generated, and gated on the +/// reference check: "never the winner's, never any Secret referenced by an +/// existing pod" (`:1259-1264`). Failure is logged, not fatal — a leaked +/// Secret is collected by the age-gated sweep. +async fn drop_own_secret(substrate: &impl Substrate, identity: &AgentIdentity, secret: &str) { + match secret_is_referenced(substrate, identity, secret).await { + Ok(false) => { + if let Err(e) = substrate.delete_secret(secret).await { + eprintln!("could not clean up own unreferenced secret {secret}: {e}"); + } + } + Ok(true) => {} + Err(e) => eprintln!("could not check whether {secret} is still referenced: {e}"), + } +} + +/// Lost the create race: adopt the elected winner. +/// +/// Observe-only by construction — this function has no delete edge for the +/// pod. A winner that is terminated or provably broken is reported, not +/// repaired; the spec's escape is "a *subsequent* deploy that walks in and +/// observes that never-started divergent winner replaces it normally" +/// (`:849-850`). +async fn adopt_winner( + substrate: &impl Substrate, + identity: &AgentIdentity, + own_secret: &str, +) -> Result { + let name = identity.pod_name(); + + // Verify before adopting. A pod under our deterministic name that fails + // the marker/annotation check is not ours to adopt, wait for, or touch — + // and it will never become ours, so this is terminal rather than a retry. + match substrate.get_pod(&name).await? { + None => {} + Some(pod) => { + let startup = settle(substrate, &pod).await?; + if observe::verify(&pod, identity, startup).is_none() { + drop_own_secret(substrate, identity, own_secret).await; + return Err(format!( + "a pod named {name} already exists in this namespace but is not \ + managed by this provider for this agent (it lacks the management \ + marker or carries a different agent identity). Remove it, or \ + deploy this agent to a different namespace." + )); + } + } + } + + drop_own_secret(substrate, identity, own_secret).await; + + // Then wait for the winner exactly as we would wait for our own pod: + // success still means the harness container started. + loop { + if substrate.elapsed() >= DEADLINE { + return Err(format!( + "startup not confirmed within {}s for {name} (another deploy of this \ + agent created it): {}", + DEADLINE.as_secs(), + latest_condition(substrate, identity).await + )); + } + match observe_pod(substrate, identity).await? { + Some(pod) if matches!(pod.startup, Startup::Started) => return Ok(pod.name), + Some(pod) if pod.deletion_marked => { + return Err(format!( + "{name} was created by another deploy of this agent and is already \ + being deleted; try again" + )) + } + Some(pod) + if matches!( + pod.startup, + Startup::Terminated | Startup::NeverStartedProvablyBroken + ) => + { + return Err(format!( + "{name} was created by another deploy of this agent and did not \ + start: {}", + latest_condition(substrate, identity).await + )) + } + // Gone again, or still coming up: keep observing under this + // operation's deadline. + _ => substrate.sleep(POLL_INTERVAL).await, + } + } +} + +/// Run the deploy state machine to a terminal outcome: the started pod's name, +/// or an in-band error carrying the latest condition. +pub async fn deploy( + substrate: &impl Substrate, + identity: &AgentIdentity, + cfg: &ProviderConfig, + env: BTreeMap, +) -> Result { + substrate.ensure_namespace(&cfg.namespace).await?; + preflight_gc(substrate, identity).await; + + let desired = crate::pod::intent_template(cfg, env.keys().cloned()).fingerprint(); + + // Has THIS call created a pod? Set once its create lands. The replacement + // rows below are for residue from a previous life; once this call has made + // its own attempt, a replace-classification means that attempt failed — + // and startup verification is part of create, so the failure is reported + // in-band rather than retried. Without this bound a deterministic startup + // failure (the harness starts, rejects its configuration, exits) is + // delete-recreated every poll for the whole deadline, minting an immutable + // Secret per cycle — measured live at 107 Secrets in one 600s call, every + // one younger than the orphan sweep's age gate. + let mut created_this_call = false; + + loop { + if substrate.elapsed() >= DEADLINE { + return Err(format!( + "startup not confirmed within {}s for {}: {}", + DEADLINE.as_secs(), + identity.pod_name(), + latest_condition(substrate, identity).await + )); + } + + let observed = observe_pod(substrate, identity).await?; + match classify::classify(observed.as_ref(), &desired) { + // The only success edge: the harness container is running. + Action::NoOp { agent_id } => return Ok(agent_id), + + // Self-healing states. Never delete, on this call or any later one + // — what replaces a never-started pod is a config change, never a + // deadline (`:717-729`). + Action::Observe { .. } => substrate.sleep(POLL_INTERVAL).await, + + // A pull that will not self-heal: report now rather than spend the + // remaining deadline on it. Still no delete authority. + Action::Report { name, failure } => { + return Err(format!( + "{name} did not start: {}", + observe::pull_failure_message(failure, cfg.image.as_str()) + )) + } + + Action::AwaitDisappearance { name } => await_disappearance(substrate, &name).await?, + + Action::Delete { name, fence } => { + // This call already made its own attempt, and that attempt is + // what the classification wants replaced: it terminated (the + // deterministic startup failure — the harness starts, rejects + // its configuration, exits) or was proven broken. Replacing it + // here retries the identical configuration against the same + // cluster: a hot delete/mint/create cycle every poll for the + // whole deadline, an immutable Secret per cycle — measured + // live at 107 Secrets in one 600s call, all younger than the + // orphan sweep's age gate. Report in-band instead. The residue + // is deliberate: the next Start's preflight GC collects the + // terminated pod and its referenced Secret together, so retry + // is gated on fresh owner intent and litter stays bounded at + // one pod + one Secret per press. + if created_this_call { + return Err(format!( + "{name} was created by this deploy and did not stay \ + running: {}. Not retrying in this call — an immediate \ + exit recurs until its cause is fixed. Check the \ + agent's configuration and press Start to try again.", + latest_condition(substrate, identity).await + )); + } + match substrate.delete_pod(&name, &fence).await? { + // Accepted or already gone: both need the disappearance + // poll before the name is free again. + DeleteOutcome::Accepted | DeleteOutcome::NotFound => { + await_disappearance(substrate, &name).await? + } + // The object changed since the observation that authorized + // this delete. Discard the action and re-classify — never + // retry with a fresher fence, which would delete something + // we never examined. Sleep before re-entering: the losing + // race is against another writer, and re-reading at full + // speed is a busy-retry with no better odds than a paced + // one. + DeleteOutcome::PreconditionFailed => substrate.sleep(POLL_INTERVAL).await, + } + } + + Action::Create => { + let generation = crate::naming::new_generation(); + let secret_name = identity.secret_name(&generation); + + // The generation is minted *per attempt*, and it is two things + // at once: the Secret's name suffix and the lifecycle + // correlator the harness reports. `build_env` stamped the + // caller's generation, so on any attempt after the first the + // two would name different generations — pod logs correlating + // to a Secret that is not the one mounted. Restamp so there is + // exactly one generation per attempt (§K8s Secrets). + let mut env = env.clone(); + env.insert(crate::env::START_NONCE_KEY.to_string(), generation.clone()); + + // Secret first: the pod's spec references this exact name, so + // payload and Secret are atomic at the pod-spec boundary. + let secret = crate::pod::build_secret(identity, &cfg.namespace, &generation, env); + substrate.create_secret(&secret).await?; + + let pod = crate::pod::build_pod(identity, cfg, &generation, &desired); + match substrate.create_pod(&pod).await? { + // Re-enter rather than wait inline: the next iteration + // observes what we just created and runs the same rows + // every other state runs through. One loop, one table. + // + // Sleep first. A just-created pod cannot already be + // started, so the immediate observation has no outcome but + // "still coming up" — and if it ever came back + // unverifiable, re-entering without advancing the clock + // would hot-spin creates against the apiserver for the + // whole deadline. + CreateOutcome::Created => { + created_this_call = true; + substrate.sleep(POLL_INTERVAL).await + } + CreateOutcome::AlreadyExists => { + return adopt_winner(substrate, identity, &secret_name).await + } + } + } + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::config::Resources; + use crate::naming::{ANNOTATION_CREATE_INTENT, ANNOTATION_PUBKEY_FULL, LABEL_MANAGED_BY}; + use k8s_openapi::api::core::v1::{ + ContainerState, ContainerStateRunning, ContainerStateTerminated, ContainerStateWaiting, + ContainerStatus, PodStatus, + }; + use k8s_openapi::apimachinery::pkg::apis::meta::v1::Time; + use std::cell::RefCell; + use std::future::Future; + use std::pin::pin; + use std::task::{Context, Poll, RawWaker, RawWakerVTable, Waker}; + + /// Mutates the pod map after a poll, so a test can script a pod that + /// starts (or vanishes) partway through an observation loop. + type PollHook = Box)>; + + /// A scripted cluster. Single-threaded on purpose: the reconciler is one + /// process per operation, and `RefCell` keeps the assertions readable. + /// + /// This drives the *shipped* `deploy` — the point of the `Substrate` seam. + /// Every fake here answers with real `k8s_openapi` objects, so the decode + /// and verification layers under test are the ones that run in a cluster. + #[derive(Default)] + struct Fake { + pods: RefCell>, + secrets: RefCell>, + /// Server clock for the GC age gate; `None` models a missing `Date`. + server_now: Option>, + /// Elapsed time, advanced only by `sleep` — a fake clock, so a 600s + /// deadline test runs instantly and cannot flake on a slow machine. + elapsed: RefCell, + /// Queued create outcomes; the default is `Created`. + create_outcomes: RefCell>, + /// Installed when a create loses the race. A winner must be *absent* + /// at the observation that decides to create and *present* by the time + /// the create lands — pre-seeding it instead makes the loop no-op + /// before it ever reaches the create edge. + winner: RefCell>, + /// Every Secret ever created, retained across deletion. + created_secrets: RefCell>, + /// Queued delete outcomes; the default is `Accepted`. + delete_outcomes: RefCell>, + /// Every mutating call, in order — the anti-mutation assertions read + /// this rather than guessing from final state. + calls: RefCell>, + /// Applied to the pod map after each poll, so a test can script a pod + /// that starts (or vanishes) partway through an observation loop. + on_poll: RefCell>, + /// `ensure_namespace` fails with this, if set. + namespace_error: Option, + } + + impl Fake { + fn with_pod(self, pod: Pod) -> Self { + self.pods + .borrow_mut() + .insert(pod.metadata.name.clone().unwrap(), pod); + self + } + fn log(&self, entry: impl Into) { + self.calls.borrow_mut().push(entry.into()); + } + fn mutations(&self) -> Vec { + self.calls.borrow().clone() + } + } + + impl Substrate for Fake { + async fn ensure_namespace(&self, namespace: &str) -> Result<(), String> { + match &self.namespace_error { + Some(e) => Err(e.clone()), + None => { + self.log(format!("ensure_namespace {namespace}")); + Ok(()) + } + } + } + + async fn list_pods( + &self, + _selector: &str, + ) -> Result<(Vec, Option>), String> { + Ok(( + self.pods.borrow().values().cloned().collect(), + self.server_now, + )) + } + + async fn list_secrets(&self, _selector: &str) -> Result, String> { + Ok(self.secrets.borrow().clone()) + } + + async fn secret_exists(&self, name: &str) -> Result { + Ok(self + .secrets + .borrow() + .iter() + .any(|s| s.metadata.name.as_deref() == Some(name))) + } + + async fn create_secret(&self, secret: &Secret) -> Result<(), String> { + let name = secret.metadata.name.clone().unwrap(); + self.log(format!("create_secret {name}")); + self.secrets.borrow_mut().push(secret.clone()); + // Kept even after the Secret is deleted: assertions about what an + // attempt *wrote* must not be silently vacuous once cleanup runs. + self.created_secrets.borrow_mut().push(secret.clone()); + Ok(()) + } + + async fn create_pod(&self, pod: &Pod) -> Result { + let name = pod.metadata.name.clone().unwrap(); + self.log(format!("create_pod {name}")); + let outcome = if self.create_outcomes.borrow().is_empty() { + CreateOutcome::Created + } else { + self.create_outcomes.borrow_mut().remove(0) + }; + if outcome == CreateOutcome::Created { + // The apiserver stamps these on admission; a builder never + // carries them. Without them the pod fails `verify`'s fence + // extraction and the loop can never see what it just created. + let mut pod = pod.clone(); + pod.metadata.uid = Some(format!("uid-created-{}", self.calls.borrow().len())); + pod.metadata.resource_version = Some("1".into()); + self.pods.borrow_mut().insert(name, pod); + } else if let Some(winner) = self.winner.borrow_mut().take() { + // The concurrent attempt's pod becomes visible exactly when our + // create is rejected — the ordering a real race produces. + self.pods.borrow_mut().insert(name, winner); + } + Ok(outcome) + } + + async fn delete_pod(&self, name: &str, fence: &Fence) -> Result { + self.log(format!( + "delete_pod {name} uid={} rv={}", + fence.uid, fence.resource_version + )); + let outcome = if self.delete_outcomes.borrow().is_empty() { + DeleteOutcome::Accepted + } else { + self.delete_outcomes.borrow_mut().remove(0) + }; + if outcome == DeleteOutcome::Accepted { + self.pods.borrow_mut().remove(name); + } + Ok(outcome) + } + + async fn delete_secret(&self, name: &str) -> Result<(), String> { + self.log(format!("delete_secret {name}")); + self.secrets + .borrow_mut() + .retain(|s| s.metadata.name.as_deref() != Some(name)); + Ok(()) + } + + async fn get_pod(&self, name: &str) -> Result, String> { + Ok(self.pods.borrow().get(name).cloned()) + } + + async fn sleep(&self, duration: Duration) { + *self.elapsed.borrow_mut() += duration; + let hooks = self.on_poll.borrow(); + let mut pods = self.pods.borrow_mut(); + for hook in hooks.iter() { + hook(&mut pods); + } + } + + fn elapsed(&self) -> Duration { + *self.elapsed.borrow() + } + } + + fn identity() -> AgentIdentity { + use nostr::nips::nip19::ToBech32; + let keys = nostr::Keys::generate(); + AgentIdentity::from_nsec(&keys.secret_key().to_bech32().unwrap()).unwrap() + } + + fn config() -> ProviderConfig { + ProviderConfig { + context: None, + namespace: "buzz-agents-test".into(), + image: crate::image::parse(&format!( + "ghcr.io/block/buzz-sprig@sha256:{}", + "a".repeat(64) + )) + .unwrap(), + resources: Resources::default(), + inactivity_seconds: Some(7200), + service_account: None, + } + } + + fn env() -> BTreeMap { + [("BUZZ_RELAY_URL".to_string(), "wss://r".to_string())] + .into_iter() + .collect() + } + + /// A pod exactly as this provider would have created it — same builder the + /// reconciler uses, so verification is exercised rather than bypassed. + fn our_pod(id: &AgentIdentity, cfg: &ProviderConfig, state: Option) -> Pod { + let fp = crate::pod::intent_template(cfg, env().keys().cloned()).fingerprint(); + let mut pod = crate::pod::build_pod(id, cfg, "gen-existing", &fp); + pod.metadata.uid = Some("uid-1".into()); + pod.metadata.resource_version = Some("100".into()); + pod.status = Some(PodStatus { + phase: Some("Running".into()), + container_statuses: state.map(|s| { + vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(s), + ..Default::default() + }] + }), + ..Default::default() + }); + pod + } + + fn running() -> ContainerState { + ContainerState { + running: Some(ContainerStateRunning::default()), + ..Default::default() + } + } + + fn terminated() -> ContainerState { + ContainerState { + terminated: Some(ContainerStateTerminated::default()), + ..Default::default() + } + } + + fn waiting(reason: &str) -> ContainerState { + ContainerState { + waiting: Some(ContainerStateWaiting { + reason: Some(reason.into()), + message: None, + }), + ..Default::default() + } + } + + fn run(fake: &Fake, id: &AgentIdentity, cfg: &ProviderConfig) -> Result { + block_on(deploy(fake, id, cfg, env())) + } + + /// Minimal executor: the fake never yields to a reactor (no timers, no + /// I/O — `sleep` just advances a counter), so polling to completion is + /// sufficient and avoids pulling a runtime into the unit job. + fn block_on(fut: impl Future) -> T { + fn noop_waker() -> Waker { + fn nop(_: *const ()) {} + fn clone(_: *const ()) -> RawWaker { + RawWaker::new(std::ptr::null(), &VTABLE) + } + static VTABLE: RawWakerVTable = RawWakerVTable::new(clone, nop, nop, nop); + unsafe { Waker::from_raw(RawWaker::new(std::ptr::null(), &VTABLE)) } + } + + let waker = noop_waker(); + let mut cx = Context::from_waker(&waker); + let mut fut = pin!(fut); + match fut.as_mut().poll(&mut cx) { + Poll::Ready(v) => v, + Poll::Pending => panic!("fake substrate future parked — it has no reactor"), + } + } + + // ---- the state machine's rows, end to end ------------------------------- + + /// First deploy: Secret before pod, and the returned id is the pod name. + #[test] + fn creates_secret_then_pod_and_returns_the_pod_name() { + let id = identity(); + let cfg = config(); + let fake = Fake::default(); + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + Box::new(move |pods: &mut BTreeMap| { + if let Some(pod) = pods.get_mut(&name) { + pod.status = Some(PodStatus { + phase: Some("Running".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(running()), + ..Default::default() + }]), + ..Default::default() + }); + } + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + + let calls = fake.mutations(); + let secret_at = calls + .iter() + .position(|c| c.starts_with("create_secret")) + .unwrap(); + let pod_at = calls + .iter() + .position(|c| c.starts_with("create_pod")) + .unwrap(); + assert!( + secret_at < pod_at, + "pod created before its Secret: {calls:?}" + ); + } + + /// The strict no-op row: a started pod returns its id having mutated + /// nothing at all. Asserted on the *call log*, not on final state — a + /// delete-then-recreate would leave identical final state. + #[test] + fn started_pod_is_a_zero_mutation_no_op() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(running()))); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert_eq!( + fake.mutations(), + [format!("ensure_namespace {}", cfg.namespace)], + "the no-op row mutated something" + ); + } + + /// ...including when the desired intent has diverged. Edits reach a + /// started pod only via the next generation (`:861-865`). + #[test] + fn started_pod_no_ops_under_divergent_intent() { + let id = identity(); + let cfg = config(); + let mut pod = our_pod(&id, &cfg, Some(running())); + pod.metadata.annotations.as_mut().unwrap().insert( + ANNOTATION_CREATE_INTENT.to_string(), + "stale-fingerprint".into(), + ); + let fake = Fake::default().with_pod(pod); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert!( + !fake.mutations().iter().any(|c| c.starts_with("delete_pod")), + "divergence deleted a started pod" + ); + } + + /// Terminated → fenced delete → disappearance → recreate. The normal + /// restart path, and the fence must carry the observed uid/rv. + #[test] + fn terminated_pod_is_replaced_with_a_fenced_delete() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(terminated()))); + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + Box::new(move |pods: &mut BTreeMap| { + if let Some(pod) = pods.get_mut(&name) { + if pod + .status + .as_ref() + .and_then(|s| s.container_statuses.as_ref()) + .is_none() + { + pod.status = Some(PodStatus { + phase: Some("Running".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(running()), + ..Default::default() + }]), + ..Default::default() + }); + } + } + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + let calls = fake.mutations(); + assert!( + calls + .iter() + .any(|c| c == &format!("delete_pod {} uid=uid-1 rv=100", id.pod_name())), + "delete was not fenced to the observed uid+resourceVersion: {calls:?}" + ); + assert!(calls.iter().any(|c| c.starts_with("create_pod"))); + } + + /// A failed precondition is neither an error nor permission to retry the + /// delete: re-enter and classify what exists now (`:775-777`). Here the + /// object has become a *started* pod, so the correct outcome is the no-op + /// row — never a second delete with a fresher fence. + #[test] + fn precondition_failure_reclassifies_instead_of_retrying() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(terminated()))); + fake.delete_outcomes + .borrow_mut() + .push(DeleteOutcome::PreconditionFailed); + // The writer we lost the race to: between our observation and our + // delete, the pod under this name became a *started* one with a new + // resourceVersion. Installed on the poll that follows the failed + // precondition, so the sequence is observe → delete → lose → re-observe. + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + let started = { + let mut p = our_pod(&id, &cfg, Some(running())); + p.metadata.resource_version = Some("200".into()); + p + }; + Box::new(move |pods: &mut BTreeMap| { + pods.insert(name.clone(), started.clone()); + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + let deletes: Vec<_> = fake + .mutations() + .into_iter() + .filter(|c| c.starts_with("delete_pod")) + .collect(); + // Two call sites legitimately target a terminated pod: preflight GC's + // sweep (which is the one that loses the precondition here) and the + // state machine's terminated row. Both are fenced on their own + // observation. What must never happen is a *third* — a retry of the + // failed delete — so the invariant is the fence, not the count: every + // delete carries rv=100, the version we observed. A retry would carry + // the racing writer's rv=200. + assert_eq!(deletes.len(), 2, "unexpected delete traffic: {deletes:?}"); + assert!( + deletes.iter().all(|d| d.contains("rv=100")), + "retried with a fresher fence: {deletes:?}" + ); + } + + /// The anti-livelock rule, at the loop level: a recoverable pod whose + /// intent matches is observed until the deadline and **never** deleted — + /// the case that would otherwise reset the pod age Cluster Autoscaler + /// keys on (`:689`). + #[test] + fn recoverable_pod_with_matching_intent_is_never_deleted() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(waiting("Unschedulable")))); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("startup not confirmed"), "got: {err}"); + assert!( + !fake.mutations().iter().any(|c| c.starts_with("delete_pod")), + "deleted a recoverable pod: {:?}", + fake.mutations() + ); + assert!(fake.elapsed() >= DEADLINE, "gave up before the deadline"); + } + + /// The same pod, provisioned late: the observation loop must *succeed* + /// when the autoscaler eventually lands the node, not merely avoid + /// deleting. + #[test] + fn recoverable_pod_that_starts_late_succeeds() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(waiting("Unschedulable")))); + let polls = RefCell::new(0); + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + Box::new(move |pods: &mut BTreeMap| { + *polls.borrow_mut() += 1; + if *polls.borrow() >= 5 { + if let Some(pod) = pods.get_mut(&name) { + pod.status.as_mut().unwrap().container_statuses = + Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(running()), + ..Default::default() + }]); + } + } + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert!(fake.elapsed() < DEADLINE); + } + + /// A permanent pull failure reports immediately rather than burning the + /// deadline — and still never deletes. + #[test] + fn permanent_pull_failure_reports_without_deleting() { + let id = identity(); + let cfg = config(); + let mut pod = our_pod(&id, &cfg, None); + pod.status = Some(PodStatus { + phase: Some("Pending".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(ContainerState { + waiting: Some(ContainerStateWaiting { + reason: Some("ImagePullBackOff".into()), + message: Some("manifest unknown".into()), + }), + ..Default::default() + }), + ..Default::default() + }]), + ..Default::default() + }); + let fake = Fake::default().with_pod(pod); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("no image at"), "got: {err}"); + assert!( + fake.elapsed() < DEADLINE, + "burned the deadline on a permanent failure" + ); + assert!(!fake.mutations().iter().any(|c| c.starts_with("delete_pod"))); + } + + /// The kubelet's pull *message* can echo a registry request; only the + /// classified outcome and the image reference reach the user. + #[test] + fn pull_failure_message_is_not_echoed_verbatim() { + let id = identity(); + let cfg = config(); + let mut pod = our_pod(&id, &cfg, None); + pod.status = Some(PodStatus { + phase: Some("Pending".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(ContainerState { + waiting: Some(ContainerStateWaiting { + reason: Some("ErrImagePull".into()), + message: Some( + "unauthorized: authentication required, token=SUPERSECRET".into(), + ), + }), + ..Default::default() + }), + ..Default::default() + }]), + ..Default::default() + }); + let fake = Fake::default().with_pod(pod); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!( + !err.contains("SUPERSECRET"), + "kubelet message echoed: {err}" + ); + } + + /// A pod being gracefully deleted stays in phase `Running` for its whole + /// grace period. The reconciler must wait it out and recreate, not return + /// an id that evaporates. + #[test] + fn deletion_marked_pod_is_awaited_then_recreated() { + let id = identity(); + let cfg = config(); + let mut dying = our_pod(&id, &cfg, Some(running())); + dying.metadata.deletion_timestamp = Some(Time(Utc::now())); + let fake = Fake::default().with_pod(dying); + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + Box::new(move |pods: &mut BTreeMap| { + match pods + .get(&name) + .map(|p| p.metadata.deletion_timestamp.is_some()) + { + // The grace period elapses: the object disappears. + Some(true) => { + pods.remove(&name); + } + // The replacement we create then starts. + Some(false) => { + let pod = pods.get_mut(&name).unwrap(); + pod.status = Some(PodStatus { + phase: Some("Running".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(running()), + ..Default::default() + }]), + ..Default::default() + }); + } + None => {} + } + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert!( + !fake.mutations().iter().any(|c| c.starts_with("delete_pod")), + "deleted a pod that was already terminating" + ); + } + + /// `CreateContainerConfigError` is settled by a most-recent Secret read, + /// never by the reason string: Secret present → recoverable (observe). + #[test] + fn config_error_with_a_present_secret_is_recoverable() { + let id = identity(); + let cfg = config(); + let pod = our_pod(&id, &cfg, Some(waiting("CreateContainerConfigError"))); + let referenced = crate::observe::referenced_secret(&pod).unwrap(); + let fake = Fake::default().with_pod(pod); + fake.secrets.borrow_mut().push(crate::pod::build_secret( + &id, + &cfg.namespace, + "gen-existing", + env(), + )); + assert_eq!(referenced, id.secret_name("gen-existing")); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("startup not confirmed"), "got: {err}"); + assert!( + !fake.mutations().iter().any(|c| c.starts_with("delete_pod")), + "deleted a pod whose Secret exists" + ); + } + + /// ...and Secret confirmed absent → provably broken → fenced replace. + #[test] + fn config_error_with_a_confirmed_absent_secret_is_replaced() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod( + &id, + &cfg, + Some(waiting("CreateContainerConfigError")), + )); + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + Box::new(move |pods: &mut BTreeMap| { + if let Some(pod) = pods.get_mut(&name) { + if pod + .status + .as_ref() + .and_then(|s| s.container_statuses.as_ref()) + .is_none() + { + pod.status = Some(PodStatus { + phase: Some("Running".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(running()), + ..Default::default() + }]), + ..Default::default() + }); + } + } + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert!(fake.mutations().iter().any(|c| c.starts_with("delete_pod"))); + } + + /// The generation is the Secret's name suffix *and* the lifecycle + /// correlator inside it. They must name the same generation, or pod logs + /// point at a Secret that was never mounted. The caller stamps one + /// generation into the env once; each create attempt mints its own, so + /// only a restamp keeps the pair together across a retry. + #[test] + fn each_attempt_stamps_its_own_generation_into_its_own_secret() { + let id = identity(); + let cfg = config(); + // First attempt loses the create race and is cleaned up; the operation + // then adopts. Two creates, two generations. + let fake = Fake::default(); + *fake.winner.borrow_mut() = Some(our_pod(&id, &cfg, Some(running()))); + fake.create_outcomes + .borrow_mut() + .push(CreateOutcome::AlreadyExists); + + run(&fake, &id, &cfg).unwrap(); + + let created = fake.created_secrets.borrow(); + // Guard the guard: this assertion is over a list that cleanup empties, + // so an empty list would make every check below vacuously true. + assert_eq!(created.len(), 1, "expected one create attempt"); + for secret in created.iter() { + let name = secret.metadata.name.as_deref().unwrap(); + let nonce = secret + .string_data + .as_ref() + .and_then(|d| d.get(crate::env::START_NONCE_KEY)) + .expect("no lifecycle correlator in the Secret"); + assert!( + name.ends_with(nonce.as_str()), + "secret {name} carries a correlator for a different generation ({nonce})" + ); + } + } + + // ---- bounded replacement ------------------------------------------------ + + /// Installs a hook that makes every pod under `name` crash-exit before the + /// next poll — a deterministic startup failure (the harness starts, rejects + /// its configuration, and exits) as observed live: the container reaches + /// `state.terminated` with a nonzero exit code. + fn crash_exits_immediately(fake: &Fake, name: String) { + fake.on_poll + .borrow_mut() + .push(Box::new(move |pods: &mut BTreeMap| { + if let Some(pod) = pods.get_mut(&name) { + if pod + .status + .as_ref() + .and_then(|s| s.container_statuses.as_ref()) + .is_none() + { + pod.status = Some(PodStatus { + phase: Some("Failed".into()), + container_statuses: Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(ContainerState { + terminated: Some(ContainerStateTerminated { + exit_code: 1, + reason: Some("Error".into()), + ..Default::default() + }), + ..Default::default() + }), + ..Default::default() + }]), + ..Default::default() + }); + } + } + })); + } + + /// A pod that this call created and that crash-exits deterministically is + /// reported in-band after exactly ONE attempt — never hot-replaced until + /// the deadline. The live failure this pins: one delete/create cycle every + /// ~4s minted 107 immutable Secrets in a single 600s deploy call, all + /// younger than the orphan sweep's age gate. + #[test] + fn a_deterministic_crash_exit_is_reported_not_hot_replaced() { + let id = identity(); + let cfg = config(); + let fake = Fake::default(); + crash_exits_immediately(&fake, id.pod_name()); + + let err = run(&fake, &id, &cfg).unwrap_err(); + + let creates = fake + .mutations() + .iter() + .filter(|c| c.starts_with("create_secret")) + .count(); + assert_eq!( + creates, 1, + "hot replacement loop: {creates} Secrets minted in one deploy call" + ); + assert!( + err.contains("exited with code 1"), + "error does not carry the exit: {err}" + ); + // The failed attempt is left in place as evidence — no cleanup on the + // error path. Its Secret stays referenced by the terminated pod, so + // the next deploy's preflight GC collects both together. + assert!( + !fake + .mutations() + .iter() + .any(|c| c.starts_with("delete_pod") || c.starts_with("delete_secret")), + "the failed attempt was cleaned up on the error path: {:?}", + fake.mutations() + ); + assert!( + fake.elapsed() < DEADLINE, + "burned the whole deadline on a deterministic failure" + ); + } + + /// The revive path is untouched: terminated residue from a *previous* + /// life is still replaced — the bound is on pods this call created, not + /// on the row. + #[test] + fn pre_existing_terminated_residue_is_still_replaced_once() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(terminated()))); + crash_exits_immediately(&fake, id.pod_name()); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("exited with code 1"), "got: {err}"); + + let calls = fake.mutations(); + let deletes = calls.iter().filter(|c| c.starts_with("delete_pod")).count(); + let creates = calls + .iter() + .filter(|c| c.starts_with("create_secret")) + .count(); + // Exactly one residue delete (the normal restart path) and one fresh + // attempt — then report, not another cycle. + assert_eq!(deletes, 1, "unexpected delete traffic: {calls:?}"); + assert_eq!(creates, 1, "unexpected create traffic: {calls:?}"); + } + + /// Retry is gated on fresh owner intent: the *next* deploy call clears + /// the crashed attempt (preflight GC collects the terminated pod and its + /// referenced Secret together) and makes exactly one new attempt — total + /// litter stays one pod + one Secret however many times Start is pressed. + #[test] + fn the_next_deploy_collects_the_crashed_attempt_before_its_own_attempt() { + let id = identity(); + let cfg = config(); + let fake = Fake::default(); + crash_exits_immediately(&fake, id.pod_name()); + + run(&fake, &id, &cfg).unwrap_err(); + // Second Start: fresh call, fresh clock. + *fake.elapsed.borrow_mut() = Duration::ZERO; + run(&fake, &id, &cfg).unwrap_err(); + + assert_eq!( + fake.secrets.borrow().len(), + 1, + "crashed attempts accumulated Secrets across calls" + ); + assert_eq!(fake.pods.borrow().len(), 1, "crashed pods accumulated"); + } + + // ---- the auto-repair fence ---------------------------------------------- + + /// An object under our deterministic name that lacks the management + /// marker is not ours. The reconciler must not adopt it, delete it, or + /// return its name — it fails closed to the operator (`:1156-1162`). + #[test] + fn an_unmarked_look_alike_is_never_touched_or_adopted() { + let id = identity(); + let cfg = config(); + let mut look_alike = our_pod(&id, &cfg, Some(running())); + look_alike + .metadata + .labels + .as_mut() + .unwrap() + .remove(LABEL_MANAGED_BY); + let fake = Fake::default(); + // Our create loses to the object already sitting on the name. + *fake.winner.borrow_mut() = Some(look_alike); + fake.create_outcomes + .borrow_mut() + .push(CreateOutcome::AlreadyExists); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("not managed by this provider"), "got: {err}"); + assert!( + !fake.mutations().iter().any(|c| c.starts_with("delete_pod")), + "deleted an object we do not own" + ); + } + + /// Same fence, other direction: the marker is present but the annotation + /// carries a different agent's pubkey. The 32-hex label is + /// collision-resistant, not collision-free (`:1152-1155`). + #[test] + fn a_pubkey_mismatch_is_never_adopted() { + let id = identity(); + let other = identity(); + let cfg = config(); + let mut foreign = our_pod(&id, &cfg, Some(running())); + foreign.metadata.annotations.as_mut().unwrap().insert( + ANNOTATION_PUBKEY_FULL.to_string(), + other.pubkey_hex().to_string(), + ); + let fake = Fake::default(); + *fake.winner.borrow_mut() = Some(foreign); + fake.create_outcomes + .borrow_mut() + .push(CreateOutcome::AlreadyExists); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("different agent identity"), "got: {err}"); + assert!(!fake.mutations().iter().any(|c| c.starts_with("delete_pod"))); + } + + // ---- create-conflict convergence --------------------------------------- + + /// The loser adopts the winner, returns the winner's id, and deletes + /// **only its own** Secret — never the winner's (`:1259-1264`). + #[test] + fn create_loser_adopts_the_winner_and_drops_only_its_own_secret() { + let id = identity(); + let cfg = config(); + let winner = our_pod(&id, &cfg, Some(running())); + let winners_secret = crate::observe::referenced_secret(&winner).unwrap(); + let fake = Fake::default(); + *fake.winner.borrow_mut() = Some(winner); + fake.secrets.borrow_mut().push(crate::pod::build_secret( + &id, + &cfg.namespace, + "gen-existing", + env(), + )); + fake.create_outcomes + .borrow_mut() + .push(CreateOutcome::AlreadyExists); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + + let deleted: Vec<_> = fake + .mutations() + .into_iter() + .filter(|c| c.starts_with("delete_secret")) + .collect(); + assert_eq!( + deleted.len(), + 1, + "expected exactly our own Secret dropped: {deleted:?}" + ); + assert!( + !deleted[0].contains(&winners_secret), + "deleted the winner's Secret: {deleted:?}" + ); + assert!( + fake.secrets + .borrow() + .iter() + .any(|s| s.metadata.name.as_deref() == Some(winners_secret.as_str())), + "the winner's Secret is gone" + ); + } + + /// The loser must not apply the divergence row to the pod that just beat + /// it — that is the ping-pong the spec forbids (`:845-850`). A divergent + /// *started* winner is adopted as-is. + #[test] + fn create_loser_does_not_replace_a_divergent_winner() { + let id = identity(); + let cfg = config(); + let mut winner = our_pod(&id, &cfg, Some(running())); + winner.metadata.annotations.as_mut().unwrap().insert( + ANNOTATION_CREATE_INTENT.to_string(), + "a-different-intent".into(), + ); + let fake = Fake::default(); + *fake.winner.borrow_mut() = Some(winner); + fake.create_outcomes + .borrow_mut() + .push(CreateOutcome::AlreadyExists); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert!( + !fake.mutations().iter().any(|c| c.starts_with("delete_pod")), + "the create loser deleted the winner" + ); + } + + /// The loser waits for the winner to *start* — adopting a not-yet-started + /// winner as success would report an agent that is not running. + #[test] + fn create_loser_waits_for_the_winner_to_start() { + let id = identity(); + let cfg = config(); + let fake = Fake::default(); + *fake.winner.borrow_mut() = Some(our_pod(&id, &cfg, Some(waiting("ContainerCreating")))); + fake.create_outcomes + .borrow_mut() + .push(CreateOutcome::AlreadyExists); + let polls = RefCell::new(0); + fake.on_poll.borrow_mut().push({ + let name = id.pod_name(); + Box::new(move |pods: &mut BTreeMap| { + *polls.borrow_mut() += 1; + if *polls.borrow() >= 3 { + if let Some(pod) = pods.get_mut(&name) { + pod.status.as_mut().unwrap().container_statuses = + Some(vec![ContainerStatus { + name: crate::observe::CONTAINER_NAME.into(), + state: Some(running()), + ..Default::default() + }]); + } + } + }) + }); + + assert_eq!(run(&fake, &id, &cfg).unwrap(), id.pod_name()); + assert!(fake.elapsed() > Duration::ZERO, "did not wait at all"); + } + + // ---- deadline and namespace -------------------------------------------- + + /// Deadline expiry reports the *latest* condition, not a generic timeout + /// (`:699-701`), and triggers no cleanup (`:720-724`). + #[test] + fn deadline_expiry_reports_the_condition_and_cleans_up_nothing() { + let id = identity(); + let cfg = config(); + let fake = Fake::default().with_pod(our_pod(&id, &cfg, Some(waiting("ContainerCreating")))); + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("ContainerCreating"), "generic timeout: {err}"); + let calls = fake.mutations(); + assert!( + !calls.iter().any(|c| c.starts_with("delete_")), + "deadline expiry triggered cleanup: {calls:?}" + ); + } + + /// An RBAC denial on namespace create fails the deploy with the literal + /// command to run, before any Secret is written (`:1002-1005`). + #[test] + fn namespace_denial_fails_before_writing_any_secret() { + let id = identity(); + let cfg = config(); + let fake = Fake { + namespace_error: Some(format!( + "not authorized to create namespaces: run `kubectl create namespace {}`", + cfg.namespace + )), + ..Default::default() + }; + + let err = run(&fake, &id, &cfg).unwrap_err(); + assert!(err.contains("kubectl create namespace"), "got: {err}"); + assert!( + fake.mutations().is_empty(), + "wrote something after a namespace denial: {:?}", + fake.mutations() + ); + } + + /// GC failures are hygiene, not deploy failures: a list the user cannot + /// perform must not block a deploy they can. + #[test] + fn a_gc_failure_does_not_fail_the_deploy() { + struct GcDenied(Fake); + impl Substrate for GcDenied { + async fn ensure_namespace(&self, ns: &str) -> Result<(), String> { + self.0.ensure_namespace(ns).await + } + async fn list_pods( + &self, + _s: &str, + ) -> Result<(Vec, Option>), String> { + Err("forbidden: cannot list pods".into()) + } + async fn list_secrets(&self, _s: &str) -> Result, String> { + Err("forbidden: cannot list secrets".into()) + } + async fn secret_exists(&self, n: &str) -> Result { + self.0.secret_exists(n).await + } + async fn create_secret(&self, s: &Secret) -> Result<(), String> { + self.0.create_secret(s).await + } + async fn create_pod(&self, p: &Pod) -> Result { + self.0.create_pod(p).await + } + async fn delete_pod(&self, n: &str, f: &Fence) -> Result { + self.0.delete_pod(n, f).await + } + async fn delete_secret(&self, n: &str) -> Result<(), String> { + self.0.delete_secret(n).await + } + async fn get_pod(&self, n: &str) -> Result, String> { + self.0.get_pod(n).await + } + async fn sleep(&self, d: Duration) { + self.0.sleep(d).await + } + fn elapsed(&self) -> Duration { + self.0.elapsed() + } + } + + let id = identity(); + let cfg = config(); + let inner = Fake::default().with_pod(our_pod(&id, &cfg, Some(running()))); + let fake = GcDenied(inner); + + assert_eq!( + block_on(deploy(&fake, &id, &cfg, env())).unwrap(), + id.pod_name() + ); + } +} diff --git a/crates/buzz-backend-kubernetes/src/wire.rs b/crates/buzz-backend-kubernetes/src/wire.rs new file mode 100644 index 0000000000..89b78b364d --- /dev/null +++ b/crates/buzz-backend-kubernetes/src/wire.rs @@ -0,0 +1,250 @@ +//! The stdin/stdout JSON protocol (spec §Provider Protocol). +//! +//! One process per operation: one JSON object in, one JSON object out. +//! These types are this provider's view of the contract; the golden fixtures +//! in `tests/fixtures/provider-wire/` are the arbiter shared with the desktop. + +use serde::{Deserialize, Serialize}; +use std::collections::BTreeMap; + +/// The wire-contract version this provider speaks (spec §Info). +pub const PROTOCOL_VERSION: u32 = 1; + +/// Request envelope. `op` discriminates; unknown ops are an in-band error. +/// +/// `request_id` is deliberately absent from every variant. The desktop sends +/// it, but the exchange is one request and one response per process, so there +/// is nothing to correlate and nothing in the response schema to echo it into +/// (`:387`, `:416` — neither response shape carries it). Serde ignores it on +/// the way in, so a caller that sends it is accepted; typing it would only +/// create a field nothing reads. +#[derive(Debug, Deserialize)] +#[serde(tag = "op", rename_all = "lowercase")] +pub enum Request { + Info, + Deploy(Box), +} + +#[derive(Debug, Deserialize)] +pub struct DeployRequest { + pub agent: AgentPayload, + #[serde(default)] + pub provider_config: serde_json::Value, +} + +/// The agent payload (spec §Deploy). +/// +/// Only the fields this binding actually consumes are typed. `name` (display +/// name), `model`, `provider`, and `turn_timeout_seconds` are deliberately +/// absent: object names derive from the pubkey, not the display name +/// (`:1150-1152`), the model/provider pair arrives already resolved inside +/// `launch`, and the timeout is ignored upstream — typing any of them would +/// invite a provider-side remap the spec forbids. +#[derive(Debug, Deserialize)] +pub struct AgentPayload { + pub relay_url: String, + pub private_key_nsec: String, + #[serde(default)] + pub auth_tag: Option, + #[serde(default)] + pub respond_to: Option, + #[serde(default)] + pub respond_to_allowlist: Option>, + /// User env, already merged global < persona < agent by the desktop and + /// already stripped of reserved keys. Superseded by `launch.env` when + /// `launch` is present — a provider MUST NOT re-merge it on top + /// (§Launch data, precedence tier 2). + #[serde(default)] + pub env_vars: BTreeMap, + /// The desktop-resolved launch contract. Absent only from a desktop + /// predating Known Defect 3's fix. + #[serde(default)] + pub launch: Option, +} + +/// Desktop-resolved launch data (spec §Launch data). +#[derive(Debug, Default, Deserialize)] +pub struct LaunchBlock { + /// Command *name*, resolved against the image's PATH — never a host path. + #[serde(default)] + pub command: Option, + #[serde(default)] + pub args: Vec, + /// Layered env: baked → runtime metadata → definition → global → persona + /// → agent. Precedence tier 2. + #[serde(default)] + pub env: BTreeMap, + /// Overridable behavior defaults. Precedence tier 1 — user env beats + /// these, matching the local spawn. + #[serde(default)] + pub policy_env: BTreeMap, + /// Resolved workspace owner (hex). The respond-to gate's one + /// irreducible input; without it or `auth_tag` the harness cannot match + /// `!shutdown`. + #[serde(default)] + pub owner_pubkey: Option, +} + +/// Response envelope. Serialized flat — `{"ok": true, …}` — because the +/// desktop reads `ok`, `error`, and `agent_id` off the top level. +#[derive(Debug, Serialize)] +#[serde(untagged)] +pub enum Response { + Info(InfoResponse), + Deploy(DeployResponse), + Error(ErrorResponse), +} + +#[derive(Debug, Serialize)] +pub struct InfoResponse { + pub ok: bool, + pub name: &'static str, + pub version: &'static str, + pub protocol_version: u32, + pub description: &'static str, + pub config_schema: serde_json::Value, +} + +#[derive(Debug, Serialize)] +pub struct DeployResponse { + pub ok: bool, + pub agent_id: String, +} + +#[derive(Debug, Serialize)] +pub struct ErrorResponse { + pub ok: bool, + pub error: String, +} + +impl Response { + pub fn error(message: impl Into) -> Self { + Response::Error(ErrorResponse { + ok: false, + error: message.into(), + }) + } + + /// The provider's self-description (spec §Info). Pure — no cluster + /// contact — because the desktop calls it to render the config form + /// before a kubeconfig is known to exist. + pub fn info() -> Self { + Response::Info(InfoResponse { + ok: true, + name: "kubernetes", + version: env!("CARGO_PKG_VERSION"), + protocol_version: PROTOCOL_VERSION, + description: "Runs agents as pods in a Kubernetes cluster", + config_schema: crate::config::config_schema(), + }) + } + + pub fn deployed(agent_id: impl Into) -> Self { + Response::Deploy(DeployResponse { + ok: true, + agent_id: agent_id.into(), + }) + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn parses_info_request() { + let r: Request = serde_json::from_str(r#"{"op":"info","request_id":"abc"}"#).unwrap(); + assert!(matches!(r, Request::Info)); + } + + /// The desktop sends `request_id` on every call, but the exchange is 1:1 + /// per process — a provider that hard-required it would fail a + /// conforming-but-minimal caller for no safety gain. + #[test] + fn request_id_is_optional() { + let r: Request = serde_json::from_str(r#"{"op":"info"}"#).unwrap(); + assert!(matches!(r, Request::Info)); + } + + #[test] + fn rejects_unknown_op() { + assert!(serde_json::from_str::(r#"{"op":"undeploy"}"#).is_err()); + } + + /// Payload fields this binding does not consume must not break parsing: + /// the desktop sends `model`, `provider`, `system_prompt` and more, and a + /// provider that rejected them would break on every real deploy. + #[test] + fn ignores_unconsumed_payload_fields() { + let json = r#"{ + "op":"deploy","request_id":"r1", + "agent":{ + "name":"a","relay_url":"wss://r","private_key_nsec":"nsec1x", + "model":"gpt-5","provider":"openai","system_prompt":"hi", + "turn_timeout_seconds":30,"parallelism":10, + "agent_command":"goose","agent_args":[] + }, + "provider_config":{"namespace":"ns"} + }"#; + let r: Request = serde_json::from_str(json).unwrap(); + let Request::Deploy(d) = r else { + panic!("wrong op") + }; + assert_eq!(d.agent.relay_url, "wss://r"); + assert!(d.agent.launch.is_none()); + } + + #[test] + fn parses_launch_block() { + let json = r#"{ + "op":"deploy", + "agent":{ + "relay_url":"wss://r","private_key_nsec":"nsec1x", + "launch":{ + "command":"goose","args":["run","--x"], + "env":{"GOOSE_MODEL":"m"}, + "policy_env":{"GOOSE_MODE":"auto"}, + "owner_pubkey":"deadbeef" + } + } + }"#; + let Request::Deploy(d) = serde_json::from_str::(json).unwrap() else { + panic!("wrong op") + }; + let l = d.agent.launch.unwrap(); + assert_eq!(l.command.as_deref(), Some("goose")); + assert_eq!(l.args, ["run", "--x"]); + assert_eq!(l.env["GOOSE_MODEL"], "m"); + assert_eq!(l.policy_env["GOOSE_MODE"], "auto"); + assert_eq!(l.owner_pubkey.as_deref(), Some("deadbeef")); + } + + /// A null `owner_pubkey`/`auth_tag` must parse (the refusal is a policy + /// decision made later, with a specific message), not fail as a type error. + #[test] + fn null_owner_fields_parse() { + let json = r#"{"op":"deploy","agent":{ + "relay_url":"wss://r","private_key_nsec":"nsec1x","auth_tag":null, + "launch":{"owner_pubkey":null} + }}"#; + let Request::Deploy(d) = serde_json::from_str::(json).unwrap() else { + panic!("wrong op") + }; + assert!(d.agent.auth_tag.is_none()); + assert!(d.agent.launch.unwrap().owner_pubkey.is_none()); + } + + /// The desktop reads `ok`/`error`/`agent_id` off the top level, so the + /// enum must serialize flat with no variant tag. + #[test] + fn responses_serialize_flat() { + let v = serde_json::to_value(Response::deployed("buzz-agent-abc")).unwrap(); + assert_eq!(v["ok"], true); + assert_eq!(v["agent_id"], "buzz-agent-abc"); + assert!(v.get("Deploy").is_none()); + + let v = serde_json::to_value(Response::error("boom")).unwrap(); + assert_eq!(v["ok"], false); + assert_eq!(v["error"], "boom"); + } +} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/README.md b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/README.md new file mode 100644 index 0000000000..56a972f894 --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/README.md @@ -0,0 +1,39 @@ +# Provider wire fixtures + +The shared arbiter for the stdin/stdout contract between the desktop +(`agents_deploy.rs`) and this provider (spec §Provider Protocol). + +Each `*.request.json` is a request the desktop can emit; each matching +`*.response.json` is the exact response this provider produces for it. The +provider side is asserted by `tests/wire_fixtures.rs`; the desktop side should +assert that its emitted payloads parse as the corresponding request. + +Three rules keep these useful rather than decorative: + +* **Requests are recorded, not invented.** A fixture that no caller emits + tests a contract nobody has. "Recorded" means *executed and transcribed* — + `deploy-full-launch.request.json` is the output of the desktop's real + `build_launch_block` → `deploy_payload_json` path, not a shape derived by + reading those functions. Deriving it is how this fixture acquired four + impossible values at once: a `respond_to` that was a pubkey where the + desktop serializes a kebab-case `RespondTo` enum, allowlist and owner + values failing `validate_respond_to_allowlist`'s 64-hex rule + (`types.rs:897`), an invented `BUZZ_ACP_PARALLELISM` where the emitter + writes `BUZZ_ACP_AGENTS` (`runtime.rs:729`), and a `launch.env` key from + no layer of `resolve_effective_harness_descriptor`. +* **The provider cannot police this file, so the desktop must.** Every field + above is one this provider is deliberately indifferent to — `respond_to` is + an opaque `Option`, the allowlist an opaque `Vec`, + `policy_env` an arbitrary map — so `the_full_desktop_payload_is_accepted` + passes on invented data exactly as happily as on recorded data. The + enforcement is the desktop's whole-object equality test, which *builds* the + payload and compares it to this file. A completeness guard (the case-list + directory scan in `wire_fixtures.rs`) stops a case from going missing; it + cannot tell you a case is false. +* **Responses are byte-compared after key-sorted re-serialization**, so a + field rename or a type change fails here rather than in a desktop that + silently reads `undefined`. + +`deploy-*` fixtures cover only responses reachable without a cluster — +refusals and malformed input. A successful deploy needs an apiserver and is +covered by the conformance suite, not by a static fixture. diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-full-launch.request.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-full-launch.request.json new file mode 100644 index 0000000000..28fe6ce90e --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-full-launch.request.json @@ -0,0 +1,52 @@ +{ + "op": "deploy", + "request_id": "req-6", + "agent": { + "agent_args": [], + "agent_command": "goose", + "auth_tag": "tag-1", + "env_vars": { + "USER_KEY": "user-value" + }, + "idle_timeout_seconds": null, + "launch": { + "args": [ + "acp" + ], + "command": "goose", + "env": { + "GOOSE_MODEL": "gpt-5", + "GOOSE_PROVIDER": "openai", + "USER_KEY": "user-value" + }, + "owner_pubkey": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", + "policy_env": { + "BUZZ_ACP_AGENTS": "10", + "BUZZ_ACP_LAZY_POOL": "true", + "BUZZ_ACP_MODEL": "gpt-5", + "BUZZ_ACP_RELAY_OBSERVER": "true", + "BUZZ_ACP_SESSION_TITLE": "worker", + "GOOSE_MODE": "auto" + } + }, + "max_turn_duration_seconds": null, + "model": "gpt-5", + "name": "worker", + "parallelism": 10, + "private_key_nsec": "nsec1vl029mgpspedva04g90vltkh6fvh240zqtv9k0t9af8935ke9laqsnlfe5", + "provider": "openai", + "relay_url": "wss://relay.example", + "respond_to": "allowlist", + "respond_to_allowlist": [ + "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", + "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb" + ], + "system_prompt": null, + "turn_timeout_seconds": 300 + }, + "provider_config": { + "namespace": "buzz-agents-test", + "image": "ghcr.io/block/buzz-sprig@sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", + "inactivity_seconds": 3600 + } +} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-no-owner.request.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-no-owner.request.json new file mode 100644 index 0000000000..4f56082bbb --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-no-owner.request.json @@ -0,0 +1,9 @@ +{ + "op": "deploy", + "request_id": "req-5", + "agent": { + "relay_url": "wss://relay.example", + "private_key_nsec": "nsec1vl029mgpspedva04g90vltkh6fvh240zqtv9k0t9af8935ke9laqsnlfe5" + }, + "provider_config": {"namespace": "buzz-agents-test", "image": "ghcr.io/block/buzz-sprig@sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"} +} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-no-owner.response.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-no-owner.response.json new file mode 100644 index 0000000000..c6011b899a --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-no-owner.response.json @@ -0,0 +1 @@ +{"ok":false,"error":"deploy refused: neither auth_tag nor launch.owner_pubkey resolved — without an owner the agent cannot honor !shutdown"} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh-padded.request.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh-padded.request.json new file mode 100644 index 0000000000..2160909b8f --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh-padded.request.json @@ -0,0 +1,11 @@ +{ + "op": "deploy", + "request_id": "req-3", + "agent": { + "relay_url": "wss://relay.example", + "private_key_nsec": "nsec1vl029mgpspedva04g90vltkh6fvh240zqtv9k0t9af8935ke9laqsnlfe5", + "auth_tag": "tag-1", + "provider": " relay-mesh " + }, + "provider_config": {"namespace": "buzz-agents-test", "image": "ghcr.io/block/buzz-sprig@sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"} +} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh-padded.response.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh-padded.response.json new file mode 100644 index 0000000000..0f0c995fd8 --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh-padded.response.json @@ -0,0 +1 @@ +{"ok":false,"error":"deploy refused: this agent is configured for shared compute (relay-mesh), which runs on the relay rather than in a pod. Switch the agent to a local runtime before deploying it to Kubernetes."} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh.request.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh.request.json new file mode 100644 index 0000000000..71ce9f067e --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh.request.json @@ -0,0 +1,12 @@ +{ + "op": "deploy", + "request_id": "req-2", + "agent": { + "name": "mesh-agent", + "relay_url": "wss://relay.example", + "private_key_nsec": "nsec1vl029mgpspedva04g90vltkh6fvh240zqtv9k0t9af8935ke9laqsnlfe5", + "auth_tag": "tag-1", + "provider": "relay-mesh" + }, + "provider_config": {"namespace": "buzz-agents-test", "image": "ghcr.io/block/buzz-sprig@sha256:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"} +} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh.response.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh.response.json new file mode 100644 index 0000000000..0f0c995fd8 --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-relay-mesh.response.json @@ -0,0 +1 @@ +{"ok":false,"error":"deploy refused: this agent is configured for shared compute (relay-mesh), which runs on the relay rather than in a pod. Switch the agent to a local runtime before deploying it to Kubernetes."} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-tag-image.request.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-tag-image.request.json new file mode 100644 index 0000000000..232d9dcf43 --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-tag-image.request.json @@ -0,0 +1,10 @@ +{ + "op": "deploy", + "request_id": "req-4", + "agent": { + "relay_url": "wss://relay.example", + "private_key_nsec": "nsec1vl029mgpspedva04g90vltkh6fvh240zqtv9k0t9af8935ke9laqsnlfe5", + "auth_tag": "tag-1" + }, + "provider_config": {"namespace": "buzz-agents-test", "image": "ghcr.io/block/buzz-sprig:latest"} +} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-tag-image.response.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-tag-image.response.json new file mode 100644 index 0000000000..e67fa6a47f --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-tag-image.response.json @@ -0,0 +1 @@ +{"ok":false,"error":"provider_config.image \"ghcr.io/block/buzz-sprig:latest\" is not digest-pinned: a tag is a mutable pointer, and this object runs with the agent's private key. Use name@sha256:<64 hex chars>"} diff --git a/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/info.request.json b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/info.request.json new file mode 100644 index 0000000000..db99c86c00 --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/info.request.json @@ -0,0 +1 @@ +{"op":"info","request_id":"req-1"} diff --git a/crates/buzz-backend-kubernetes/tests/wire_fixtures.rs b/crates/buzz-backend-kubernetes/tests/wire_fixtures.rs new file mode 100644 index 0000000000..b3049a98ef --- /dev/null +++ b/crates/buzz-backend-kubernetes/tests/wire_fixtures.rs @@ -0,0 +1,222 @@ +//! Golden wire fixtures (spec §Provider Protocol). +//! +//! These drive the **built binary** over a real pipe rather than calling an +//! in-process function: the contract the desktop depends on is +//! `stdin → one JSON object on stdout → exit code`, and an in-process test +//! would assert the shape of a value while skipping the three things that +//! actually break — the process writing nothing, writing two objects, or +//! signalling the outcome through the exit code. + +use std::io::Write; +use std::path::{Path, PathBuf}; +use std::process::{Command, Stdio}; + +fn fixtures() -> PathBuf { + Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/fixtures/provider-wire") +} + +/// Feed one request to the binary; return `(stdout, exit code)`. +fn run(request: &str) -> (String, i32) { + let mut child = Command::new(env!("CARGO_BIN_EXE_buzz-backend-kubernetes")) + // A kubeconfig that does not exist, so a fixture that accidentally + // reaches the cluster fails loudly here instead of depending on + // whatever cluster the developer is pointed at. + .env("KUBECONFIG", "/nonexistent/kubeconfig-for-fixture-tests") + .stdin(Stdio::piped()) + .stdout(Stdio::piped()) + .spawn() + .expect("could not run the provider binary"); + child + .stdin + .take() + .expect("no stdin") + .write_all(request.as_bytes()) + .expect("could not write the request"); + let out = child.wait_with_output().expect("provider did not exit"); + ( + String::from_utf8(out.stdout).expect("stdout was not UTF-8"), + out.status.code().unwrap_or(-1), + ) +} + +fn read(name: &str) -> String { + std::fs::read_to_string(fixtures().join(name)) + .unwrap_or_else(|e| panic!("could not read fixture {name}: {e}")) +} + +/// Every response fixture, byte-compared after key-sorted re-serialization so +/// a field rename fails here rather than in a desktop reading `undefined`. +#[test] +fn responses_match_their_fixtures() { + let cases = [ + "deploy-relay-mesh", + "deploy-relay-mesh-padded", + "deploy-tag-image", + "deploy-no-owner", + ]; + // The list must cover every response fixture on disk. A literal array is + // never empty, so `!is_empty()` would assert nothing; what can actually go + // wrong is a fixture added to the directory and never added here, which + // reads as a passing suite that exercises one case fewer than it appears to. + let mut on_disk: Vec = std::fs::read_dir(fixtures()) + .expect("could not read the fixture directory") + .filter_map(|entry| entry.ok()?.file_name().into_string().ok()) + .filter_map(|name| Some(name.strip_suffix(".response.json")?.to_string())) + .collect(); + on_disk.sort(); + let mut listed: Vec = cases.iter().map(|c| c.to_string()).collect(); + listed.sort(); + assert_eq!(on_disk, listed, "response fixtures and cases disagree"); + + for case in cases { + let (stdout, code) = run(&read(&format!("{case}.request.json"))); + assert_eq!(code, 0, "{case}: a produced response must exit 0"); + + // Exactly one object, terminated by exactly one newline. Two responses + // would leave the desktop's reader holding a second one forever. + assert_eq!( + stdout.matches('\n').count(), + 1, + "{case}: expected exactly one line, got {stdout:?}" + ); + + let actual: serde_json::Value = + serde_json::from_str(&stdout).unwrap_or_else(|e| panic!("{case}: {e}: {stdout:?}")); + let expected: serde_json::Value = + serde_json::from_str(&read(&format!("{case}.response.json"))).unwrap(); + assert_eq!( + actual, expected, + "{case}: response drifted from its fixture" + ); + } +} + +/// `info` is checked on the fields the desktop reads rather than byte-for-byte: +/// the namespace default is randomly generated per call (§K8s Namespace), so a +/// golden copy of it would be a test that fails every run. +#[test] +fn info_response_carries_the_contract_fields() { + let (stdout, code) = run(&read("info.request.json")); + assert_eq!(code, 0); + let info: serde_json::Value = serde_json::from_str(&stdout).unwrap(); + assert_eq!(info["ok"], true); + assert_eq!(info["protocol_version"], 1); + assert_eq!(info["name"], "kubernetes"); + let schema = &info["config_schema"]; + assert_eq!( + schema["required"], + serde_json::json!(["namespace", "image"]) + ); + let default = schema["properties"]["namespace"]["default"] + .as_str() + .expect("no generated namespace default"); + assert!( + default.starts_with("buzz-agents-"), + "unexpected namespace default: {default}" + ); + let image_default = schema["properties"]["image"]["default"] + .as_str() + .expect("no image default"); + assert!( + image_default.starts_with("ghcr.io/block/buzz-sprig:") + && image_default.contains("@sha256:"), + "unexpected image default: {image_default}" + ); +} + +/// The desktop's richest payload must parse. No response fixture: this one +/// reaches the cluster, so its outcome depends on a kubeconfig. What it +/// guards is that every field the desktop sends is *accepted* — a payload the +/// provider rejects at parse time is a deploy that never starts. +#[test] +fn the_full_desktop_payload_is_accepted() { + let (stdout, code) = run(&read("deploy-full-launch.request.json")); + assert_eq!(code, 0); + let response: serde_json::Value = serde_json::from_str(&stdout).unwrap(); + let error = response["error"].as_str().unwrap_or_default(); + // It fails — there is no cluster — but it must fail at the *connection*, + // having accepted every field above it. + assert!( + error.contains("kubeconfig"), + "the full payload was rejected before reaching the cluster: {error}" + ); +} + +/// Sami's pre-registered respond-to matrix, driven through the built binary. +/// +/// `build_env` runs at `main.rs:124`, `client::connect` at `:132`, so under a +/// kubeconfig that cannot exist the error string *is* the ordering assertion: +/// "kubeconfig" means the gate passed and we reached the cluster, anything +/// else means we refused before writing a Secret. A test asserting only +/// `ok: false` would pass on the connection error and prove nothing. +/// +/// The cases are applied to the real full-launch request so each one differs +/// from a known-good deploy in exactly the field under test. +#[test] +fn the_respond_to_gate_matches_the_harness_acceptance_surface() { + let key_a = "a".repeat(64); + let padded_upper = format!(" {} ", "A".repeat(64)); + // (name, respond_to, allowlist, must reach the cluster) + let cases: Vec<(&str, &str, Option>, bool)> = vec![ + ("allowlist + []", "allowlist", Some(vec![]), false), + ("allowlist + absent", "allowlist", None, false), + ( + "allowlist + junk", + "allowlist", + Some(vec!["beefcafe".into()]), + false, + ), + ("unparseable mode", "npub1abc", None, false), + ("padded mode", " allowlist ", None, false), + ( + "allowlist + two valid", + "allowlist", + Some(vec![key_a.clone(), "b".repeat(64)]), + true, + ), + ( + "owner-only + junk list", + "owner-only", + Some(vec!["beefcafe".into()]), + true, + ), + ( + "allowlist + padded upper", + "allowlist", + Some(vec![padded_upper]), + true, + ), + ("nobody", "nobody", None, true), + ("anyone", "anyone", None, true), + ]; + + let base: serde_json::Value = + serde_json::from_str(&read("deploy-full-launch.request.json")).unwrap(); + + for (name, mode, allowlist, reaches_cluster) in cases { + let mut request = base.clone(); + let agent = &mut request["agent"]; + agent["respond_to"] = serde_json::json!(mode); + agent["respond_to_allowlist"] = match &allowlist { + Some(list) => serde_json::json!(list), + None => serde_json::Value::Null, + }; + + let (stdout, code) = run(&request.to_string()); + assert_eq!(code, 0, "{name}: provider did not exit cleanly"); + let response: serde_json::Value = serde_json::from_str(&stdout).unwrap(); + let error = response["error"].as_str().unwrap_or_default(); + let reached = error.contains("kubeconfig"); + + assert_eq!( + reached, reaches_cluster, + "{name}: expected reaches_cluster={reaches_cluster}, got error: {error}" + ); + if !reaches_cluster { + assert!( + error.contains("deploy refused"), + "{name}: refused, but not by the gate: {error}" + ); + } + } +} diff --git a/crates/buzz-cli/src/commands/mod.rs b/crates/buzz-cli/src/commands/mod.rs index 8691590636..1ccc37a702 100644 --- a/crates/buzz-cli/src/commands/mod.rs +++ b/crates/buzz-cli/src/commands/mod.rs @@ -12,9 +12,37 @@ pub mod notes; pub mod pack; pub mod patches; pub mod pr; +pub mod projects; pub mod reactions; pub mod repos; pub mod social; pub mod upload; pub mod users; pub mod workflows; + +use crate::{client::normalize_write_response, error::CliError}; + +/// Parse a relay write-response JSON blob, mapping a duplicate (dominated) +/// write to [`CliError::Conflict`] with the caller-supplied message. +/// +/// Used by every command that publishes an NIP-33 addressable event and +/// needs to tell accepted from duplicate/dominated. +pub fn parse_write_response(raw: &str, conflict_msg: &str) -> Result { + let response: serde_json::Value = serde_json::from_str(raw) + .map_err(|e| CliError::Other(format!("relay response is not JSON: {e} ({raw})")))?; + let accepted = response + .get("accepted") + .and_then(serde_json::Value::as_bool) + .unwrap_or(false); + let message = response + .get("message") + .and_then(serde_json::Value::as_str) + .unwrap_or(""); + if !accepted { + return Err(CliError::Other(format!("relay rejected event: {message}"))); + } + if message == "duplicate" || message.starts_with("duplicate:") { + return Err(CliError::Conflict(conflict_msg.to_string())); + } + Ok(normalize_write_response(raw)) +} diff --git a/crates/buzz-cli/src/commands/projects.rs b/crates/buzz-cli/src/commands/projects.rs new file mode 100644 index 0000000000..e6798dbfc4 --- /dev/null +++ b/crates/buzz-cli/src/commands/projects.rs @@ -0,0 +1,1198 @@ +//! `buzz projects` commands — NIP-MP kind:30621 write path. +//! +//! All mutations follow a read-modify-write pattern: +//! 1. Fetch the caller's own live head via `kinds:[30621] + authors:[self] + #d:[slug]`. +//! 2. Mutate the tag set (strip `auth`, apply change). +//! 3. Re-validate the full envelope through Layer A before submitting. +//! 4. Set `created_at = head.created_at + 1` (never wall-clock) to avoid +//! overwriting a concurrently advancing head. +//! +//! Limitations recorded in this phase: +//! - Relay hints are read-preserved but not authored (`--repo` carries +//! a coordinate only; existing hinted tags survive RMW unchanged). +//! - `delete` targets signer-self only (NIP-OA owner-delete path deferred). +//! - Deletion durability against later arrival (watermark follow-up) is +//! not in scope. + +use buzz_core::kind::KIND_PROJECT; +use buzz_sdk::{ + build_delete_addressable, build_project, build_project_with_tags, ProjectMemberCoord, + PROJECT_D_MAX_LEN, +}; +use nostr::{Event, EventBuilder, Tag, Timestamp}; + +use crate::client::BuzzClient; +use crate::commands::parse_write_response; +use crate::error::CliError; + +// ── Buzz repo-ID grammar (bare --repo shorthand) ───────────────────────────── + +/// Pattern for a Buzz-hosted repo identifier (bare `--repo` shorthand). +/// `[a-zA-Z0-9._-]{1,64}` — no colons, so guaranteed collision-free with +/// `30617::` full coordinates. +fn is_bare_repo_id(s: &str) -> bool { + !s.is_empty() + && s.len() <= 64 + && s.chars() + .all(|c| c.is_ascii_alphanumeric() || c == '.' || c == '_' || c == '-') +} + +/// Expand a CLI `--repo` argument into a full `30617::` coordinate. +/// +/// Bare form (`[a-zA-Z0-9._-]{1,64}`): owner defaults to the caller's pubkey. +/// Full form (`30617::`): used verbatim. +fn expand_repo_coord(s: &str, caller_pubkey: &str) -> Result { + if is_bare_repo_id(s) { + // Bare form: expand to full coordinate with caller as owner. + let full = format!("30617:{caller_pubkey}:{s}"); + ProjectMemberCoord::parse_full(&full) + .map_err(|e| CliError::Usage(format!("invalid repo coordinate: {e}"))) + } else { + // Full form: must be parseable as a complete coordinate. + ProjectMemberCoord::parse_full(s) + .map_err(|e| CliError::Usage(format!("invalid repo coordinate: {e}"))) + } +} + +// ── Head-fetch helper ───────────────────────────────────────────────────────── + +fn parse_events(json: &str) -> Result, CliError> { + serde_json::from_str(json) + .map_err(|e| CliError::Other(format!("failed to parse relay response: {e}"))) +} + +/// Fetch the caller's own live kind:30621 head for `slug`. +async fn fetch_own_project(client: &BuzzClient, slug: &str) -> Result, CliError> { + fetch_project(client, slug, None).await +} + +/// Fetch a project head by slug and optional owner pubkey. +async fn fetch_project( + client: &BuzzClient, + slug: &str, + owner: Option<&str>, +) -> Result, CliError> { + let pubkey = match owner { + Some(pk) => { + crate::validate::validate_hex64(pk)?; + pk.to_string() + } + None => client.keys().public_key().to_hex(), + }; + let filter = serde_json::json!({ + "kinds": [KIND_PROJECT], + "authors": [pubkey], + "#d": [slug], + "limit": 1, + }); + let raw = client.query(&filter).await?; + let mut events = parse_events(&raw)?; + events.sort_by_key(|e| std::cmp::Reverse(e.created_at)); + Ok(events.into_iter().next()) +} + +// ── Tag helpers ─────────────────────────────────────────────────────────────── + +fn tag_name(tag: &Tag) -> Option<&str> { + tag.as_slice().first().map(String::as_str) +} + +fn tag_value(tag: &Tag) -> Option<&str> { + tag.as_slice().get(1).map(String::as_str) +} + +fn make_tag(parts: &[&str]) -> Result { + Tag::parse(parts.iter().copied()) + .map_err(|e| CliError::Other(format!("tag construction failed: {e}"))) +} + +// ── Submit helper ───────────────────────────────────────────────────────────── + +async fn submit_project(client: &BuzzClient, builder: EventBuilder) -> Result<(), CliError> { + let event = client.sign_event(builder)?; + let raw = client.submit_event(event).await?; + println!( + "{}", + parse_write_response(&raw, "project changed concurrently; retry")? + ); + Ok(()) +} + +// ── Build helpers ───────────────────────────────────────────────────────────── + +/// Advance the `created_at` counter off an observed head. +fn next_timestamp(head: &Event) -> Result { + head.created_at + .as_secs() + .checked_add(1) + .map(Timestamp::from) + .ok_or_else(|| CliError::Other("project timestamp cannot be advanced".into())) +} + +/// Strip `auth` from a tag list and pass the resulting envelope through +/// Layer A validation. Returns a validated `EventBuilder` at `next_ts`. +fn rebuild_project( + content: &str, + tags: Vec, + next_ts: Timestamp, +) -> Result { + // Strip auth tags. + let clean_tags: Vec = tags + .into_iter() + .filter(|t| tag_name(t) != Some("auth")) + .collect(); + + build_project_with_tags(content, clean_tags) + .map_err(|e| CliError::Other(format!("envelope validation failed: {e}"))) + .map(|b| b.custom_created_at(next_ts)) +} + +// ── Command implementations ─────────────────────────────────────────────────── + +/// `buzz projects create` +pub async fn cmd_create( + client: &BuzzClient, + slug: &str, + repos: &[String], + name: Option<&str>, + description: Option<&str>, + channel: Option<&str>, + visibility: Option<&str>, +) -> Result<(), CliError> { + // ── Local validation (all checks before any .await) ─────────────────── + validate_project_slug(slug)?; + + let caller_pubkey = client.keys().public_key().to_hex(); + + // Expand and validate repo coordinates. + let members: Vec = repos + .iter() + .map(|r| expand_repo_coord(r, &caller_pubkey)) + .collect::, _>>()?; + + // Dedupe: preserve first occurrence, reject duplicates with Usage. + let mut seen = std::collections::HashSet::new(); + for m in &members { + if !seen.insert(m.coord.clone()) { + return Err(CliError::Usage(format!( + "duplicate --repo coordinate in this invocation: {:?}", + m.coord + ))); + } + } + + // Validate optional metadata (early, before any network call). + if let Some(ch) = channel { + crate::validate::validate_uuid(ch)?; + } + if let Some(vis) = visibility { + validate_visibility(vis)?; + } + if let Some(n) = name { + if n.len() > 256 { + return Err(CliError::Usage(format!( + "project name must not exceed 256 bytes (got {})", + n.len() + ))); + } + } + + // ── Network: collision preflight ────────────────────────────────────── + if fetch_own_project(client, slug).await?.is_some() { + return Err(CliError::Conflict(format!( + "project {slug:?} already exists; use 'buzz projects update' to modify it" + ))); + } + + // ── Build via Layer B (enforces all writer policy) ──────────────────── + let builder = build_project(slug, name, description, &members, channel, visibility) + .map_err(|e| CliError::Usage(e.to_string()))?; + submit_project(client, builder).await +} + +/// `buzz projects get` +pub async fn cmd_get(client: &BuzzClient, slug: &str, owner: Option<&str>) -> Result<(), CliError> { + validate_project_slug(slug)?; + let resp = match fetch_project(client, slug, owner).await? { + Some(event) => serde_json::json!({ + "event_id": event.id.to_hex(), + "pubkey": event.pubkey.to_hex(), + "created_at": event.created_at.as_secs(), + "kind": event.kind.as_u16(), + "tags": event.tags.iter().map(|t| t.as_slice().to_vec()).collect::>(), + "content": event.content, + }), + None => { + let owner_desc = owner.unwrap_or("current identity"); + return Err(CliError::NotFound(format!( + "project {slug:?} not found for {owner_desc}" + ))); + } + }; + println!("{resp}"); + Ok(()) +} + +/// `buzz projects list` +pub async fn cmd_list( + client: &BuzzClient, + owner: Option<&str>, + limit: Option, +) -> Result<(), CliError> { + let pubkey = match owner { + Some(pk) => { + crate::validate::validate_hex64(pk)?; + pk.to_string() + } + None => client.keys().public_key().to_hex(), + }; + let mut filter = serde_json::json!({ + "kinds": [KIND_PROJECT], + "authors": [pubkey], + }); + if let Some(n) = limit { + filter["limit"] = serde_json::json!(n); + } + let resp = client.query(&filter).await?; + println!("{resp}"); + Ok(()) +} + +/// `buzz projects add-repo` +pub async fn cmd_add_repo( + client: &BuzzClient, + slug: &str, + repos: &[String], +) -> Result<(), CliError> { + validate_project_slug(slug)?; + let caller_pubkey = client.keys().public_key().to_hex(); + + // ── Local validation before any .await ──────────────────────────────── + let new_members: Vec = repos + .iter() + .map(|r| expand_repo_coord(r, &caller_pubkey)) + .collect::, _>>()?; + + // Dedupe within this invocation: first occurrence wins, duplicate → Usage. + let mut seen = std::collections::HashSet::new(); + for m in &new_members { + if !seen.insert(m.coord.clone()) { + return Err(CliError::Usage(format!( + "duplicate --repo coordinate in this invocation: {:?}", + m.coord + ))); + } + } + + // ── Network: fetch head ─────────────────────────────────────────────── + let head = fetch_own_project(client, slug) + .await? + .ok_or_else(|| CliError::NotFound(format!("project {slug:?} not found")))?; + let next_ts = next_timestamp(&head)?; + + // Build the new tag set: keep existing tags (including hinted members), + // append new members only if not already present (by coordinate). + let mut tags: Vec = head.tags.iter().cloned().collect(); + let existing_coords: std::collections::HashSet = head + .tags + .iter() + .filter(|t| tag_name(t) == Some("a")) + .filter_map(|t| tag_value(t).map(String::from)) + .collect(); + let mut added = 0usize; + for m in &new_members { + if !existing_coords.contains(m.coord.as_str()) { + let parts = m.to_tag_parts(); + let parts_ref: Vec<&str> = parts.iter().map(String::as_str).collect(); + tags.push( + Tag::parse(parts_ref.iter().copied()) + .map_err(|e| CliError::Other(format!("member tag construction failed: {e}")))?, + ); + added += 1; + } + } + + // All requested coordinates were already present — no change to publish. + if added == 0 { + return Err(CliError::Conflict(format!( + "all requested repositories are already members of project {slug:?}" + ))); + } + + let builder = rebuild_project(&head.content, tags, next_ts)?; + submit_project(client, builder).await +} + +/// `buzz projects remove-repo` +pub async fn cmd_remove_repo( + client: &BuzzClient, + slug: &str, + repos: &[String], +) -> Result<(), CliError> { + validate_project_slug(slug)?; + let caller_pubkey = client.keys().public_key().to_hex(); + + // ── Local validation before any .await ──────────────────────────────── + let to_remove: Vec = repos + .iter() + .map(|r| expand_repo_coord(r, &caller_pubkey)) + .collect::, _>>()?; + + // ── Network: fetch head ─────────────────────────────────────────────── + let head = fetch_own_project(client, slug) + .await? + .ok_or_else(|| CliError::NotFound(format!("project {slug:?} not found")))?; + let next_ts = next_timestamp(&head)?; + + // Verify all requested repos exist in the project. + let existing_coords: std::collections::HashSet = head + .tags + .iter() + .filter(|t| tag_name(t) == Some("a")) + .filter_map(|t| tag_value(t).map(String::from)) + .collect(); + for m in &to_remove { + if !existing_coords.contains(m.coord.as_str()) { + return Err(CliError::NotFound(format!( + "project {slug:?} does not contain member {:?}", + m.coord + ))); + } + } + + let remove_coords: std::collections::HashSet<&str> = + to_remove.iter().map(|m| m.coord.as_str()).collect(); + + // Keep all tags except auth and the removed members. + let tags: Vec = head + .tags + .iter() + .filter(|t| { + if tag_name(t) == Some("auth") { + return false; + } + if tag_name(t) == Some("a") { + if let Some(coord) = tag_value(t) { + return !remove_coords.contains(coord); + } + } + true + }) + .cloned() + .collect(); + + // Single rebuild validates the full envelope and strips any remaining auth. + let builder = rebuild_project(&head.content, tags, next_ts)?; + submit_project(client, builder).await +} + +/// `buzz projects update` +/// +/// Requires at least one setter or clearer; a no-op call is a usage error. +#[allow(clippy::too_many_arguments)] +pub async fn cmd_update( + client: &BuzzClient, + slug: &str, + name: Option<&str>, + clear_name: bool, + description: Option<&str>, + clear_description: bool, + channel: Option<&str>, + clear_channel: bool, + visibility: Option<&str>, + clear_visibility: bool, +) -> Result<(), CliError> { + // Guard: at least one mutation required. The clap `ArgGroup` with + // `required(true).multiple(true)` enforces this at parse time; this + // runtime check is a defense-in-depth safety net for callers that invoke + // `cmd_update` directly (e.g. tests and future programmatic callers). + let has_mutation = name.is_some() + || clear_name + || description.is_some() + || clear_description + || channel.is_some() + || clear_channel + || visibility.is_some() + || clear_visibility; + if !has_mutation { + return Err(CliError::Usage( + "buzz projects update requires at least one of: \ + --name, --clear-name, --description, --clear-description, \ + --channel, --clear-channel, --visibility, --clear-visibility" + .into(), + )); + } + + validate_project_slug(slug)?; + if let Some(ch) = channel { + crate::validate::validate_uuid(ch)?; + } + if let Some(vis) = visibility { + validate_visibility(vis)?; + } + + let head = fetch_own_project(client, slug) + .await? + .ok_or_else(|| CliError::NotFound(format!("project {slug:?} not found")))?; + let next_ts = next_timestamp(&head)?; + + // Build the new tag set. For each singleton metadata field: + // - setter present: replace value (strip old, append new) + // - clear flag set: drop the tag + // - neither: keep existing + // Non-singleton / non-metadata tags (d, a, unknown) are preserved as-is. + let singleton_fields = ["name", "description", "buzz-channel", "buzz-visibility"]; + let mut tags: Vec = head + .tags + .iter() + .filter(|t| { + if tag_name(t) == Some("auth") { + return false; + } + // Drop singletons we're replacing or clearing. + if let Some(field) = tag_name(t) { + if singleton_fields.contains(&field) { + let clear = match field { + "name" => clear_name || name.is_some(), + "description" => clear_description || description.is_some(), + "buzz-channel" => clear_channel || channel.is_some(), + "buzz-visibility" => clear_visibility || visibility.is_some(), + _ => false, + }; + return !clear; + } + } + true + }) + .cloned() + .collect(); + + // Append new singleton values. + if let Some(n) = name { + tags.push(make_tag(&["name", n])?); + } + if let Some(d) = description { + tags.push(make_tag(&["description", d])?); + } + if let Some(ch) = channel { + tags.push(make_tag(&["buzz-channel", ch])?); + } + if let Some(vis) = visibility { + tags.push(make_tag(&["buzz-visibility", vis])?); + } + + let builder = build_project_with_tags(&head.content, tags) + .map_err(|e| CliError::Other(format!("envelope validation failed: {e}")))? + .custom_created_at(next_ts); + submit_project(client, builder).await +} + +/// `buzz projects delete` +/// +/// Head-based and verified: +/// 1. Fetch own live head — `NotFound` if absent. +/// 2. Build tombstone at `head.created_at + 1`. +/// 3. Submit. +/// 4. Re-query the coordinate; if a newer head survived → `Conflict`. +pub async fn cmd_delete(client: &BuzzClient, slug: &str) -> Result<(), CliError> { + validate_project_slug(slug)?; + + let head = fetch_own_project(client, slug) + .await? + .ok_or_else(|| CliError::NotFound(format!("project {slug:?} not found")))?; + let next_ts = next_timestamp(&head)?; + + let pubkey_hex = client.keys().public_key().to_hex(); + let tombstone = build_delete_addressable(KIND_PROJECT, &pubkey_hex, slug) + .map_err(|e| CliError::Other(format!("failed to build delete event: {e}")))? + .custom_created_at(next_ts); + + let event = client.sign_event(tombstone)?; + let raw = client.submit_event(event).await?; + parse_write_response(&raw, "delete event was dominated; a newer head exists")?; + + // Post-submit verification: re-query to confirm the head is gone. + if let Some(survivor) = fetch_own_project(client, slug).await? { + // A newer head survived the tombstone. + return Err(CliError::Conflict(format!( + "project {slug:?} still exists (head at {}); a concurrent write raced the delete", + survivor.created_at.as_secs() + ))); + } + + println!("{}", serde_json::json!({ "deleted": slug, "status": "ok" })); + Ok(()) +} + +// ── Validation helpers ──────────────────────────────────────────────────────── + +/// Validate a project slug: non-empty, ≤1024 bytes, verbatim. +/// Does NOT impose the Buzz repo-ID grammar — project slugs are more permissive. +fn validate_project_slug(slug: &str) -> Result<(), CliError> { + if slug.is_empty() { + return Err(CliError::Usage("project slug must not be empty".into())); + } + if slug.len() > PROJECT_D_MAX_LEN { + return Err(CliError::Usage(format!( + "project slug must not exceed {PROJECT_D_MAX_LEN} bytes (got {})", + slug.len() + ))); + } + Ok(()) +} + +/// Validate a `buzz-visibility` value at the writer level. +fn validate_visibility(vis: &str) -> Result<(), CliError> { + if vis != "listed" && vis != "unlisted" { + return Err(CliError::Usage(format!( + "visibility must be 'listed' or 'unlisted' (got {vis:?})" + ))); + } + Ok(()) +} + +// ── Dispatch ────────────────────────────────────────────────────────────────── + +pub async fn dispatch(cmd: crate::ProjectsCmd, client: &BuzzClient) -> Result<(), CliError> { + use crate::ProjectsCmd; + match cmd { + ProjectsCmd::Create { + slug, + repo, + name, + description, + channel, + visibility, + } => { + cmd_create( + client, + &slug, + &repo, + name.as_deref(), + description.as_deref(), + channel.as_deref(), + visibility.map(|v| v.as_str()), + ) + .await + } + ProjectsCmd::Get { slug, owner } => cmd_get(client, &slug, owner.as_deref()).await, + ProjectsCmd::List { owner, limit } => cmd_list(client, owner.as_deref(), limit).await, + ProjectsCmd::AddRepo { slug, repo } => cmd_add_repo(client, &slug, &repo).await, + ProjectsCmd::RemoveRepo { slug, repo } => cmd_remove_repo(client, &slug, &repo).await, + ProjectsCmd::Update { + slug, + name, + clear_name, + description, + clear_description, + channel, + clear_channel, + visibility, + clear_visibility, + } => { + cmd_update( + client, + &slug, + name.as_deref(), + clear_name, + description.as_deref(), + clear_description, + channel.as_deref(), + clear_channel, + visibility.map(|v| v.as_str()), + clear_visibility, + ) + .await + } + ProjectsCmd::Delete { slug } => cmd_delete(client, &slug).await, + } +} + +// ── Tests ───────────────────────────────────────────────────────────────────── + +#[cfg(test)] +mod tests { + use buzz_sdk::{validate_project_envelope, PROJECT_MEMBER_CAP}; + use nostr::Tag; + + use super::*; + + // ── Coordinate expansion ────────────────────────────────────────────────── + + const OWNER_HEX: &str = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"; + const OWNER_B_HEX: &str = "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"; + + #[test] + fn expand_repo_coord_bare_expands_with_caller_pubkey() { + let coord = expand_repo_coord("my-repo", OWNER_HEX).unwrap(); + assert_eq!(coord.coord, format!("30617:{OWNER_HEX}:my-repo")); + } + + #[test] + fn expand_repo_coord_full_passes_through() { + let full = format!("30617:{OWNER_HEX}:some-repo"); + let coord = expand_repo_coord(&full, OWNER_B_HEX).unwrap(); + // Owner from the full coord, not the caller. + assert_eq!(coord.coord, full); + } + + #[test] + fn expand_repo_coord_full_cross_owner() { + let full = format!("30617:{OWNER_B_HEX}:infra"); + let coord = expand_repo_coord(&full, OWNER_HEX).unwrap(); + assert_eq!(coord.coord, full); + } + + #[test] + fn expand_repo_coord_rejects_uppercase_owner() { + let upper = "30617:AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA:buzz"; + assert!(expand_repo_coord(upper, OWNER_HEX).is_err()); + } + + #[test] + fn expand_repo_coord_rejects_coordinate_shaped_bare_value() { + // A value with a colon is never a bare id. + let not_bare = "30617:something"; + // parse_full will fail because it's not a valid full coordinate either. + assert!(expand_repo_coord(not_bare, OWNER_HEX).is_err()); + } + + // ── validate_project_slug ───────────────────────────────────────────────── + + #[test] + fn validate_project_slug_accepts_normal() { + assert!(validate_project_slug("my-project").is_ok()); + assert!(validate_project_slug("platform:v2").is_ok()); // colons allowed — more permissive than repo-id + } + + #[test] + fn validate_project_slug_rejects_empty() { + assert!(validate_project_slug("").is_err()); + } + + #[test] + fn validate_project_slug_rejects_over_1024() { + let long = "a".repeat(1025); + assert!(validate_project_slug(&long).is_err()); + } + + #[test] + fn validate_project_slug_accepts_1024() { + let at_limit = "a".repeat(1024); + assert!(validate_project_slug(&at_limit).is_ok()); + } + + // ── validate_visibility ─────────────────────────────────────────────────── + + #[test] + fn validate_visibility_accepts_listed_and_unlisted() { + assert!(validate_visibility("listed").is_ok()); + assert!(validate_visibility("unlisted").is_ok()); + } + + #[test] + fn validate_visibility_rejects_unknown_token() { + assert!(validate_visibility("chartreuse").is_err()); + assert!(validate_visibility("").is_err()); + } + + // ── is_bare_repo_id ─────────────────────────────────────────────────────── + + #[test] + fn bare_repo_id_accepts_valid() { + assert!(is_bare_repo_id("buzz")); + assert!(is_bare_repo_id("my-repo_1.0")); + } + + #[test] + fn bare_repo_id_rejects_colon() { + assert!(!is_bare_repo_id("30617:something")); + assert!(!is_bare_repo_id("has:colon")); + } + + #[test] + fn bare_repo_id_rejects_empty() { + assert!(!is_bare_repo_id("")); + } + + #[test] + fn bare_repo_id_rejects_over_64() { + let long = "a".repeat(65); + assert!(!is_bare_repo_id(&long)); + } + + // ── tag helpers ─────────────────────────────────────────────────────────── + + fn make_test_tag(parts: &[&str]) -> Tag { + Tag::parse(parts.iter().copied()).unwrap() + } + + // ── rebuild_project: hinted / unknown tag preservation ─────────────────── + + #[test] + fn rebuild_project_preserves_hinted_member_tags() { + // A member 'a' tag with a relay hint must survive RMW untouched. + let coord = format!("30617:{OWNER_HEX}:buzz"); + let hint = "wss://relay.example.com"; + let tags = vec![ + make_test_tag(&["d", "platform"]), + Tag::parse(["a", &coord, hint]).unwrap(), + ]; + let ts = Timestamp::from(1_700_000_001u64); + let b = rebuild_project("", tags, ts).unwrap(); + let ev = b.sign_with_keys(&nostr::Keys::generate()).expect("sign"); + let a_tag = ev + .tags + .iter() + .find(|t| tag_name(t) == Some("a")) + .expect("a tag present"); + assert_eq!( + a_tag.as_slice(), + &["a".to_string(), coord, hint.to_string()], + "relay hint must survive rebuild" + ); + } + + #[test] + fn rebuild_project_preserves_unknown_tags() { + let tags = vec![ + make_test_tag(&["d", "platform"]), + make_test_tag(&["future-metadata", "value"]), + ]; + let ts = Timestamp::from(1_700_000_001u64); + let b = rebuild_project("", tags, ts).unwrap(); + let ev = b.sign_with_keys(&nostr::Keys::generate()).expect("sign"); + assert!(ev + .tags + .iter() + .any(|t| tag_name(t) == Some("future-metadata"))); + } + + #[test] + fn rebuild_project_strips_auth_tag() { + let tags = vec![ + make_test_tag(&["d", "platform"]), + make_test_tag(&["auth", &"a".repeat(64), "kind=30617", &"b".repeat(128)]), + ]; + let ts = Timestamp::from(1_700_000_001u64); + let b = rebuild_project("", tags, ts).unwrap(); + let ev = b.sign_with_keys(&nostr::Keys::generate()).expect("sign"); + assert!( + !ev.tags.iter().any(|t| tag_name(t) == Some("auth")), + "auth tag must be stripped" + ); + } + + #[test] + fn rebuild_project_rejects_over_cap_foreign_head() { + // A foreign head with 65 members must fail Layer A on republish. + let mut tags = vec![make_test_tag(&["d", "wide"])]; + for i in 0..=64u32 { + let coord = format!("30617:{OWNER_HEX}:repo-{i:02}"); + tags.push(make_test_tag(&["a", &coord])); + } + assert_eq!( + tags.iter().filter(|t| tag_name(t) == Some("a")).count(), + 65, + "65 a-tags" + ); + let ts = Timestamp::from(1_700_000_001u64); + // rebuild_project strips auth, but 65 a-tags still exceeds cap. + assert!( + rebuild_project("", tags, ts).is_err(), + "over-cap foreign head must fail rebuild" + ); + } + + #[test] + fn rebuild_project_at_exact_cap_succeeds() { + let mut tags = vec![make_test_tag(&["d", "wide"])]; + for i in 0..PROJECT_MEMBER_CAP { + let coord = format!("30617:{OWNER_HEX}:repo-{i:02}"); + tags.push(make_test_tag(&["a", &coord])); + } + let ts = Timestamp::from(1_700_000_001u64); + assert!(rebuild_project("", tags, ts).is_ok()); + } + + // ── clear-flag semantics ────────────────────────────────────────────────── + + /// Build a minimal head Event for testing update semantics without the relay. + fn make_head_tags(extra: &[Tag]) -> Vec { + let mut tags = vec![make_test_tag(&["d", "platform"])]; + tags.extend_from_slice(extra); + tags + } + + #[allow(clippy::too_many_arguments)] + fn apply_update_tags( + head_tags: Vec, + name: Option<&str>, + clear_name: bool, + description: Option<&str>, + clear_description: bool, + channel: Option<&str>, + clear_channel: bool, + visibility: Option<&str>, + clear_visibility: bool, + ) -> Vec { + // Replicate the tag-mutation logic from cmd_update (sans relay I/O). + let singleton_fields = ["name", "description", "buzz-channel", "buzz-visibility"]; + let mut tags: Vec = head_tags + .iter() + .filter(|t| { + if tag_name(t) == Some("auth") { + return false; + } + if let Some(field) = tag_name(t) { + if singleton_fields.contains(&field) { + let clear = match field { + "name" => clear_name || name.is_some(), + "description" => clear_description || description.is_some(), + "buzz-channel" => clear_channel || channel.is_some(), + "buzz-visibility" => clear_visibility || visibility.is_some(), + _ => false, + }; + return !clear; + } + } + true + }) + .cloned() + .collect(); + if let Some(n) = name { + tags.push(make_test_tag(&["name", n])); + } + if let Some(d) = description { + tags.push(make_test_tag(&["description", d])); + } + if let Some(ch) = channel { + tags.push(make_test_tag(&["buzz-channel", ch])); + } + if let Some(vis) = visibility { + tags.push(make_test_tag(&["buzz-visibility", vis])); + } + tags + } + + #[test] + fn update_omission_preserves_existing_field() { + let head = make_head_tags(&[make_test_tag(&["name", "Old Name"])]); + let result = apply_update_tags(head, None, false, None, false, None, false, None, false); + assert!(result.iter().any(|t| tag_value(t) == Some("Old Name"))); + } + + #[test] + fn update_setter_replaces_existing_field() { + let head = make_head_tags(&[make_test_tag(&["name", "Old Name"])]); + let result = apply_update_tags( + head, + Some("New Name"), + false, + None, + false, + None, + false, + None, + false, + ); + assert!(result.iter().any(|t| tag_value(t) == Some("New Name"))); + assert!(!result.iter().any(|t| tag_value(t) == Some("Old Name"))); + } + + #[test] + fn update_clear_drops_existing_field() { + let head = make_head_tags(&[make_test_tag(&["name", "Old Name"])]); + let result = apply_update_tags(head, None, true, None, false, None, false, None, false); + assert!(!result.iter().any(|t| tag_name(t) == Some("name"))); + } + + #[test] + fn update_clear_visibility_drops_tag() { + let head = make_head_tags(&[make_test_tag(&["buzz-visibility", "unlisted"])]); + let result = apply_update_tags(head, None, false, None, false, None, false, None, true); + assert!(!result + .iter() + .any(|t| tag_name(t) == Some("buzz-visibility"))); + } + + #[test] + fn update_exactly_one_singleton_after_replace() { + // Start with a buzz-channel; replace with a new one; must have exactly one. + let uuid1 = "3580ca9b-47b4-4af9-b22a-1068778f26c6"; + let uuid2 = "00000000-0000-0000-0000-000000000000"; + let head = make_head_tags(&[make_test_tag(&["buzz-channel", uuid1])]); + let result = apply_update_tags( + head, + None, + false, + None, + false, + Some(uuid2), + false, + None, + false, + ); + let channels: Vec<_> = result + .iter() + .filter(|t| tag_name(t) == Some("buzz-channel")) + .collect(); + assert_eq!(channels.len(), 1); + assert_eq!(tag_value(channels[0]), Some(uuid2)); + } + + // ── duplicate-member rejection on republish ─────────────────────────────── + + #[test] + fn duplicate_member_in_foreign_head_fails_rebuild() { + let coord = format!("30617:{OWNER_HEX}:buzz"); + let tags = vec![ + make_test_tag(&["d", "platform"]), + make_test_tag(&["a", &coord]), + make_test_tag(&["a", &coord]), // duplicate + ]; + let ts = Timestamp::from(1_700_000_001u64); + assert!(rebuild_project("", tags, ts).is_err()); + } + + // ── validate_project_envelope integration ──────────────────────────────── + + #[test] + fn validate_project_envelope_accepts_hinted_member() { + let coord = format!("30617:{OWNER_HEX}:buzz"); + let tags = vec![ + make_test_tag(&["d", "platform"]), + Tag::parse(["a", &coord, "wss://relay.example.com"]).unwrap(), + ]; + assert!(validate_project_envelope(&tags, "").is_ok()); + } + + #[test] + fn validate_project_envelope_rejects_four_element_member() { + let coord = format!("30617:{OWNER_HEX}:buzz"); + let tags = vec![ + make_test_tag(&["d", "platform"]), + Tag::parse(["a", &coord, "wss://relay.example.com", "extra"]).unwrap(), + ]; + assert!(validate_project_envelope(&tags, "").is_err()); + } + + // ── next_timestamp ordering ─────────────────────────────────────────────── + + /// `next_timestamp` must return `head.created_at + 1` regardless of the wall + /// clock. NIP-MP Deletion rule: a tombstone older than the live head does + /// NOT remove it, so we must advance strictly off the observed head — never + /// use wall-clock time, which could be behind a head that was bumped + /// multiple times in the same second. + #[test] + fn next_timestamp_returns_head_plus_one_when_head_is_ahead_of_wall_clock() { + // Build a minimal signed event with a created_at far in the future. + let keys = nostr::Keys::generate(); + let far_future_ts = Timestamp::from(9_999_999_999u64); // year 2286 + let tags = vec![ + make_test_tag(&["d", "platform"]), + make_test_tag(&["a", &format!("30617:{OWNER_HEX}:buzz")]), + ]; + let builder = rebuild_project("", tags, far_future_ts).expect("valid head envelope"); + let head = builder.sign_with_keys(&keys).expect("sign"); + // Verify the event actually has our future timestamp. + assert_eq!(head.created_at, far_future_ts); + + // next_timestamp must return far_future + 1, not now(). + let next = next_timestamp(&head).expect("no overflow"); + assert_eq!( + next.as_secs(), + far_future_ts.as_secs() + 1, + "tombstone must be strictly after head, even when head is far in the future" + ); + } + + // ── empty update guard ──────────────────────────────────────────────────── + + /// `cmd_update` with no setters or clearers must return `CliError::Usage` + /// before making any network call. The guard is synchronous (before the + /// first `.await`) so we can drive it with a dummy client whose address + /// would reject any real connection attempt. + #[tokio::test] + async fn empty_update_returns_usage_error_before_any_network_call() { + let keys = nostr::Keys::generate(); + // Port 9 is the discard protocol — any real connect will be refused + // immediately, but the guard fires before the first await so this + // never reaches the network. + let client = crate::client::BuzzClient::new("http://127.0.0.1:9".into(), keys, None, None) + .expect("client construction"); + + let err = cmd_update( + &client, "my-slug", None, false, // name / clear_name + None, false, // description / clear_description + None, false, // channel / clear_channel + None, false, // visibility / clear_visibility + ) + .await + .expect_err("empty update must fail"); + + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage, got {err:?}" + ); + } + + // ── no-network malformed-input tests ───────────────────────────────────── + // + // All three cases use port 9 (discard protocol): any real connection is + // refused immediately, but local validation fires before the first .await + // so the network is never touched. + + fn discard_client() -> crate::client::BuzzClient { + let keys = nostr::Keys::generate(); + crate::client::BuzzClient::new("http://127.0.0.1:9".into(), keys, None, None) + .expect("client construction") + } + + /// Invalid visibility token must return Usage before touching the relay. + #[tokio::test] + async fn create_invalid_visibility_returns_usage_before_any_network_call() { + let client = discard_client(); + let err = cmd_create( + &client, + "my-slug", + &["buzz".to_string()], + None, + None, + None, + Some("chartreuse"), + ) + .await + .expect_err("invalid visibility must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for invalid visibility, got {err:?}" + ); + } + + /// A name longer than 256 bytes must return Usage before touching the relay. + #[tokio::test] + async fn create_overlong_name_returns_usage_before_any_network_call() { + let client = discard_client(); + let long_name = "a".repeat(257); + let err = cmd_create( + &client, + "my-slug", + &["buzz".to_string()], + Some(&long_name), + None, + None, + None, + ) + .await + .expect_err("overlong name must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for overlong name, got {err:?}" + ); + } + + /// A malformed --repo coordinate must return Usage before touching the relay. + #[tokio::test] + async fn create_malformed_repo_returns_usage_before_any_network_call() { + let client = discard_client(); + let err = cmd_create( + &client, + "my-slug", + &["nope:bad".to_string()], + None, + None, + None, + None, + ) + .await + .expect_err("malformed repo must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for malformed repo, got {err:?}" + ); + } + + /// A malformed --repo coordinate on add-repo must return Usage before touching the relay. + #[tokio::test] + async fn add_repo_malformed_coord_returns_usage_before_any_network_call() { + let client = discard_client(); + let err = cmd_add_repo(&client, "my-slug", &["nope:bad".to_string()]) + .await + .expect_err("malformed repo must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for malformed repo on add-repo, got {err:?}" + ); + } + + /// A malformed --repo coordinate on remove-repo must return Usage before touching the relay. + #[tokio::test] + async fn remove_repo_malformed_coord_returns_usage_before_any_network_call() { + let client = discard_client(); + let err = cmd_remove_repo(&client, "my-slug", &["nope:bad".to_string()]) + .await + .expect_err("malformed repo must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for malformed repo on remove-repo, got {err:?}" + ); + } + + // ── duplicate --repo within one invocation ──────────────────────────────── + + /// Supplying the same coordinate twice in one create call must return Usage + /// (names the duplicate) before any network call. + #[tokio::test] + async fn create_duplicate_repo_returns_usage_before_any_network_call() { + let client = discard_client(); + let coord = format!("30617:{OWNER_HEX}:buzz"); + let err = cmd_create( + &client, + "my-slug", + &[coord.clone(), coord.clone()], + None, + None, + None, + None, + ) + .await + .expect_err("duplicate repo must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for duplicate repo, got {err:?}" + ); + // Error message must name the duplicate coordinate. + assert!( + format!("{err}").contains("buzz"), + "Usage message must name the duplicate coordinate, got {err:?}" + ); + } + + /// Supplying the same coordinate twice in one add-repo call must return Usage + /// (names the duplicate) before any network call. + #[tokio::test] + async fn add_repo_duplicate_coord_returns_usage_before_any_network_call() { + let client = discard_client(); + let coord = format!("30617:{OWNER_HEX}:buzz"); + let err = cmd_add_repo(&client, "my-slug", &[coord.clone(), coord.clone()]) + .await + .expect_err("duplicate repo must fail"); + assert!( + matches!(err, CliError::Usage(_)), + "expected CliError::Usage for duplicate repo on add-repo, got {err:?}" + ); + } + + // ── create collision guard ──────────────────────────────────────────────── + + // The create-collision Conflict path is pinned by the live transcript + // (step: duplicate create → Conflict, exit=5). No relay mock is available + // for a unit test; the no-network tests above cover all pre-await paths. + + // ── add-repo no-op guard ────────────────────────────────────────────────── + + // The add-repo no-op Conflict path is pinned by the live transcript + // (step 7: buzz already present → exit=5). No relay mock is available + // for a unit test; the async no-network tests above cover all pre-await paths. +} diff --git a/crates/buzz-cli/src/commands/repos.rs b/crates/buzz-cli/src/commands/repos.rs index 608d495055..15e064d9c3 100644 --- a/crates/buzz-cli/src/commands/repos.rs +++ b/crates/buzz-cli/src/commands/repos.rs @@ -4,7 +4,8 @@ use buzz_core::{ }; use nostr::{Event, EventBuilder, Tag, Timestamp}; -use crate::client::{normalize_write_response, BuzzClient}; +use crate::client::BuzzClient; +use crate::commands::parse_write_response; use crate::error::CliError; use crate::validate::validate_repo_id; @@ -186,25 +187,10 @@ fn protection_rules_json(event: &Event) -> Result { } fn validate_write_response(raw: &str) -> Result { - let response: serde_json::Value = serde_json::from_str(raw) - .map_err(|error| CliError::Other(format!("relay response is not JSON: {error} ({raw})")))?; - let accepted = response - .get("accepted") - .and_then(serde_json::Value::as_bool) - .unwrap_or(false); - let message = response - .get("message") - .and_then(serde_json::Value::as_str) - .unwrap_or(""); - if !accepted { - return Err(CliError::Other(format!("relay rejected event: {message}"))); - } - if message == "duplicate" || message.starts_with("duplicate:") { - return Err(CliError::Conflict( - "repository changed concurrently; fetch the latest rules and retry".into(), - )); - } - Ok(normalize_write_response(raw)) + parse_write_response( + raw, + "repository changed concurrently; fetch the latest rules and retry", + ) } async fn submit_repo_update(client: &BuzzClient, builder: EventBuilder) -> Result<(), CliError> { diff --git a/crates/buzz-cli/src/lib.rs b/crates/buzz-cli/src/lib.rs index 0726406d29..f745e7b280 100644 --- a/crates/buzz-cli/src/lib.rs +++ b/crates/buzz-cli/src/lib.rs @@ -212,6 +212,9 @@ enum Cmd { /// Announce and discover git repositories (NIP-34) #[command(subcommand)] Repos(ReposCmd), + /// Create and manage multi-repo projects (NIP-MP) + #[command(subcommand)] + Projects(ProjectsCmd), /// Send, get, list, and set status on git patches (NIP-34) #[command(subcommand)] Patches(PatchesCmd), @@ -1227,6 +1230,122 @@ pub enum RepoPushRole { Member, } +/// Visibility of a multi-repo project listing. +#[derive(Clone, Copy, Debug, clap::ValueEnum)] +pub enum ProjectVisibility { + /// Project appears in public listings (default). + Listed, + /// Project is hidden from public listings. + Unlisted, +} + +impl ProjectVisibility { + pub fn as_str(self) -> &'static str { + match self { + ProjectVisibility::Listed => "listed", + ProjectVisibility::Unlisted => "unlisted", + } + } +} + +#[derive(Subcommand)] +pub enum ProjectsCmd { + /// Create a new multi-repo project (NIP-MP kind:30621) + /// + /// Requires at least one --repo. Fails with Conflict if the project already exists. + Create { + /// Project identifier (slug), up to 1024 bytes + slug: String, + /// Member repository coordinate: bare Buzz repo id (e.g. `buzz`) or full + /// `30617::` for cross-owner or colon-bearing repo ids. + /// At least one --repo is required. + #[arg(long = "repo", required = true)] + repo: Vec, + /// Display name (≤256 bytes) + #[arg(long)] + name: Option, + /// Description (≤2048 bytes) + #[arg(long)] + description: Option, + /// Associated Buzz channel UUID + #[arg(long)] + channel: Option, + /// Visibility: `listed` (default) or `unlisted` + #[arg(long)] + visibility: Option, + }, + /// Get a project by slug + Get { + /// Project slug + slug: String, + /// Owner pubkey (64-char hex). Defaults to the current identity. + #[arg(long)] + owner: Option, + }, + /// List projects + List { + /// Owner pubkey (64-char hex). Defaults to the current identity. + #[arg(long)] + owner: Option, + /// Maximum number of results + #[arg(long)] + limit: Option, + }, + /// Add one or more member repositories to a project + #[command(name = "add-repo")] + AddRepo { + /// Project slug + slug: String, + /// Member repository coordinate (bare id or full `30617::`) + #[arg(long = "repo", required = true)] + repo: Vec, + }, + /// Remove one or more member repositories from a project + #[command(name = "remove-repo")] + RemoveRepo { + /// Project slug + slug: String, + /// Member repository coordinate to remove (bare id or full `30617::`) + #[arg(long = "repo", required = true)] + repo: Vec, + }, + /// Update project metadata (at least one setter or clearer required) + #[command(group = clap::ArgGroup::new("mutation").required(true).multiple(true))] + Update { + /// Project slug + slug: String, + /// Set the display name + #[arg(long, group = "mutation")] + name: Option, + /// Remove the display name + #[arg(long, group = "mutation", conflicts_with = "name")] + clear_name: bool, + /// Set the description + #[arg(long, group = "mutation")] + description: Option, + /// Remove the description + #[arg(long, group = "mutation", conflicts_with = "description")] + clear_description: bool, + /// Set the associated Buzz channel UUID + #[arg(long, group = "mutation")] + channel: Option, + /// Remove the associated channel + #[arg(long, group = "mutation", conflicts_with = "channel")] + clear_channel: bool, + /// Set visibility: `listed` or `unlisted` + #[arg(long, group = "mutation")] + visibility: Option, + /// Remove the visibility tag (absence defaults to `listed`) + #[arg(long, group = "mutation", conflicts_with = "visibility")] + clear_visibility: bool, + }, + /// Delete a project (head-based tombstone; verified after submit) + Delete { + /// Project slug + slug: String, + }, +} + #[derive(Subcommand)] pub enum PatchesCmd { /// Send a git patch (NIP-34 kind:1617) @@ -1768,6 +1887,41 @@ pub enum ModerationCmd { }, } +/// Normalize hand-authored `BUZZ_AUTH_TAG` input to strict JSON. +/// +/// `.env` files and shell exports sometimes carry the tag in the unquoted +/// shorthand `[auth,,,]` (quotes dropped by hand). +/// When the input is not valid JSON but is bracket-delimited, rewrite it as +/// a JSON array of the comma-separated fields (an empty field `,,` becomes +/// `""`, matching the canonical form `["auth","hex","","hex"]`). +/// +/// This is presentation-layer leniency at the configuration edge only: the +/// output is always fed through the SDK's strict `parse_auth_tag` / +/// `verify_auth_tag`, which enforce structure, hex, the conditions grammar, +/// and the BIP-340 signature. Inputs that are already valid JSON — or not +/// recognizable as the shorthand — are returned unchanged so the strict +/// parser reports the error on the original bytes. +fn normalize_auth_tag_input(input: &str) -> String { + let trimmed = input.trim(); + if serde_json::from_str::(trimmed).is_ok() { + return trimmed.to_owned(); + } + if trimmed.starts_with('[') && trimmed.ends_with(']') { + let fields: Vec<&str> = trimmed[1..trimmed.len() - 1] + .split(',') + .map(str::trim) + .collect(); + // Only a plausible 4-field auth tag is rewritten; anything else is + // passed through untouched for the strict parser to reject with an + // error that references the caller's original input. + if fields.len() == 4 && !fields.iter().any(|f| f.contains('"')) { + // serde_json cannot fail serializing a Vec<&str>. + return serde_json::to_string(&fields).expect("string array serializes"); + } + } + trimmed.to_owned() +} + async fn run(cli: Cli) -> Result<(), CliError> { let relay_url = client::normalize_relay_url(&cli.relay); @@ -1788,17 +1942,28 @@ async fn run(cli: Cli) -> Result<(), CliError> { .map_err(|e| CliError::Key(format!("invalid BUZZ_PRIVATE_KEY: {e}")))?; // NIP-OA: parse and verify the auth tag if provided. + // + // `BUZZ_AUTH_TAG` is hand-authored configuration, so the unquoted raw + // shorthand `[auth,hex,,hex]` is normalized to JSON here — at this input + // edge only. The SDK grammar and the `x-auth-tag` wire format stay strict + // JSON; all validation and signature verification happen on the strict + // path below, unchanged. let (auth_tag, auth_tag_json) = match cli.auth_tag { - Some(ref json) if !json.is_empty() => { - let tag = buzz_sdk::nip_oa::parse_auth_tag(json) + Some(ref input) if !input.is_empty() => { + let json = normalize_auth_tag_input(input); + let tag = buzz_sdk::nip_oa::parse_auth_tag(&json) .map_err(|e| CliError::Auth(format!("BUZZ_AUTH_TAG is malformed: {e}")))?; - buzz_sdk::nip_oa::verify_auth_tag(json, &keys.public_key()).map_err(|e| { + buzz_sdk::nip_oa::verify_auth_tag(&json, &keys.public_key()).map_err(|e| { CliError::Auth(format!( "BUZZ_AUTH_TAG verification failed for pubkey {}: {e}", keys.public_key().to_hex() )) })?; - (Some(tag), Some(json.clone())) + // Canonical wire form derives from the parsed-and-verified tag + // (same shape as buzz-acp's RestClient), never from raw input. + let canonical = serde_json::to_string(tag.as_slice()) + .map_err(|e| CliError::Auth(format!("BUZZ_AUTH_TAG serialization failed: {e}")))?; + (Some(tag), Some(canonical)) } _ => (None, None), }; @@ -1819,6 +1984,7 @@ async fn run(cli: Cli) -> Result<(), CliError> { Cmd::Social(sub) => commands::social::dispatch(sub, &client).await, Cmd::Notes(sub) => commands::notes::dispatch(sub, &client).await, Cmd::Repos(sub) => commands::repos::dispatch(sub, &client).await, + Cmd::Projects(sub) => commands::projects::dispatch(sub, &client).await, Cmd::Patches(sub) => commands::patches::dispatch(sub, &client).await, Cmd::Issues(sub) => commands::issues::dispatch(sub, &client).await, Cmd::Pr(sub) => commands::pr::dispatch(sub, &client).await, @@ -1835,6 +2001,51 @@ mod tests { use super::*; use clap::CommandFactory; + /// Raw shorthand `[auth,hex,,hex]` normalizes to strict JSON; the empty + /// conditions field becomes `""`. + #[test] + fn normalize_auth_tag_raw_shorthand() { + let owner = "a".repeat(64); + let sig = "b".repeat(128); + + let raw = format!("[auth,{owner},,{sig}]"); + let json = normalize_auth_tag_input(&raw); + let parsed: Vec = serde_json::from_str(&json).expect("output must be JSON"); + assert_eq!(parsed, vec!["auth", &owner, "", &sig]); + + // With conditions and surrounding whitespace (shell/.env artifacts). + let raw = format!(" [auth, {owner} , kind=9, {sig}] \n"); + let json = normalize_auth_tag_input(&raw); + let parsed: Vec = serde_json::from_str(&json).expect("output must be JSON"); + assert_eq!(parsed, vec!["auth", &owner, "kind=9", &sig]); + } + + /// Valid JSON input passes through byte-identical (modulo outer trim) — + /// the normalizer must never rewrite well-formed input. + #[test] + fn normalize_auth_tag_json_passthrough() { + let owner = "a".repeat(64); + let sig = "b".repeat(128); + let json_in = serde_json::json!(["auth", owner, "kind=9", sig]).to_string(); + assert_eq!(normalize_auth_tag_input(&json_in), json_in); + } + + /// Inputs that are neither JSON nor a plausible 4-field shorthand pass + /// through unchanged, so the strict parser rejects the original bytes. + #[test] + fn normalize_auth_tag_leaves_garbage_untouched() { + for garbage in [ + "not a tag", + "[auth,too,few]", + "[a,b,c,d,e]", + r#"[auth,"quoted",x,y]"#, // quote chars => not the shorthand + "[]", + "{\"auth\":1}", + ] { + assert_eq!(normalize_auth_tag_input(garbage), garbage.trim()); + } + } + /// Smoke test: CLI definition is valid and parseable. #[test] fn cli_definition_is_valid() { @@ -1883,6 +2094,7 @@ mod tests { "pack", "patches", "pr", + "projects", "reactions", "repos", "social", @@ -2038,6 +2250,18 @@ mod tests { names(&cmd, "patches"), vec!["get", "list", "send", "status"] ); + assert_eq!( + names(&cmd, "projects"), + vec![ + "add-repo", + "create", + "delete", + "get", + "list", + "remove-repo", + "update" + ] + ); assert_eq!( names(&cmd, "issues"), vec!["create", "get", "list", "status"] @@ -2075,6 +2299,7 @@ mod tests { ("pack", 2), ("patches", 4), ("pr", 5), + ("projects", 7), ("reactions", 3), ("repos", 5), ("social", 7), @@ -2142,4 +2367,111 @@ mod tests { .join("\n") ); } + + // ── projects update mutation group ──────────────────────────────────────── + + /// Multiple independent fields must be accepted in the same invocation. + #[test] + fn projects_update_multi_field_is_accepted() { + assert!( + Cli::try_parse_from([ + "buzz", + "projects", + "update", + "my-slug", + "--name", + "X", + "--description", + "Y", + ]) + .is_ok(), + "--name and --description together must be accepted" + ); + } + + /// A setter for one field and a clearer for a different field must be accepted. + #[test] + fn projects_update_setter_with_other_clearer_is_accepted() { + assert!( + Cli::try_parse_from([ + "buzz", + "projects", + "update", + "my-slug", + "--name", + "X", + "--clear-description", + ]) + .is_ok(), + "--name with --clear-description must be accepted" + ); + } + + /// A setter and its own clearer are mutually exclusive — clap must reject this. + #[test] + fn projects_update_setter_with_own_clearer_is_rejected() { + assert!( + Cli::try_parse_from([ + "buzz", + "projects", + "update", + "my-slug", + "--name", + "X", + "--clear-name", + ]) + .is_err(), + "--name and --clear-name together must be rejected by clap" + ); + } + + /// Providing no mutation options at all must be rejected by clap (required group). + #[test] + fn projects_update_no_mutation_is_rejected_by_clap() { + // Without credentials, a valid parse would reach authentication and fail + // with auth_error — but a clap-level rejection happens before any I/O. + // We verify it's a clap error (not just any error) by checking the error + // kind is not a runtime/auth failure — Cli::try_parse_from returns Err + // immediately for argument violations. + assert!( + Cli::try_parse_from(["buzz", "projects", "update", "my-slug"]).is_err(), + "update with no setters or clearers must be rejected at parse time" + ); + } + + /// An unrecognised visibility token must be rejected by clap before any I/O. + #[test] + fn projects_create_invalid_visibility_is_rejected_by_clap() { + assert!( + Cli::try_parse_from([ + "buzz", + "projects", + "create", + "my-slug", + "--repo", + "buzz", + "--visibility", + "chartreuse", + ]) + .is_err(), + "--visibility chartreuse must be rejected at parse time" + ); + } + + /// An unrecognised visibility token on update must be rejected by clap before any I/O. + #[test] + fn projects_update_invalid_visibility_is_rejected_by_clap() { + assert!( + Cli::try_parse_from([ + "buzz", + "projects", + "update", + "my-slug", + "--visibility", + "chartreuse", + ]) + .is_err(), + "--visibility chartreuse on update must be rejected at parse time" + ); + } } diff --git a/crates/buzz-db/src/lib.rs b/crates/buzz-db/src/lib.rs index 245e49bb2d..9b26876747 100644 --- a/crates/buzz-db/src/lib.rs +++ b/crates/buzz-db/src/lib.rs @@ -3999,8 +3999,33 @@ impl Db { } /// Returns `true` if `pubkey` (64-char hex) is a member of `community`. + /// + /// Replica-routed on the bounded arm — the one PERMISSION read routed by + /// explicit product decision (bounded-stale membership beats the 10s + /// cache it replaced). Admits and revokes may lag by at most the budget + /// `B`; everything else fails closed to the writer, exactly like + /// [`Db::query_events_routed_bounded`]. Not precedent for routing other + /// permission reads. pub async fn is_relay_member(&self, community: CommunityId, pubkey: &str) -> Result { - relay_members::is_relay_member(&self.pool, community, pubkey).await + let path = "relay_membership"; + match self.route_read(path, RoutePredicate::Bounded).await { + RouteDecision::Replica(mut tx, _entry, reason) => { + match relay_members::is_relay_member_on(&mut tx, community, pubkey).await { + Ok(is_member) => { + Self::record_route(path, "replica", reason); + Ok(is_member) + } + Err(e) => { + tracing::warn!(path, "replica read failed; re-running on writer: {e}"); + Self::record_route(path, "writer", "replica_error"); + relay_members::is_relay_member(&self.pool, community, pubkey).await + } + } + } + RouteDecision::Writer => { + relay_members::is_relay_member(&self.pool, community, pubkey).await + } + } } /// Returns the relay member record for `pubkey` in `community`, or `None` if not found. @@ -4096,6 +4121,12 @@ impl Db { relay_members::bootstrap_owner(&self.pool, community, owner_pubkey).await } + /// Returns `true` if any member of `community` holds the `admin` or + /// `owner` role. + pub async fn has_admin_or_owner(&self, community: CommunityId) -> Result { + relay_members::has_admin_or_owner(&self.pool, community).await + } + /// Atomically transfers ownership of `community` to `new_owner_pubkey`, /// demoting the previous owner(s) to `member`. Verifies /// `expected_owner_pubkey` matches the current owner inside the same @@ -7408,6 +7439,78 @@ mod tests { drop_scratch_db(&admin, writer, &wname).await; } + /// Routed relay-membership check: budget unset ⇒ writer; budget set + + /// fresh proved entry ⇒ replica (bounded arm); over-budget entry ⇒ + /// writer. Divergent membership rows prove which pool answered. + #[tokio::test] + #[ignore = "requires Postgres"] + async fn is_relay_member_is_bounded_routed_and_fails_closed() { + let admin = PgPool::connect(&admin_url().await) + .await + .expect("connect admin"); + let (writer, wname) = create_scratch_db(&admin, "mem_w").await; + let (replica, rname) = create_scratch_db(&admin, "mem_r").await; + + let community = Uuid::new_v4(); + for pool in [&writer, &replica] { + sqlx::query("INSERT INTO communities (id, host) VALUES ($1, $2)") + .bind(community) + .bind(format!("member-routing-{}.example", community.simple())) + .execute(pool) + .await + .expect("insert community"); + } + let cid = CommunityId::from_uuid(community); + let writer_only = "aa".repeat(32); + let replica_only = "bb".repeat(32); + relay_members::add_relay_member(&writer, cid, &writer_only, "member", None) + .await + .expect("seed writer member"); + relay_members::add_relay_member(&replica, cid, &replica_only, "member", None) + .await + .expect("seed replica member"); + + let mut db = Db::from_pools(writer.clone(), replica.clone()); + db.fence().force_open_for_tests(chrono::Utc::now()); + + // Budget unset ⇒ bounded arm disabled ⇒ writer. + assert!( + db.is_relay_member(cid, &writer_only) + .await + .expect("gate off"), + "budget unset must answer from the writer" + ); + assert!(!db.is_relay_member(cid, &replica_only).await.unwrap()); + + // Budget set + fresh entry ⇒ replica. + db.set_replica_read_max_age_for_tests(Some(std::time::Duration::from_secs(5))); + assert!( + db.is_relay_member(cid, &replica_only) + .await + .expect("gate on"), + "budget set must answer from the replica" + ); + assert!(!db.is_relay_member(cid, &writer_only).await.unwrap()); + + // Entry older than the budget ⇒ fail closed to the writer. Close + // first so no prior fresh entry can be the one proved (matches the + // count test; today `force_open_for_tests_at` also clears the ring). + db.fence().close(); + db.fence().force_open_for_tests_at( + chrono::Utc::now(), + std::time::Instant::now() - std::time::Duration::from_secs(10), + ); + assert!( + db.is_relay_member(cid, &writer_only) + .await + .expect("entry too old"), + "an over-budget entry must fail closed to the writer" + ); + + drop_scratch_db(&admin, replica, &rname).await; + drop_scratch_db(&admin, writer, &wname).await; + } + /// Community separation across every routed seam, verified on /// REPLICA-SERVED reads. /// diff --git a/crates/buzz-db/src/relay_members.rs b/crates/buzz-db/src/relay_members.rs index bfc56f82de..402229cdec 100644 --- a/crates/buzz-db/src/relay_members.rs +++ b/crates/buzz-db/src/relay_members.rs @@ -29,14 +29,41 @@ pub struct RelayMember { /// Returns `true` if `pubkey` (64-char hex) is a member of `community`. pub async fn is_relay_member(pool: &PgPool, community: CommunityId, pubkey: &str) -> Result { + let mut conn = pool.acquire().await?; + is_relay_member_on(&mut conn, community, pubkey).await +} + +/// [`is_relay_member`] on a specific session — the replica-routing path runs +/// the lookup on the exact reader connection whose heartbeat observation +/// proved fence coverage. +pub(crate) async fn is_relay_member_on( + conn: &mut sqlx::PgConnection, + community: CommunityId, + pubkey: &str, +) -> Result { let row = sqlx::query("SELECT 1 FROM relay_members WHERE community_id = $1 AND pubkey = $2") .bind(community.as_uuid()) .bind(pubkey) - .fetch_optional(pool) + .fetch_optional(conn) .await?; Ok(row.is_some()) } +/// Returns `true` if any member of `community` holds the `admin` or `owner` +/// role. Open relays don't *enforce* the roster, but startup +/// (`bootstrap_owner`) and operator provisioning still populate it — this is +/// how the workspace-profile gate detects whether a steward exists. +pub async fn has_admin_or_owner(pool: &PgPool, community: CommunityId) -> Result { + let row = sqlx::query( + "SELECT 1 FROM relay_members \ + WHERE community_id = $1 AND role IN ('admin', 'owner') LIMIT 1", + ) + .bind(community.as_uuid()) + .fetch_optional(pool) + .await?; + Ok(row.is_some()) +} + /// Returns the relay member record for `pubkey` in `community`, or `None`. pub async fn get_relay_member( pool: &PgPool, diff --git a/crates/buzz-relay/src/api/git/cas_publish.rs b/crates/buzz-relay/src/api/git/cas_publish.rs index c213e2913e..50bb36d818 100644 --- a/crates/buzz-relay/src/api/git/cas_publish.rs +++ b/crates/buzz-relay/src/api/git/cas_publish.rs @@ -1370,6 +1370,15 @@ mod tests { ); } + #[test] + fn published_head_moves_to_surviving_branch_after_current_branch_deletion() { + let refs = BTreeMap::from([("refs/heads/master".to_string(), "1".repeat(40))]); + assert_eq!( + resolve_published_head(&refs, "refs/heads/main".to_string(), "refs/heads/main"), + "refs/heads/master" + ); + } + #[test] fn digest_from_key_strips_prefix() { let k = format!("manifests/{}", "a".repeat(64)); diff --git a/crates/buzz-relay/src/api/git/transport.rs b/crates/buzz-relay/src/api/git/transport.rs index f8e0300277..d3118d8a76 100644 --- a/crates/buzz-relay/src/api/git/transport.rs +++ b/crates/buzz-relay/src/api/git/transport.rs @@ -1064,7 +1064,7 @@ pub async fn receive_pack( state.config.bind_addr.port() ); let hooks_dir = repo.path().join("hooks").display().to_string(); - let hook_env = vec![ + let mut hook_env = vec![ ("BUZZ_HOOK_URL", hook_url), ( "BUZZ_HOOK_SECRET", @@ -1077,13 +1077,8 @@ pub async fn receive_pack( auth.tenant.community().as_uuid().to_string(), ), ("BUZZ_PUSHER_PUBKEY", pusher_hex.clone()), - // Override any repo-local core.hooksPath setting; defense in - // depth even though the hydrated workspace has no inherited - // config. - ("GIT_CONFIG_COUNT", "1".to_string()), - ("GIT_CONFIG_KEY_0", "core.hooksPath".to_string()), - ("GIT_CONFIG_VALUE_0", hooks_dir), ]; + hook_env.extend(receive_pack_git_config(hooks_dir)); // Run receive-pack against the tempdir. Returns the *owned* subprocess // output (PackOutput) — crucially NOT a Response, so the post-push @@ -1111,6 +1106,23 @@ pub async fn receive_pack( Ok(finalize_push(&state, ctx).await) } +/// Per-process git configuration for the hydrated receive-pack workspace. +fn receive_pack_git_config(hooks_dir: String) -> Vec<(&'static str, String)> { + vec![ + // Override any repo-local core.hooksPath setting; defense in depth + // even though the hydrated workspace has no inherited config. + ("GIT_CONFIG_COUNT", "2".to_string()), + ("GIT_CONFIG_KEY_0", "core.hooksPath".to_string()), + ("GIT_CONFIG_VALUE_0", hooks_dir), + // A bare repository rejects deletion of its symbolic HEAD branch by + // default. Hydrated repositories are ephemeral, and cas_publish + // selects a surviving branch for the next manifest HEAD, so allow + // receive-pack to apply the deletion before that selection runs. + ("GIT_CONFIG_KEY_1", "receive.denyDeleteCurrent".to_string()), + ("GIT_CONFIG_VALUE_1", "ignore".to_string()), + ] +} + /// Buffered output of a `git --stateless-rpc` subprocess. /// /// The handler holds this as an owned value between subprocess completion @@ -1921,11 +1933,148 @@ mod track_c_tests { use buzz_core::CommunityId; use nostr::{EventBuilder, Keys, Kind, Tag}; use std::collections::BTreeMap; + use std::io::Write; + use std::process::Output; fn oid_sha1() -> String { "cb09a769da1c01f458fa6959d4e8eded38fac8d3".to_string() } + fn run_test_git(cwd: &Path, args: &[&str], extra_env: &[(&str, String)]) -> Output { + let mut cmd = std::process::Command::new("git"); + cmd.current_dir(cwd) + .args(args) + .env_clear() + .env("PATH", std::env::var("PATH").unwrap_or_default()) + .env("GIT_CONFIG_NOSYSTEM", "1") + .env("GIT_CONFIG_GLOBAL", "/dev/null") + .env("HOME", "/dev/null"); + for (key, value) in extra_env { + cmd.env(key, value); + } + cmd.output().expect("run git") + } + + fn run_test_receive_pack(repo: &Path, request: &[u8], extra_env: &[(&str, String)]) -> Output { + let mut cmd = std::process::Command::new("git"); + cmd.arg("receive-pack") + .arg("--stateless-rpc") + .arg(repo) + .stdin(std::process::Stdio::piped()) + .stdout(std::process::Stdio::piped()) + .stderr(std::process::Stdio::piped()) + .env_clear() + .env("PATH", std::env::var("PATH").unwrap_or_default()) + .env("GIT_CONFIG_NOSYSTEM", "1") + .env("GIT_CONFIG_GLOBAL", "/dev/null") + .env("HOME", "/dev/null"); + for (key, value) in extra_env { + cmd.env(key, value); + } + + let mut child = cmd.spawn().expect("spawn receive-pack"); + child + .stdin + .take() + .expect("receive-pack stdin") + .write_all(request) + .expect("write receive-pack request"); + child.wait_with_output().expect("wait for receive-pack") + } + + fn assert_git_success(output: Output, operation: &str) { + assert!( + output.status.success(), + "{operation} failed: {}", + String::from_utf8_lossy(&output.stderr) + ); + } + + #[test] + fn receive_pack_config_allows_deleting_current_branch() { + let root = tempfile::TempDir::new().expect("tempdir"); + let remote = root.path().join("remote.git"); + let source = root.path().join("source"); + let remote_arg = remote.to_str().expect("utf-8 remote path"); + let source_arg = source.to_str().expect("utf-8 source path"); + + assert_git_success( + run_test_git( + root.path(), + &["init", "--bare", "--initial-branch=main", remote_arg], + &[], + ), + "initialize bare remote", + ); + assert_git_success( + run_test_git( + root.path(), + &["init", "--initial-branch=main", source_arg], + &[], + ), + "initialize source repository", + ); + assert_git_success( + run_test_git(source.as_path(), &["config", "user.name", "Buzz Test"], &[]), + "configure user name", + ); + assert_git_success( + run_test_git( + source.as_path(), + &["config", "user.email", "buzz-test@example.com"], + &[], + ), + "configure user email", + ); + std::fs::write(source.join("README.md"), "test\n").expect("write fixture"); + assert_git_success( + run_test_git(source.as_path(), &["add", "README.md"], &[]), + "stage fixture", + ); + assert_git_success( + run_test_git(source.as_path(), &["commit", "-m", "fixture"], &[]), + "commit fixture", + ); + assert_git_success( + run_test_git( + source.as_path(), + &["push", remote_arg, "main:main", "main:master"], + &[], + ), + "seed main and master", + ); + + let oid_output = run_test_git(remote.as_path(), &["rev-parse", "refs/heads/main"], &[]); + assert!(oid_output.status.success()); + let old_oid = String::from_utf8(oid_output.stdout) + .expect("utf-8 oid") + .trim() + .to_string(); + let command = format!( + "{old_oid} {} refs/heads/main\0report-status\n", + "0".repeat(40) + ); + let mut request = format!("{:04x}", command.len() + 4).into_bytes(); + request.extend_from_slice(command.as_bytes()); + request.extend_from_slice(b"0000"); + + let git_config = receive_pack_git_config(remote.join("hooks").display().to_string()); + let output = run_test_receive_pack(remote.as_path(), &request, &git_config); + assert!( + output.status.success(), + "receive-pack failed: {}", + String::from_utf8_lossy(&output.stderr) + ); + assert!( + !receive_pack_report_rejected(&output.stdout), + "receive-pack rejected the deletion: {}", + String::from_utf8_lossy(&output.stdout) + ); + + assert!(!remote.join("refs/heads/main").exists()); + assert!(remote.join("refs/heads/master").exists()); + } + /// A gzip-encoded request body is transparently inflated before it /// reaches the git subprocess. Git's smart-HTTP client gzips the /// upload-pack/receive-pack request body past a size threshold (fires diff --git a/crates/buzz-relay/src/handlers/relay_admin.rs b/crates/buzz-relay/src/handlers/relay_admin.rs index 3f58a9c2aa..3782f2c516 100644 --- a/crates/buzz-relay/src/handlers/relay_admin.rs +++ b/crates/buzz-relay/src/handlers/relay_admin.rs @@ -10,7 +10,7 @@ //! | 9030 | Add member | admin or owner | //! | 9031 | Remove member | admin or owner | //! | 9032 | Change role | owner only | -//! | 9033 | Set workspace profile (icon) | admin or owner | +//! | 9033 | Set workspace profile (icon) | admin or owner; on an open relay whose community has no admin/owner row at all, any authenticated sender (see [`may_set_workspace_profile`]) | use std::sync::Arc; @@ -94,6 +94,35 @@ fn validate_workspace_icon(icon: &str) -> Result<(), String> { Ok(()) } +/// Whether `sender_role` may set the workspace profile (kind:9033). +/// +/// Closed relays (`membership_enforced == true`) require an `admin`/`owner` +/// row in `relay_members` — the enforced roster is the authority. Open relays +/// don't *enforce* the roster, but the data can still exist: startup +/// bootstraps `RELAY_OWNER_PUBKEY` as `owner` regardless of the flag +/// (`main.rs`), as does operator provisioning. So the rule is steward-wins: +/// +/// - a steward (any admin/owner row) exists → admin/owner only, exactly like +/// a closed relay. An open relay with a configured owner keeps its icon +/// owner-controlled instead of last-write-wins for every authenticated key. +/// - genuinely rosterless (e.g. a community created by +/// `ensure_configured_community`, which writes no owner row) → any +/// NIP-42-authenticated sender may set the icon, mirroring how open relays +/// gate every other write. Without this the icon is permanently unsettable: +/// the desktop deliberately shows the icon editor on open relays (see +/// `canEditIcon` in `EditCommunityDialog.tsx`, #2640) and defers to this +/// relay-side check, which used to always say no. +fn may_set_workspace_profile( + sender_role: &str, + membership_enforced: bool, + community_has_steward: bool, +) -> bool { + if !membership_enforced && !community_has_steward { + return true; + } + sender_role == "admin" || sender_role == "owner" +} + /// A relay-admin command failure, carrying the *category* of the failure so /// the ingest seam can map it to the right NIP-01 prefix and HTTP status. /// @@ -230,9 +259,33 @@ async fn execute_relay_admin_command( // kind:9033 — Set workspace profile (icon). Handled before p-tag // extraction: it targets the relay itself, not a member pubkey. if kind == RELAY_ADMIN_SET_WORKSPACE_PROFILE { - if sender_role != "admin" && sender_role != "owner" { + // Steward detection only matters on open relays (closed relays gate on + // the sender's own role either way), so skip the extra query there. + let community_has_steward = if state.config.require_relay_membership { + true + } else { + state + .db + .has_admin_or_owner(tenant.community()) + .await + .map_err(|e| format!("database error: {e}"))? + }; + if !may_set_workspace_profile( + sender_role, + state.config.require_relay_membership, + community_has_steward, + ) { return Err("actor not authorized: must be admin or owner".to_string()); } + if sender_role != "admin" && sender_role != "owner" { + // Rosterless-open-relay admit: 9033 writes no audit row and + // publishes no announcement event (unlike 9030/9031), so this warn + // is the only durable attribution of who changed the icon. + warn!( + sender = %sender_hex, + "workspace profile change admitted without a roster role (open relay, no steward)" + ); + } // Empty or missing icon tag clears the workspace icon. let icon = extract_tag_value(event, "icon").unwrap_or_default(); @@ -562,6 +615,46 @@ mod tests { assert!(validate_workspace_icon("").is_ok()); } + /// Closed relay (membership enforced): only an admin/owner row in + /// `relay_members` may set the workspace profile — a plain member, or a + /// pubkey with no row at all (empty role), must be refused. The steward + /// flag is irrelevant when membership is enforced (call sites pass `true`, + /// but the rule must not depend on it). + #[test] + fn closed_relay_requires_admin_or_owner_for_workspace_profile() { + for steward in [true, false] { + assert!(may_set_workspace_profile("owner", true, steward)); + assert!(may_set_workspace_profile("admin", true, steward)); + assert!(!may_set_workspace_profile("member", true, steward)); + assert!(!may_set_workspace_profile("", true, steward)); + } + } + + /// Open relay with a steward: startup bootstraps `RELAY_OWNER_PUBKEY` as + /// `owner` regardless of `require_relay_membership`, so an open relay's + /// community can hold admin/owner rows. When one exists, the icon stays + /// steward-only — the fix must not widen an owner-controlled icon to + /// every authenticated key. + #[test] + fn open_relay_with_steward_keeps_workspace_profile_steward_only() { + assert!(may_set_workspace_profile("owner", false, true)); + assert!(may_set_workspace_profile("admin", false, true)); + assert!(!may_set_workspace_profile("member", false, true)); + assert!(!may_set_workspace_profile("", false, true)); + } + + /// Open relay, genuinely rosterless (no admin/owner row anywhere): any + /// authenticated sender may set the icon — including the roleless (empty + /// role) case, which is *every* sender there. This is the bug being + /// fixed: the desktop shows the icon editor on open relays (#2640) but + /// the relay refused every 9033. + #[test] + fn rosterless_open_relay_admits_any_authenticated_sender_for_workspace_profile() { + assert!(may_set_workspace_profile("", false, false)); + assert!(may_set_workspace_profile("member", false, false)); + assert!(may_set_workspace_profile("owner", false, false)); + } + #[test] fn workspace_icon_https_ok() { assert!(validate_workspace_icon("https://example.com/icon.png").is_ok()); @@ -591,4 +684,216 @@ mod tests { let long_data = format!("data:image/png;base64,{}", "A".repeat(98_304)); assert!(validate_workspace_icon(&long_data).is_err()); } + + // ─── Call-site integration: the 9033 gate wired to real config + DB ──── + // + // The unit tests above pin `may_set_workspace_profile`'s truth table, but + // not its wiring: mutation-testing showed that inverting + // `state.config.require_relay_membership` at the call site — an exact + // inversion of the security contract — survives the default suite. These + // tests drive `handle_relay_admin_event` with a real `AppState` against + // Postgres, on both relay modes, so the wiring itself is pinned. Selected + // explicitly in CI's Backend Integration job; requires local Postgres + // (and hard-fails rather than skipping when it is unreachable). + + const TEST_DB_URL: &str = "postgres://buzz:buzz_dev@localhost:5432/buzz"; // sadscan:disable np.postgres.1 + + /// Build a real `AppState` + tenant for a fresh community on `host`, with + /// `require_relay_membership` set as given. Mirrors + /// `api::invites::tests::invite_test_state`. + async fn workspace_profile_test_state( + host: &str, + require_relay_membership: bool, + ) -> (Arc, TenantContext) { + let mut config = crate::config::Config::from_env().expect("config from env"); + let database_url = std::env::var("BUZZ_TEST_DATABASE_URL") + .or_else(|_| std::env::var("DATABASE_URL")) + .unwrap_or_else(|_| TEST_DB_URL.to_string()); + config.database_url = database_url.clone(); + config.redis_url = "redis://127.0.0.1:1".to_string(); + config.relay_url = format!("wss://{host}"); + config.require_relay_membership = require_relay_membership; + + let pool = sqlx::PgPool::connect(&database_url) + .await + .expect("requires reachable Postgres"); + let db = buzz_db::Db::from_pool(pool.clone()); + let record = db + .ensure_configured_community(host) + .await + .expect("ensure community"); + let tenant = TenantContext::resolved(record.id, host); + + let redis_pool = deadpool_redis::Config::from_url(&config.redis_url) + .create_pool(Some(deadpool_redis::Runtime::Tokio1)) + .expect("redis pool config"); + let pubsub = Arc::new( + buzz_pubsub::PubSubManager::new(&config.redis_url, redis_pool.clone()) + .await + .expect("pubsub manager"), + ); + let audit = buzz_audit::AuditService::new(pool.clone()); + let auth = buzz_auth::AuthService::new(config.auth.clone()); + let search = buzz_search::SearchService::new(pool.clone()); + let workflow_engine = Arc::new(buzz_workflow::WorkflowEngine::new( + db.clone(), + buzz_workflow::WorkflowConfig::default(), + )); + let media_storage = buzz_media::MediaStorage::new(&config.media).expect("media storage"); + let (state, _audit_shutdown) = AppState::new( + config, + db, + redis_pool, + audit, + pubsub, + auth, + search, + workflow_engine, + Keys::generate(), + media_storage, + ); + (Arc::new(state), tenant) + } + + /// Sign a fresh kind:9033 with `icon` and run it through the real + /// admission + command path. + async fn submit_9033( + state: &Arc, + tenant: &TenantContext, + keys: &Keys, + icon: &str, + ) -> Result<(), RelayAdminError> { + let event = EventBuilder::new(Kind::Custom(9033), "") + .tags(vec![Tag::parse(["icon", icon]).expect("icon tag")]) + .sign_with_keys(keys) + .expect("sign 9033"); + handle_relay_admin_event(tenant, state, &event).await + } + + async fn stored_icon(state: &Arc, tenant: &TenantContext) -> Option { + state + .db + .get_community_icon(tenant.community()) + .await + .expect("read icon") + } + + /// Open relay (`require_relay_membership = false`): a rosterless + /// community admits any authenticated sender, but the moment a steward + /// (admin/owner row) exists the gate reverts to steward-only. + /// + /// Discriminating: fails if the call site inverts or drops + /// `require_relay_membership`, or stops consulting `has_admin_or_owner`. + #[tokio::test] + #[ignore = "requires Postgres"] + async fn open_relay_9033_admits_roleless_only_until_a_steward_exists() { + let host = format!("icon-gate-open-{}.example", uuid::Uuid::new_v4().simple()); + let (state, tenant) = workspace_profile_test_state(&host, false).await; + let roleless = Keys::generate(); + let owner = Keys::generate(); + + // Rosterless: the roleless sender may set the icon. + submit_9033(&state, &tenant, &roleless, "https://example.com/open.png") + .await + .expect("rosterless open relay must admit an authenticated sender"); + assert_eq!( + stored_icon(&state, &tenant).await.as_deref(), + Some("https://example.com/open.png"), + "icon must actually be stored" + ); + + // Seed a steward — the same roleless sender must now be refused, and + // the previously stored icon must survive the refused attempt. + state + .db + .add_relay_member( + tenant.community(), + &owner.public_key().to_hex(), + "owner", + None, + ) + .await + .expect("seed owner"); + let refused = submit_9033(&state, &tenant, &roleless, "https://evil.example/pwn.png").await; + assert_eq!( + refused, + Err(RelayAdminError::Rejected( + "actor not authorized: must be admin or owner".to_string() + )), + "an open relay with a steward must refuse a roleless sender" + ); + assert_eq!( + stored_icon(&state, &tenant).await.as_deref(), + Some("https://example.com/open.png"), + "refused attempt must not mutate the icon" + ); + + // The steward still can. + submit_9033(&state, &tenant, &owner, "https://example.com/owner.png") + .await + .expect("the steward must retain icon control"); + assert_eq!( + stored_icon(&state, &tenant).await.as_deref(), + Some("https://example.com/owner.png") + ); + } + + /// Closed relay (`require_relay_membership = true`): admin/owner only — + /// a plain member and a roleless key are refused even though the + /// community also *looks* rosterless-then-stewarded to the open-relay + /// branch. Together with the open-relay test this kills the inverted-flag + /// mutant: no assignment of the flag satisfies both. + #[tokio::test] + #[ignore = "requires Postgres"] + async fn closed_relay_9033_still_requires_admin_or_owner() { + let host = format!("icon-gate-closed-{}.example", uuid::Uuid::new_v4().simple()); + let (state, tenant) = workspace_profile_test_state(&host, true).await; + let roleless = Keys::generate(); + let member = Keys::generate(); + let admin = Keys::generate(); + state + .db + .add_relay_member( + tenant.community(), + &member.public_key().to_hex(), + "member", + None, + ) + .await + .expect("seed member"); + state + .db + .add_relay_member( + tenant.community(), + &admin.public_key().to_hex(), + "admin", + None, + ) + .await + .expect("seed admin"); + + for (keys, label) in [(&roleless, "roleless"), (&member, "member")] { + let refused = submit_9033(&state, &tenant, keys, "https://evil.example/pwn.png").await; + assert_eq!( + refused, + Err(RelayAdminError::Rejected( + "actor not authorized: must be admin or owner".to_string() + )), + "closed relay must refuse a {label} sender" + ); + } + assert_eq!( + stored_icon(&state, &tenant).await, + None, + "refused attempts must not set an icon" + ); + + submit_9033(&state, &tenant, &admin, "https://example.com/closed.png") + .await + .expect("closed-relay admin must set the icon"); + assert_eq!( + stored_icon(&state, &tenant).await.as_deref(), + Some("https://example.com/closed.png") + ); + } } diff --git a/crates/buzz-sdk/src/builders.rs b/crates/buzz-sdk/src/builders.rs index 8cc9c8650a..9a139f0377 100644 --- a/crates/buzz-sdk/src/builders.rs +++ b/crates/buzz-sdk/src/builders.rs @@ -11,8 +11,8 @@ use buzz_core::{ KIND_GIT_STATUS_CLOSED, KIND_GIT_STATUS_DRAFT, KIND_GIT_STATUS_MERGED, KIND_GIT_STATUS_OPEN, KIND_IA_ARCHIVE_REQUEST, KIND_IA_UNARCHIVE_REQUEST, KIND_MODERATION_BAN, KIND_MODERATION_RESOLVE_REPORT, KIND_MODERATION_TIMEOUT, - KIND_MODERATION_UNBAN, KIND_MODERATION_UNTIMEOUT, KIND_PRESENCE_UPDATE, KIND_USER_STATUS, - KIND_WORKFLOW_DEF, KIND_WORKFLOW_TRIGGER, + KIND_MODERATION_UNBAN, KIND_MODERATION_UNTIMEOUT, KIND_PRESENCE_UPDATE, KIND_PROJECT, + KIND_USER_STATUS, KIND_WORKFLOW_DEF, KIND_WORKFLOW_TRIGGER, }, observer::{ content_looks_like_nip44, OBSERVER_AGENT_TAG, OBSERVER_FRAME_CONTROL, OBSERVER_FRAME_TAG, @@ -1499,12 +1499,7 @@ pub fn build_workflow_delete( author_pubkey: &str, workflow_id: Uuid, ) -> Result { - let pk = check_pubkey_hex(author_pubkey, "author_pubkey")?; - let tags = vec![tag(&[ - "a", - &format!("{}:{pk}:{workflow_id}", KIND_WORKFLOW_DEF), - ])?]; - Ok(EventBuilder::new(Kind::Custom(KIND_DELETION as u16), "").tags(tags)) + build_delete_addressable(KIND_WORKFLOW_DEF, author_pubkey, &workflow_id.to_string()) } /// Build a workflow trigger event (kind 46020). @@ -1838,6 +1833,364 @@ pub fn build_unarchive_identity_request( ) } +// ─── NIP-MP: Multi-repo projects (kind:30621) ──────────────────────────────── +// +// Public surface: +// • `validate_project_envelope` — Layer A protocol validator (8 ingest rules) +// • `build_project_with_tags` — Layer A raw builder (content + tags, no canonicalization) +// • `ProjectMemberCoord` — parsed member coordinate + optional relay hint +// • `build_project` — Layer B writer-policy builder +// • `build_delete_addressable` — generic NIP-09 kind:5 coordinate delete +// +// Byte-length bounds from NIP-MP §Relay Processing: +/// Maximum byte length of a project `d` tag value. +pub const PROJECT_D_MAX_LEN: usize = 1024; +/// Maximum byte length of a project `name` tag value. +pub const PROJECT_NAME_MAX: usize = 256; +/// Maximum byte length of a project `description` tag value. +pub const PROJECT_DESCRIPTION_MAX: usize = 2048; +/// Maximum byte length of a project `buzz-channel` tag value. +pub const PROJECT_CHANNEL_MAX: usize = 256; +/// Maximum byte length of a project `buzz-visibility` tag value. +pub const PROJECT_VISIBILITY_MAX: usize = 256; +/// Maximum number of `a` member tags per project event (checked before dedup). +pub const PROJECT_MEMBER_CAP: usize = 64; + +/// A validated NIP-MP member `a`-tag coordinate with an optional relay hint. +/// +/// Equality and `Hash` are by `coord` only (per spec: duplicate detection ignores hint). +#[derive(Clone, Debug)] +pub struct ProjectMemberCoord { + /// The full `30617::` coordinate string. + pub coord: String, + /// Optional opaque relay hint (third `a`-tag element, never validated by content). + pub hint: Option, +} + +impl PartialEq for ProjectMemberCoord { + fn eq(&self, other: &Self) -> bool { + self.coord == other.coord + } +} + +impl Eq for ProjectMemberCoord {} + +impl std::hash::Hash for ProjectMemberCoord { + fn hash(&self, state: &mut H) { + self.coord.hash(state); + } +} + +impl ProjectMemberCoord { + /// Parse a full `30617::` coordinate string. + /// + /// Accepts an optional relay hint as the third colon-separated element + /// after the split, but the split is always first-two-colons: kind, owner, + /// everything-else-as-repo-d. + /// + /// Rules enforced: + /// - Exactly three segments after splitting on the first two colons + /// - First segment must be the literal string `"30617"` + /// - Second segment must be exactly 64 lowercase hex characters + /// - Third segment (repo-d) must be non-empty + /// - Uppercase owners are rejected (never normalized) + pub fn parse_full(coord: &str) -> Result { + // Split on first two colons only: kind:owner:rest + let mut parts = coord.splitn(3, ':'); + let kind_part = parts.next().unwrap_or(""); + let owner_part = parts.next().unwrap_or(""); + let rest = parts.next().unwrap_or(""); + + if kind_part != "30617" { + return Err(SdkError::InvalidInput(format!( + "member coordinate must start with '30617:' (got kind {kind_part:?})" + ))); + } + if owner_part.len() != 64 || !owner_part.chars().all(|c| c.is_ascii_hexdigit()) { + return Err(SdkError::InvalidInput(format!( + "member owner must be a 64-character hex pubkey (got {owner_part:?})" + ))); + } + // Reject uppercase (spec: lowercase hex required) + if owner_part.chars().any(|c| c.is_ascii_uppercase()) { + return Err(SdkError::InvalidInput( + "member owner hex must be lowercase".into(), + )); + } + if rest.is_empty() { + return Err(SdkError::InvalidInput( + "member coordinate repo-d must not be empty".into(), + )); + } + Ok(ProjectMemberCoord { + coord: format!("30617:{owner_part}:{rest}"), + hint: None, + }) + } + + /// Returns the `a`-tag element slice: `[coord]` or `[coord, hint]`. + pub fn to_tag_parts(&self) -> Vec { + let mut parts = vec!["a".to_string(), self.coord.clone()]; + if let Some(h) = &self.hint { + parts.push(h.clone()); + } + parts + } +} + +/// **Layer A**: Validate a complete kind:30621 envelope against the 8 NIP-MP +/// ingest rules. This is the single source of protocol truth used by both +/// `build_project_with_tags` (raw path) and `build_project` (policy path). +/// +/// Rules enforced (matches relay `buzz-db` ingest logic): +/// 1. `d` cardinality: exactly one `d` tag. +/// 2. `d` value: non-empty, ≤1024 bytes. +/// 3. Member cap: raw count of every `a` tag ≤ 64 (checked **before** per-tag +/// parsing, matching relay rule order). +/// 4. Member tag arity: every `a` tag has 2 or 3 elements (no more, no fewer). +/// 5. Member coordinate grammar: first-two-colons split; kind literal `"30617"`; +/// owner lowercase 64-hex; repo-d non-empty verbatim. +/// 6. Member deduplication: coordinate equality only (hint ignored); any +/// coordinate that appears more than once is a duplicate. +/// 7. Singleton metadata: each of `name`, `description`, `buzz-channel`, +/// `buzz-visibility` appears at most once. +/// 8. Metadata byte lengths: `name` ≤256, `description` ≤2048, +/// `buzz-channel` ≤256, `buzz-visibility` ≤256. +pub fn validate_project_envelope(tags: &[Tag], _content: &str) -> Result<(), SdkError> { + // --- Rule 1 & 2: d tag --- + let d_tags: Vec<&Tag> = tags.iter().filter(|t| tag_name(t) == Some("d")).collect(); + match d_tags.len() { + 0 => { + return Err(SdkError::InvalidInput( + "project must have exactly one 'd' tag (rule: d-cardinality)".into(), + )) + } + 1 => {} + _ => { + return Err(SdkError::InvalidInput( + "project must have exactly one 'd' tag (rule: d-cardinality)".into(), + )) + } + } + let d_val = tag_value(d_tags[0]).unwrap_or(""); + if d_val.is_empty() { + return Err(SdkError::InvalidInput( + "project 'd' tag must not be empty (rule: d-empty)".into(), + )); + } + if d_val.len() > PROJECT_D_MAX_LEN { + return Err(SdkError::InvalidInput(format!( + "project 'd' tag exceeds {PROJECT_D_MAX_LEN} bytes (rule: d-empty)" + ))); + } + + let a_tags: Vec<&Tag> = tags.iter().filter(|t| tag_name(t) == Some("a")).collect(); + + // --- Rule 3: member cap (checked before per-tag parsing, matching relay rule order) --- + if a_tags.len() > PROJECT_MEMBER_CAP { + return Err(SdkError::InvalidInput(format!( + "project exceeds member cap of {PROJECT_MEMBER_CAP} (got {}) (rule: member-cap)", + a_tags.len() + ))); + } + + // --- Rule 4: member arity --- + for a in &a_tags { + let len = a.as_slice().len() - 1; // exclude the "a" name element + if !(1..=2).contains(&len) { + return Err(SdkError::InvalidInput(format!( + "member 'a' tag must have 1 or 2 value elements (got {len}) (rule: member-tag-arity)" + ))); + } + } + + // --- Rules 5 & 6: coordinate grammar + deduplication --- + let mut seen_coords: std::collections::HashSet = std::collections::HashSet::new(); + for a in &a_tags { + let coord_val = tag_value(a).unwrap_or(""); + ProjectMemberCoord::parse_full(coord_val).map_err(|e| { + SdkError::InvalidInput(format!("{e} (rule: member-coordinate-malformed)")) + })?; + if !seen_coords.insert(coord_val.to_string()) { + return Err(SdkError::InvalidInput(format!( + "duplicate member coordinate {coord_val:?} (rule: member-duplicate)" + ))); + } + } + + // --- Rules 7 & 8: singleton metadata + byte bounds --- + let singleton_fields = [ + ( + "name", + PROJECT_NAME_MAX, + "metadata-cardinality", + "metadata-length", + ), + ( + "description", + PROJECT_DESCRIPTION_MAX, + "metadata-cardinality", + "metadata-length", + ), + ( + "buzz-channel", + PROJECT_CHANNEL_MAX, + "metadata-cardinality", + "metadata-length", + ), + ( + "buzz-visibility", + PROJECT_VISIBILITY_MAX, + "metadata-cardinality", + "metadata-length", + ), + ]; + for (field, max_bytes, card_rule, len_rule) in singleton_fields { + let matches: Vec<&Tag> = tags.iter().filter(|t| tag_name(t) == Some(field)).collect(); + if matches.len() > 1 { + return Err(SdkError::InvalidInput(format!( + "project must have at most one '{field}' tag (rule: {card_rule})" + ))); + } + if let Some(t) = matches.first() { + let val = tag_value(t).unwrap_or(""); + if val.len() > max_bytes { + return Err(SdkError::InvalidInput(format!( + "'{field}' tag exceeds {max_bytes} bytes (rule: {len_rule})" + ))); + } + } + } + + Ok(()) +} + +/// Helper: tag name (first element). +fn tag_name(tag: &Tag) -> Option<&str> { + tag.as_slice().first().map(String::as_str) +} + +/// Helper: tag value (second element). +fn tag_value(tag: &Tag) -> Option<&str> { + tag.as_slice().get(1).map(String::as_str) +} + +/// **Layer A raw builder**: Build a kind:30621 project event from a raw +/// `content` string and a raw `tags` slice, without any canonicalization. +/// +/// Validates the entire envelope through `validate_project_envelope` before +/// accepting it. The caller is responsible for supplying the correct `d` tag. +/// This is the path exercised by fixture conformance tests and by read-modify- +/// write mutations in the CLI. +pub fn build_project_with_tags(content: &str, tags: Vec) -> Result { + validate_project_envelope(&tags, content)?; + Ok(EventBuilder::new(Kind::Custom(KIND_PROJECT as u16), content).tags(tags)) +} + +/// **Layer B writer-policy builder**: Build a kind:30621 project event with +/// enforced writer policy: +/// - The `d` tag is constructed from `slug`; `check_project_slug` rejects +/// an empty or over-length slug. +/// - `channel` must be a valid UUID string. +/// - `visibility` must be `"listed"` or `"unlisted"`. +/// - Content is always empty. +/// - Member coordinates are parsed through `ProjectMemberCoord::parse_full`. +/// +/// The resulting envelope is validated through Layer A before the builder is +/// returned. +pub fn build_project( + slug: &str, + name: Option<&str>, + description: Option<&str>, + members: &[ProjectMemberCoord], + channel: Option<&str>, + visibility: Option<&str>, +) -> Result { + // Slug validation + if slug.is_empty() { + return Err(SdkError::InvalidInput( + "project slug must not be empty".into(), + )); + } + if slug.len() > PROJECT_D_MAX_LEN { + return Err(SdkError::InvalidInput(format!( + "project slug must not exceed {PROJECT_D_MAX_LEN} bytes (got {})", + slug.len() + ))); + } + + // Channel UUID validation + if let Some(ch) = channel { + uuid::Uuid::parse_str(ch).map_err(|_| { + SdkError::InvalidInput(format!("buzz-channel must be a valid UUID (got {ch:?})")) + })?; + } + + // Visibility enum validation + if let Some(vis) = visibility { + if vis != "listed" && vis != "unlisted" { + return Err(SdkError::InvalidInput(format!( + "buzz-visibility must be 'listed' or 'unlisted' (got {vis:?})" + ))); + } + } + + let mut tags: Vec = Vec::new(); + tags.push(tag(&["d", slug])?); + + if let Some(n) = name { + tags.push(tag(&["name", n])?); + } + if let Some(d) = description { + tags.push(tag(&["description", d])?); + } + for m in members { + let tag_parts = m.to_tag_parts(); + let parts: Vec<&str> = tag_parts.iter().map(|s| s.as_str()).collect(); + // Safety: to_tag_parts always produces ["a", coord, ...hint] + tags.push( + Tag::parse(parts.iter().copied()).map_err(|e| SdkError::InvalidTag(e.to_string()))?, + ); + } + if let Some(ch) = channel { + tags.push(tag(&["buzz-channel", ch])?); + } + if let Some(vis) = visibility { + tags.push(tag(&["buzz-visibility", vis])?); + } + + build_project_with_tags("", tags) +} + +/// **Generic NIP-09 coordinate delete**: Build a kind:5 deletion event with +/// a single `a`-tag addressing `::`. +/// +/// Validates: +/// - `kind` is an addressable kind (10000–19999 or 30000–39999). +/// - `pubkey` is a 64-character lowercase hex string. +/// - `d` is non-empty. +/// +/// `build_workflow_delete` delegates to this function. +pub fn build_delete_addressable( + kind: u32, + pubkey: &str, + d: &str, +) -> Result { + let is_addressable = (10000..20000).contains(&kind) || (30000..40000).contains(&kind); + if !is_addressable { + return Err(SdkError::InvalidInput(format!( + "kind {kind} is not an addressable kind (must be 10000–19999 or 30000–39999)" + ))); + } + let pk = check_pubkey_hex(pubkey, "pubkey")?; + if d.is_empty() { + return Err(SdkError::InvalidInput("d must not be empty".into())); + } + let coord = format!("{kind}:{pk}:{d}"); + let tags = vec![tag(&["a", &coord])?]; + Ok(EventBuilder::new(Kind::Custom(KIND_DELETION as u16), "").tags(tags)) +} + #[cfg(test)] mod tests { use super::*; @@ -3884,4 +4237,280 @@ mod tests { .iter() .any(|t| t.as_slice().first().map(String::as_str) == Some("replaced-by"))); } + + // ── NIP-MP cap-before-arity ordering ───────────────────────────────────── + + /// When an envelope exceeds the member cap AND contains a malformed `a` tag, + /// the validator must fire `member-cap` (rule 3) — not `member-tag-arity` + /// (rule 4). This matches the relay's ingest ordering and means a client + /// sending an oversized list never receives a per-tag parse error. + #[test] + fn validate_project_envelope_cap_wins_over_arity_when_both_fail() { + let owner = "a".repeat(64); + // Build 65 well-formed `a` tags — enough to trigger the cap. + let mut tags = vec![Tag::parse(["d", "platform"]).unwrap()]; + for i in 0..65usize { + let coord = format!("30617:{owner}:repo-{i}"); + tags.push(Tag::parse(["a", &coord]).unwrap()); + } + // Also add one malformed tag (four elements) that would fire + // member-tag-arity if evaluated before the cap check. + let coord_extra = format!("30617:{owner}:repo-extra"); + tags.push( + Tag::parse([ + "a", + &coord_extra, + "wss://relay.example.com", + "extra-element", + ]) + .unwrap(), + ); + + let err = validate_project_envelope(&tags, "").unwrap_err(); + let msg = err.to_string(); + assert!( + msg.contains("member-cap"), + "expected member-cap to win, got: {msg}" + ); + assert!( + !msg.contains("member-tag-arity"), + "arity rule must not fire before cap rule, got: {msg}" + ); + } + + // ── Layer B writer-policy builder ─────────────────────────────────────── + + const OWNER64: &str = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"; + const VALID_UUID: &str = "3580ca9b-47b4-4af9-b22a-1068778f26c6"; + + fn member_coord(repo: &str) -> ProjectMemberCoord { + ProjectMemberCoord::parse_full(&format!("30617:{OWNER64}:{repo}")).unwrap() + } + + #[test] + fn build_project_emitted_envelope_has_correct_shape() { + // slug, name, description, channel, visibility, and one member. + let m = member_coord("buzz"); + let ev = sign( + build_project( + "my-proj", + Some("My Project"), + Some("A description"), + &[m], + Some(VALID_UUID), + Some("listed"), + ) + .expect("Layer B must accept valid inputs"), + ); + + // Kind must be 30621. + assert_eq!(ev.kind.as_u16(), KIND_PROJECT as u16); + // Content must be empty (Layer B policy). + assert!(ev.content.is_empty(), "content must be empty"); + + let all_tags: Vec> = ev.tags.iter().map(|t| t.as_slice().to_vec()).collect(); + + // d tag must be present exactly once. + let d_tags: Vec<_> = all_tags.iter().filter(|t| t[0] == "d").collect(); + assert_eq!(d_tags.len(), 1); + assert_eq!(d_tags[0][1], "my-proj"); + + // name, description, buzz-channel, buzz-visibility present. + let name_tags: Vec<_> = all_tags.iter().filter(|t| t[0] == "name").collect(); + assert_eq!(name_tags.len(), 1); + assert_eq!(name_tags[0][1], "My Project"); + + let desc_tags: Vec<_> = all_tags.iter().filter(|t| t[0] == "description").collect(); + assert_eq!(desc_tags.len(), 1); + assert_eq!(desc_tags[0][1], "A description"); + + let ch_tags: Vec<_> = all_tags.iter().filter(|t| t[0] == "buzz-channel").collect(); + assert_eq!(ch_tags.len(), 1); + assert_eq!(ch_tags[0][1], VALID_UUID); + + let vis_tags: Vec<_> = all_tags + .iter() + .filter(|t| t[0] == "buzz-visibility") + .collect(); + assert_eq!(vis_tags.len(), 1); + assert_eq!(vis_tags[0][1], "listed"); + + // member a tag. + let a_tags: Vec<_> = all_tags.iter().filter(|t| t[0] == "a").collect(); + assert_eq!(a_tags.len(), 1); + assert_eq!(a_tags[0][1], format!("30617:{OWNER64}:buzz")); + } + + #[test] + fn build_project_optional_fields_absent_when_not_supplied() { + let m = member_coord("core"); + let ev = sign( + build_project("my-proj", None, None, &[m], None, None) + .expect("minimal build must succeed"), + ); + let names: Vec<_> = ev + .tags + .iter() + .filter(|t| t.as_slice().first().map(|s| s.as_str()) == Some("name")) + .collect(); + assert!(names.is_empty(), "name tag must not be emitted when absent"); + } + + #[test] + fn build_project_rejects_empty_slug() { + let m = member_coord("r"); + let err = build_project("", None, None, &[m], None, None).unwrap_err(); + assert!( + matches!(err, SdkError::InvalidInput(_)), + "empty slug must be InvalidInput, got: {err:?}" + ); + assert!(err.to_string().contains("empty")); + } + + #[test] + fn build_project_rejects_overlong_slug() { + let long_slug = "a".repeat(PROJECT_D_MAX_LEN + 1); + let m = member_coord("r"); + let err = build_project(&long_slug, None, None, &[m], None, None).unwrap_err(); + assert!(matches!(err, SdkError::InvalidInput(_))); + } + + #[test] + fn build_project_rejects_invalid_channel_uuid() { + let m = member_coord("r"); + let err = build_project("slug", None, None, &[m], Some("not-a-uuid"), None).unwrap_err(); + assert!(matches!(err, SdkError::InvalidInput(_))); + assert!(err.to_string().contains("UUID") || err.to_string().contains("uuid")); + } + + #[test] + fn build_project_rejects_invalid_visibility_token() { + let m = member_coord("r"); + let err = build_project("slug", None, None, &[m], None, Some("chartreuse")).unwrap_err(); + assert!(matches!(err, SdkError::InvalidInput(_))); + assert!(err.to_string().contains("listed") || err.to_string().contains("unlisted")); + } + + #[test] + fn build_project_rejects_over_cap_members() { + let members: Vec<_> = (0..=PROJECT_MEMBER_CAP) + .map(|i| member_coord(&format!("repo-{i}"))) + .collect(); + let err = build_project("slug", None, None, &members, None, None).unwrap_err(); + assert!(matches!(err, SdkError::InvalidInput(_))); + assert!( + err.to_string().contains("member-cap"), + "over-cap must report member-cap, got: {err}" + ); + } + + #[test] + fn build_project_rejects_duplicate_members() { + let m = member_coord("same"); + let err = build_project("slug", None, None, &[m.clone(), m], None, None).unwrap_err(); + assert!(matches!(err, SdkError::InvalidInput(_))); + assert!( + err.to_string().contains("dedup") || err.to_string().contains("duplicate"), + "duplicate member must report dedup, got: {err}" + ); + } + + #[test] + fn build_project_content_is_always_empty() { + // build_project forces content="" regardless; Layer A also enforces + // that the envelope is valid. Any non-empty content would be dropped. + // This test pins the Layer B content-forced-empty policy. + let m = member_coord("r"); + let ev = sign(build_project("slug", None, None, &[m], None, None).unwrap()); + assert!( + ev.content.is_empty(), + "Layer B must always emit empty content" + ); + } + + // ── NIP-MP conformance fixtures ────────────────────────────────────────── + // `build_project_with_tags` directly. Accept cases must build; reject + // cases must fail with an error message containing the expected rule name. + // A count assertion guards against silent omissions. + // + // `include_str!` path is relative to this source file. + fn nip_mp_fixture_tags(json_tags: &serde_json::Value) -> Vec { + json_tags + .as_array() + .unwrap() + .iter() + .map(|t| { + let parts: Vec = t + .as_array() + .unwrap() + .iter() + .map(|v| v.as_str().unwrap().to_string()) + .collect(); + let parts_ref: Vec<&str> = parts.iter().map(String::as_str).collect(); + Tag::parse(parts_ref.iter().copied()) + .unwrap_or_else(|e| panic!("fixture tag parse error: {e}\n raw: {t}")) + }) + .collect() + } + + #[test] + fn nip_mp_fixtures_all_31_cases_exercised() { + const FIXTURE_JSON: &str = include_str!("../../../docs/nips/NIP-MP.fixtures.json"); + + let data: serde_json::Value = + serde_json::from_str(FIXTURE_JSON).expect("fixture JSON must parse"); + let cases = data["cases"].as_array().expect("cases must be array"); + + // Count gate: the spec says "required to test against this one file" + // with the exact count as-shipped. + assert_eq!( + cases.len(), + 31, + "expected 31 fixture cases, got {} — was NIP-MP.fixtures.json edited?", + cases.len() + ); + + let mut accept_count = 0usize; + let mut reject_count = 0usize; + + for case in cases { + let name = case["name"].as_str().unwrap(); + let expect = case["expect"].as_str().unwrap(); + let template = &case["template"]; + let content = template["content"].as_str().unwrap_or(""); + let tags = nip_mp_fixture_tags(&template["tags"]); + + match expect { + "accept" => { + build_project_with_tags(content, tags).unwrap_or_else(|e| { + panic!("fixture '{name}' (accept) must build successfully, got: {e}") + }); + accept_count += 1; + } + "reject" => { + let reject_rules = case["reject_rules"] + .as_array() + .expect("reject case must have reject_rules") + .iter() + .map(|r| r.as_str().unwrap().to_string()) + .collect::>(); + + let err = build_project_with_tags(content, tags).unwrap_err(); + let err_msg = err.to_string(); + + // The error must mention at least one of the expected rules. + let rule_matched = reject_rules.iter().any(|r| err_msg.contains(r.as_str())); + assert!( + rule_matched, + "fixture '{name}' rejected with wrong rule.\n expected one of: {reject_rules:?}\n got error: {err_msg}" + ); + reject_count += 1; + } + other => panic!("fixture '{name}' has unknown expect value: {other:?}"), + } + } + + assert_eq!(accept_count, 11, "expected 11 accept cases"); + assert_eq!(reject_count, 20, "expected 20 reject cases"); + } } diff --git a/deploy/charts/buzz/examples/argocd-app.yaml b/deploy/charts/buzz/examples/argocd-app.yaml index 8f6cb76228..a29a6919b7 100644 --- a/deploy/charts/buzz/examples/argocd-app.yaml +++ b/deploy/charts/buzz/examples/argocd-app.yaml @@ -16,9 +16,14 @@ metadata: spec: project: default source: - repoURL: oci://ghcr.io/block/buzz/charts - chart: buzz - targetRevision: 0.1.0 + # Argo CD >= 3.1 native OCI sources: repoURL must be the FULL chart + # artifact path — with the `repoURL: …/charts` + `chart: buzz` split + # form, the `chart` field is ignored for oci:// URLs and the fetch + # fails with a 403 (`repository:block/buzz/charts:pull` denied). The + # spec validator still requires `path`; use "." for OCI sources. + repoURL: oci://ghcr.io/block/buzz/charts/buzz + path: . + targetRevision: 0.1.7 helm: releaseName: buzz values: | diff --git a/desktop/package.json b/desktop/package.json index e8145f5468..d0b99c53f6 100644 --- a/desktop/package.json +++ b/desktop/package.json @@ -1,7 +1,7 @@ { "name": "buzz", "private": true, - "version": "0.5.3", + "version": "0.5.4", "type": "module", "scripts": { "dev": "vite", diff --git a/desktop/playwright.config.ts b/desktop/playwright.config.ts index 7ce7f48389..796fdede1e 100644 --- a/desktop/playwright.config.ts +++ b/desktop/playwright.config.ts @@ -41,6 +41,7 @@ export default defineConfig({ "**/channel-mute.spec.ts", "**/channel-star.spec.ts", "**/channel-controls.spec.ts", + "**/channel-activity-popover.spec.ts", "**/active-turn-resilience.spec.ts", "**/profile-active-turn.spec.ts", "**/config-bridge-screenshots.spec.ts", @@ -126,12 +127,14 @@ export default defineConfig({ "**/agent-provider-dropdowns.spec.ts", "**/agent-lifecycle-feedback.spec.ts", "**/agent-access-warning.spec.ts", + "**/edit-agent-run-on.spec.ts", "**/inbox-live-update.spec.ts", "**/mesh-compute.spec.ts", "**/observer-archive-policy.spec.ts", "**/harness-management.spec.ts", "**/harness-catalog-screenshots.spec.ts", "**/inline-custom-harness.spec.ts", + "**/where-to-run-config.spec.ts", "**/huddle-transcription.spec.ts", ], use: { diff --git a/desktop/scripts/build-release-config.mjs b/desktop/scripts/build-release-config.mjs index 389d18aec5..d1cd8181eb 100644 --- a/desktop/scripts/build-release-config.mjs +++ b/desktop/scripts/build-release-config.mjs @@ -52,6 +52,15 @@ const releaseConfig = { }, }; +// Tauri applies --config after platform-specific config using RFC 7396. +// Any externalBin value here would therefore replace the platform sidecar list, +// while null would silently delete it. This delta must never own that key. +if (Object.hasOwn(releaseConfig.bundle, "externalBin")) { + throw new Error( + "Release config must not define bundle.externalBin; sidecars are platform-specific", + ); +} + console.log(`Updater enabled -> ${updaterEndpoint}`); writeFileSync(outputConfigPath, `${JSON.stringify(releaseConfig, null, 2)}\n`); diff --git a/desktop/src-tauri/Cargo.lock b/desktop/src-tauri/Cargo.lock index 00d3fba3b5..365c8f712e 100644 --- a/desktop/src-tauri/Cargo.lock +++ b/desktop/src-tauri/Cargo.lock @@ -1036,7 +1036,7 @@ dependencies = [ [[package]] name = "buzz-desktop" -version = "0.5.3" +version = "0.5.4" dependencies = [ "anyhow", "arboard", @@ -1044,6 +1044,7 @@ dependencies = [ "audioadapter-buffers", "axum", "base64 0.22.1", + "block2", "buzz-agent", "buzz-core", "buzz-media", @@ -1106,6 +1107,7 @@ dependencies = [ "tauri-plugin-single-instance", "tauri-plugin-updater", "tauri-plugin-window-state", + "tauri-utils", "tempfile", "tokio", "tokio-tungstenite 0.29.0", @@ -6122,9 +6124,9 @@ dependencies = [ [[package]] name = "nostr" -version = "0.44.6" +version = "0.44.7" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "e826dd648489de2c5b293920e20b92932ef820302007c1987c758d4d06eeb2cf" +checksum = "c7d3d987ea7078dc36947cde532637c472a229426702e4331dd7667325378bd9" dependencies = [ "base64 0.22.1", "bech32", @@ -6166,9 +6168,9 @@ dependencies = [ [[package]] name = "nostr-relay-pool" -version = "0.44.1" +version = "0.44.3" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "91b2c039df4f96c4bf7dae52a74fd5516ad6dda83a11c0c69dea91b5255a4f37" +checksum = "c85c54d6ca9aae4ae2bf19a7663ba9db5f45f783f1d24aff55f006386b8b99a1" dependencies = [ "async-utility", "async-wsocket", diff --git a/desktop/src-tauri/Cargo.toml b/desktop/src-tauri/Cargo.toml index b80684f955..bf3a4ffe96 100644 --- a/desktop/src-tauri/Cargo.toml +++ b/desktop/src-tauri/Cargo.toml @@ -2,7 +2,7 @@ [package] name = "buzz-desktop" -version = "0.5.3" +version = "0.5.4" description = "Buzz desktop app" authors = ["you"] edition = "2021" @@ -46,8 +46,9 @@ notify-rust = "4" webkit2gtk = { version = "=2.0.2", features = ["v2_22"] } [target.'cfg(target_os = "macos")'.dependencies] +block2 = { version = "0.6", default-features = false, features = ["std"] } objc2 = { version = "0.6.4", default-features = false } -objc2-app-kit = { version = "0.3.2", default-features = false, features = ["NSHapticFeedback", "NSMenu", "NSMenuItem", "NSStatusItem"] } +objc2-app-kit = { version = "0.3.2", default-features = false, features = ["NSEvent", "NSHapticFeedback", "NSMenu", "NSMenuItem", "NSStatusItem", "block2"] } objc2-foundation = { version = "0.3.2", default-features = false, features = ["NSProcessInfo", "NSString"] } keyring = { version = "3.6.3", default-features = false, features = ["apple-native", "vendored"], optional = true } security-framework = { version = "3.7.0", features = ["OSX_10_15"] } @@ -138,6 +139,7 @@ strip-ansi-escapes = "0.2" tracing = "0.1" [dev-dependencies] +tauri-utils = "2" # `test-util` enables tokio's paused-clock (`start_paused`) so the relay # admission gate tests can assert exact wait durations without real sleeps. tokio = { version = "1", features = ["test-util"] } diff --git a/desktop/src-tauri/assets/card_template.png b/desktop/src-tauri/assets/card_template.png new file mode 100644 index 0000000000..2225d1d442 Binary files /dev/null and b/desktop/src-tauri/assets/card_template.png differ diff --git a/desktop/src-tauri/src/commands/agent_config.rs b/desktop/src-tauri/src/commands/agent_config.rs index 5a26f0f645..7aded79599 100644 --- a/desktop/src-tauri/src/commands/agent_config.rs +++ b/desktop/src-tauri/src/commands/agent_config.rs @@ -8,14 +8,14 @@ use crate::{ read_goose_file_config, reader::read_config_surface, types::{ - AcpConfigOptionEntry, AcpConfigOptionValue, AcpModelEntry, ConfigOrigin, - NormalizedField, RuntimeConfigSurface, SessionConfigCache, + AcpConfigOptionEntry, AcpConfigOptionValue, AcpModelEntry, InheritedConfigTiers, + RuntimeConfigSurface, SessionConfigCache, }, }, - current_instance_id, known_acp_runtime, load_managed_agents, load_personas, - resolve_effective_prompt_model_provider, save_managed_agents, sync_managed_agent_processes, + current_instance_id, is_reserved_env_key, is_well_formed_env_key, known_acp_runtime, + load_managed_agents, load_personas, save_managed_agents, sync_managed_agent_processes, AgentDefinition, GlobalAgentConfig, KnownAcpRuntime, ManagedAgentRecord, - ManagedAgentRuntimeKey, + ManagedAgentRuntimeKey, MAX_ENV_VALUE_BYTES, }, }; @@ -36,28 +36,86 @@ pub struct RuntimeFileConfigSubset { pub satisfied_env_keys: Vec, } -/// Resolve the config surface with persona and global default values applied. +/// Sanitize a raw env map from an inherited tier (persona or global) with the +/// same rules `merged_user_env` applies at spawn time: reserved keys, malformed +/// keys, NUL-byte values, and oversize values are stripped silently. +fn sanitize_inherited_env( + raw: &std::collections::BTreeMap, +) -> std::collections::BTreeMap { + raw.iter() + .filter(|(k, v)| { + !is_reserved_env_key(k) + && is_well_formed_env_key(k) + && !v.contains('\0') + && v.len() <= MAX_ENV_VALUE_BYTES + }) + .map(|(k, v)| (k.clone(), v.clone())) + .collect() +} + +/// Normalize a structured field value: blank/whitespace-only collapses to +/// `None`, matching `effective_config`'s `non_blank` helper. +fn non_blank(v: Option<&str>) -> Option { + v.filter(|s| !s.trim().is_empty()).map(str::to_owned) +} + +/// Build a sanitized `InheritedConfigTiers` snapshot at the command boundary. /// -/// Linked instances are definition-authoritative: the record's own -/// system_prompt/model/provider are cleared before applying, so a stale -/// materialized snapshot can never shadow the persona's current values or a -/// blank-definition fallthrough to global defaults (mirrors -/// `effective_config::resolve_linked`). Definition-less instances keep their -/// own explicit values. +/// Persona env, global env, and harness definition env are sanitized with +/// spawn-equivalent rules. Structured fields are normalized (blank → None). +/// A missing persona (orphaned link) yields empty persona tiers — the panel +/// still renders from record/global while spawn independently refuses. +fn build_inherited_tiers( + record_persona_id: Option<&str>, + record_runtime: Option<&str>, + personas: &[AgentDefinition], + global: &GlobalAgentConfig, +) -> InheritedConfigTiers { + let persona = record_persona_id.and_then(|pid| personas.iter().find(|p| p.id == pid)); + + let persona_env = persona + .map(|p| sanitize_inherited_env(&p.env_vars)) + .unwrap_or_default(); + let global_env = sanitize_inherited_env(&global.env_vars); + + // Definition env: same resolution as spawn (record.runtime → persona.runtime → ""). + // Reserved keys stripped; no malformed-key / NUL / oversize check needed because + // harness definitions are local admin-authored JSON, not user-provided data — but + // we apply `sanitize_inherited_env` for defense-in-depth (same rules as the other tiers). + let definition_env = { + let runtime_id = record_runtime + .or_else(|| persona.and_then(|p| p.runtime.as_deref())) + .unwrap_or(""); + crate::managed_agents::custom_harnesses::lookup_loaded_harness_by_id(runtime_id) + .map(|def| sanitize_inherited_env(&def.env)) + .unwrap_or_default() + }; + + let persona_model = persona.and_then(|p| non_blank(p.model.as_deref())); + let persona_provider = persona.and_then(|p| non_blank(p.provider.as_deref())); + let persona_prompt = persona.and_then(|p| non_blank(Some(&p.system_prompt))); + let global_model = non_blank(global.model.as_deref()); + let global_provider = non_blank(global.provider.as_deref()); + + InheritedConfigTiers { + persona_env, + global_env, + definition_env, + persona_model, + persona_provider, + persona_prompt, + global_model, + global_provider, + } +} + +/// Resolve the config surface with inherited persona and global tiers applied. /// -/// The pipeline: resolve the linked persona's prompt/model/provider, inject -/// each into the record only where the record lacks its own value, let -/// `read_config_surface` tag those injected fields `BuzzExplicit`, then re-tag -/// exactly the injected fields to `PersonaDefault`. -/// -/// Global defaults fill in when neither the record nor the linked persona -/// provides a value. They are re-tagged to `GlobalDefault` so the UI can -/// display "inherited from global defaults". -/// -/// The re-tag is triple-gated — a field is re-tagged only when (a) the record -/// did not already have it (`!had_*`), (b) the surface produced the field, and -/// (c) the reader tagged it `BuzzExplicit`. A value the user set explicitly in -/// Buzz keeps `had_* == true` and is never re-tagged. +/// Persona-linked instances have their system_prompt/model/provider cleared +/// first (definition-authoritative): stale materialized snapshots can never +/// shadow live persona values. The reader then resolves each field through its +/// full candidate list (record env > ACP > persona env > global env > structured +/// persona/global > config file) via `resolve_with_override`. fn resolve_config_surface( mut record: ManagedAgentRecord, personas: &[AgentDefinition], @@ -65,146 +123,23 @@ fn resolve_config_surface( session_cache: Option<&SessionConfigCache>, global: &GlobalAgentConfig, ) -> RuntimeConfigSurface { - // Linked instances are definition-authoritative (mirrors - // `effective_config::resolve_linked`): the record's own - // system_prompt/model/provider fields are, at best, a stale materialized - // snapshot from the last `apply_persona_snapshot` — never a legitimate - // live override, since `update_managed_agent` blocks writing these three - // fields for linked instances. Clear them before computing `had_*` below - // so a stale byte can never masquerade as BuzzExplicit and suppress - // definition/global injection. Env var overrides (set via the advanced - // env-vars editor) are untouched — those remain a legitimate - // per-instance override regardless of link status. + // Linked instances are definition-authoritative: clear stale materialized + // model/provider/prompt so they can never masquerade as BuzzExplicit and + // shadow definition values. Env var overrides are untouched. if record.persona_id.is_some() { record.system_prompt = None; record.model = None; record.provider = None; } - let had_prompt = - record.system_prompt.is_some() || record.env_vars.contains_key("BUZZ_ACP_SYSTEM_PROMPT"); - let had_model = record.model.is_some(); - - let provider_env_key = runtime_meta.and_then(|m| m.provider_env_var).unwrap_or(""); - let had_provider = record.env_vars.contains_key(provider_env_key); - - let (persona_prompt, persona_model, persona_provider) = resolve_effective_prompt_model_provider( + let tiers = build_inherited_tiers( record.persona_id.as_deref(), + record.runtime.as_deref(), personas, - record.system_prompt.clone(), - record.model.clone(), - record.provider.clone(), + global, ); - // Build the baseline the reader overrides a live model against, paired with - // its true origin so the secondary is tagged correctly. Two sources: - // - persona-linked, no explicit record model: the persona model is the - // baseline (PersonaDefault). - // - genuine-explicit (record had its own model) that live-switched: the - // record's own model is the baseline (BuzzExplicit). Gated behind - // `model_overridden` so a persona edited mid-life (override flag false) - // never synthesizes a baseline and false-positives an override. - // An explicit pick with no live switch has no baseline to override. - let model_overridden = session_cache.is_some_and(|c| c.model_overridden); - let baseline = if had_model { - if model_overridden { - record - .model - .clone() - .map(|m| (m, ConfigOrigin::BuzzExplicit)) - } else { - None - } - } else { - // Prefer persona as baseline, fall back to global when persona has none - // and the model was overridden mid-session (global-default agent). - persona_model - .clone() - .map(|m| (m, ConfigOrigin::PersonaDefault)) - .or_else(|| { - if model_overridden { - global - .model - .clone() - .map(|m| (m, ConfigOrigin::GlobalDefault)) - } else { - None - } - }) - }; - - // Inject resolved persona values into the record where absent. - if !had_prompt { - if let Some(p) = persona_prompt { - record - .env_vars - .insert("BUZZ_ACP_SYSTEM_PROMPT".to_string(), p); - } - } - if !had_model { - record.model = persona_model.clone(); - } - if !had_provider && !provider_env_key.is_empty() { - if let Some(prov) = persona_provider { - record.env_vars.insert(provider_env_key.to_string(), prov); - } - } - - // Inject global defaults where neither the record nor the persona had a value. - // Track injection so we can re-tag to GlobalDefault after the reader. - let inject_global_model = !had_model && record.model.is_none(); - let inject_global_provider = !had_provider - && !provider_env_key.is_empty() - && !record.env_vars.contains_key(provider_env_key); - - if inject_global_model { - record.model = global.model.clone(); - } - if inject_global_provider { - if let Some(ref gprov) = global.provider { - record - .env_vars - .insert(provider_env_key.to_string(), gprov.clone()); - } - } - - let mut surface = read_config_surface( - &record, - runtime_meta, - session_cache, - baseline.as_ref().map(|(m, o)| (m.as_str(), o.clone())), - ); - - // Re-tag persona-sourced fields from BuzzExplicit to PersonaDefault. - if !had_prompt { - retag_persona_default(&mut surface.normalized.system_prompt); - } - if !had_model && !inject_global_model { - retag_persona_default(&mut surface.normalized.model); - } - if !had_provider && !provider_env_key.is_empty() && !inject_global_provider { - retag_persona_default(&mut surface.normalized.provider); - } - - // Re-tag global-sourced fields from BuzzExplicit to GlobalDefault. - if inject_global_model { - retag_global_default(&mut surface.normalized.model); - } - if inject_global_provider { - retag_global_default(&mut surface.normalized.provider); - } - - surface -} - -/// Re-tag a field's origin from `BuzzExplicit` to `PersonaDefault`, leaving any -/// other origin untouched. No-op when the field is absent. -fn retag_persona_default(field: &mut Option) { - if let Some(field) = field { - if field.origin == ConfigOrigin::BuzzExplicit { - field.origin = ConfigOrigin::PersonaDefault; - } - } + read_config_surface(&record, runtime_meta, session_cache, &tiers) } /// Get the file-layer config for a runtime — used by the Create/Edit/Persona @@ -327,16 +262,6 @@ pub fn get_baked_build_env() -> Vec { .collect() } -/// Re-tag a field's origin from `BuzzExplicit` to `GlobalDefault`, leaving any -/// other origin untouched. No-op when the field is absent. -fn retag_global_default(field: &mut Option) { - if let Some(field) = field { - if field.origin == ConfigOrigin::BuzzExplicit { - field.origin = ConfigOrigin::GlobalDefault; - } - } -} - /// Get the full config surface for a managed agent. /// /// Returns normalized + advanced config from all available tiers. @@ -601,511 +526,5 @@ fn parse_models(raw: Option<&serde_json::Value>) -> (Vec, Option< } #[cfg(test)] -mod tests { - use super::*; - use crate::managed_agents::{BackendKind, RespondTo}; - - fn goose_runtime() -> &'static KnownAcpRuntime { - &KnownAcpRuntime { - id: "goose", - label: "Goose", - commands: &["goose"], - aliases: &[], - avatar_url: "", - mcp_command: None, - mcp_hooks: false, - underlying_cli: None, - cli_install_commands: &[], - cli_install_commands_windows: &[], - adapter_install_commands: &[], - cli_install_instructions_url: "", - adapter_install_instructions_url: "", - cli_install_hint: "", - adapter_install_hint: "", - skill_dir: None, - supports_acp_model_switching: false, - model_env_var: Some("GOOSE_MODEL"), - provider_env_var: Some("GOOSE_PROVIDER"), - provider_locked: false, - default_env: &[], - config_file_path: Some("~/.config/goose/config.yaml"), - config_file_format: Some("yaml"), - supports_acp_native_config: true, - thinking_env_var: Some("GOOSE_THINKING_EFFORT"), - max_tokens_env_var: Some("GOOSE_MAX_TOKENS"), - context_limit_env_var: Some("GOOSE_CONTEXT_LIMIT"), - required_normalized_fields: &["model", "provider"], - login_hint: None, - auth_probe_args: None, - } - } - - fn agent_record() -> ManagedAgentRecord { - ManagedAgentRecord { - pubkey: "agent".to_string(), - name: "Agent".to_string(), - persona_id: Some("persona-1".to_string()), - private_key_nsec: "".to_string(), - auth_tag: None, - relay_url: "ws://localhost:3000".to_string(), - avatar_url: None, - acp_command: "buzz-acp".to_string(), - agent_command: "goose".to_string(), - agent_args: vec![], - mcp_command: "".to_string(), - turn_timeout_seconds: 300, - idle_timeout_seconds: None, - max_turn_duration_seconds: None, - parallelism: 1, - system_prompt: None, - model: None, - env_vars: Default::default(), - start_on_app_launch: false, - auto_restart_on_config_change: true, - runtime_pid: None, - backend: BackendKind::Local, - backend_agent_id: None, - provider_binary_path: None, - team_id: None, - persona_team_dir: None, - persona_name_in_team: None, - created_at: "".to_string(), - updated_at: "".to_string(), - last_started_at: None, - last_stopped_at: None, - last_exit_code: None, - last_error: None, - last_error_code: None, - respond_to: RespondTo::OwnerOnly, - respond_to_allowlist: vec![], - display_name: None, - slug: None, - runtime: None, - name_pool: Vec::new(), - is_builtin: false, - is_active: true, - shared: false, - source_team: None, - source_team_persona_slug: None, - catalog_source: None, - definition_respond_to: None, - definition_respond_to_allowlist: Vec::new(), - definition_parallelism: None, - relay_mesh: None, - agent_command_override: None, - persona_source_version: None, - provider: None, - } - } - - fn persona_with_model(model: &str) -> AgentDefinition { - AgentDefinition { - id: "persona-1".to_string(), - display_name: "Persona".to_string(), - avatar_url: None, - system_prompt: "You are a persona.".to_string(), - runtime: None, - model: Some(model.to_string()), - provider: None, - name_pool: Vec::new(), - is_builtin: false, - is_active: true, - shared: false, - source_team: None, - source_team_persona_slug: None, - catalog_source: None, - env_vars: Default::default(), - respond_to: None, - respond_to_allowlist: Vec::new(), - parallelism: None, - created_at: "".to_string(), - updated_at: "".to_string(), - } - } - - /// A post-spawn session cache whose live model is `current_model` and whose - /// `model_overridden` flag records whether a `SwitchModel` control signal set - /// it (the live-switch signal). - fn session_cache(current_model: &str, model_overridden: bool) -> SessionConfigCache { - SessionConfigCache { - config_options: vec![], - available_modes: vec![], - available_models: vec![], - current_model: Some(current_model.to_string()), - model_overridden, - goose_native_config: None, - captured_at: "".to_string(), - } - } - - /// Definition-authoritative: a stale materialized `record.model` on a - /// linked instance must never outrank (or even be consulted against) the - /// linked persona's model. `update_managed_agent` already blocks writing - /// model/provider/prompt for linked instances, so a non-`None` value here - /// can only be leftover snapshot bytes from before a persona edit — the - /// panel must report the persona's current model, tagged `PersonaDefault`, - /// not the stale byte as `BuzzExplicit`. - #[test] - fn linked_stale_record_model_never_outranks_persona_model() { - let mut record = agent_record(); - record.model = Some("stale-explicit-model".to_string()); - let personas = vec![persona_with_model("persona-model")]; - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - None, - &Default::default(), - ); - - let model = surface.normalized.model.as_ref().expect("model resolved"); - assert_eq!(model.value.as_deref(), Some("persona-model")); - assert_eq!(model.origin, ConfigOrigin::PersonaDefault); - } - - /// Definition-authoritative, blank-definition case: a linked instance - /// whose persona has no model of its own must fall through to the global - /// default, tagged `GlobalDefault` — mirroring - /// `effective_config::resolve_linked`'s `None => global` arm. A stale - /// materialized record model must not shadow this fallthrough either. - #[test] - fn linked_blank_definition_model_falls_through_to_global_default() { - let mut record = agent_record(); - record.model = Some("stale-explicit-model".to_string()); - let mut persona = persona_with_model("unused"); - persona.model = None; - let personas = vec![persona]; - let global = crate::managed_agents::GlobalAgentConfig { - model: Some("global-model".to_string()), - ..Default::default() - }; - - let surface = - resolve_config_surface(record, &personas, Some(goose_runtime()), None, &global); - - let model = surface.normalized.model.as_ref().expect("model resolved"); - assert_eq!(model.value.as_deref(), Some("global-model")); - assert_eq!(model.origin, ConfigOrigin::GlobalDefault); - } - - /// A definition-less (no `persona_id`) instance's own explicit model IS - /// authoritative — the stale-record clearing above is scoped to linked - /// instances only. - #[test] - fn definition_less_explicit_record_model_keeps_buzz_explicit_origin() { - let mut record = agent_record(); - record.persona_id = None; - record.model = Some("explicit-model".to_string()); - let personas = vec![persona_with_model("persona-model")]; - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - None, - &Default::default(), - ); - - let model = surface.normalized.model.as_ref().expect("model resolved"); - assert_eq!(model.value.as_deref(), Some("explicit-model")); - assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); - } - - /// Part A — pending-pick: a genuine-explicit pick X with a divergent live - /// model Y but `model_overridden == false` (the live switch is not yet - /// applied — a restart is pending) must keep X as the primary and must NOT - /// surface Y as an override row. The live `acp_model` does not win. This - /// FAILS against a let-live-acp-win variant (one that dropped the - /// `model_overridden` gate), so it is not vacuous. - #[test] - fn pending_pick_keeps_explicit_x_and_does_not_surface_live_y() { - let mut record = agent_record(); - record.persona_id = None; - record.model = Some("model-x".to_string()); - let personas: Vec = vec![]; - let cache = session_cache("model-y", false); - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - Some(&cache), - &Default::default(), - ); - let model = surface.normalized.model.expect("model resolved"); - - assert_eq!(model.value.as_deref(), Some("model-x")); - assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); - assert_ne!(model.origin, ConfigOrigin::RuntimeOverride); - assert_ne!(model.overridden_value.as_deref(), Some("model-y")); - } - - /// W2 — genuine-explicit live switch: record.model = X, no persona, - /// `model_overridden == true`, live model = Y. The live Y must render as the - /// primary with a `RuntimeOverride` origin and X as the secondary tagged - /// `BuzzExplicit` (its true source — NOT `PersonaDefault`). FAILS against the - /// shipped no-persona early-return, which left X as primary and Y struck. - #[test] - fn genuine_explicit_live_switch_renders_y_over_x_buzz_explicit_secondary() { - let mut record = agent_record(); - record.persona_id = None; - record.model = Some("model-x".to_string()); - let personas: Vec = vec![]; - let cache = session_cache("model-y", true); - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - Some(&cache), - &Default::default(), - ); - let model = surface.normalized.model.expect("model resolved"); - - assert_eq!(model.value.as_deref(), Some("model-y")); - assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); - assert_eq!(model.overridden_value.as_deref(), Some("model-x")); - assert_eq!(model.overridden_origin, Some(ConfigOrigin::BuzzExplicit)); - } - - /// Y==X collision: a genuine-explicit agent live-switches to the SAME value - /// it already had. There is no real divergence, so the field must be a clean - /// single value with NO secondary row. FAILS against a naive `return base` - /// that would leak the `AcpConfigOption` row `build_model_field` populates. - #[test] - fn genuine_explicit_live_switch_to_same_model_yields_clean_field() { - let mut record = agent_record(); - record.persona_id = None; - record.model = Some("model-x".to_string()); - let personas: Vec = vec![]; - let cache = session_cache("model-x", true); - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - Some(&cache), - &Default::default(), - ); - let model = surface.normalized.model.expect("model resolved"); - - assert_eq!(model.value.as_deref(), Some("model-x")); - assert_eq!(model.overridden_value, None); - assert_eq!(model.overridden_origin, None); - } - - /// Persona parity (regression): a persona-linked agent with no explicit - /// record model that live-switches still renders the persona model as the - /// secondary tagged `PersonaDefault` — the typed-baseline change must NOT - /// regress the persona arm to a different origin. - #[test] - fn persona_linked_live_switch_keeps_persona_default_secondary() { - let record = agent_record(); - let personas = vec![persona_with_model("persona-model")]; - let cache = session_cache("model-y", true); - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - Some(&cache), - &Default::default(), - ); - let model = surface.normalized.model.expect("model resolved"); - - assert_eq!(model.value.as_deref(), Some("model-y")); - assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); - assert_eq!(model.overridden_value.as_deref(), Some("persona-model")); - assert_eq!(model.overridden_origin, Some(ConfigOrigin::PersonaDefault)); - } - - /// Fix 2 regression: a global-default-only agent (no record model, no - /// persona model, but global has a model) that live-switches mid-session - /// must render the global model as the secondary tagged `GlobalDefault`. - /// Before the fix, `baseline` was `None` in the `!had_model` arm when - /// persona has no model, so `read_config_surface` had no secondary to - /// surface. Fails against pre-fix code where the baseline arm returned - /// `None` when `!had_model && persona_model.is_none() && model_overridden`. - #[test] - fn global_default_live_switch_renders_global_model_as_secondary_global_default() { - // Record has no model, no persona, global provides the model. - let mut record = agent_record(); - record.persona_id = None; - // record.model = None (set by agent_record()) - let personas: Vec = vec![]; - let cache = session_cache("model-y", true); - let global = crate::managed_agents::GlobalAgentConfig { - model: Some("global-model".to_string()), - ..Default::default() - }; - - let surface = resolve_config_surface( - record, - &personas, - Some(goose_runtime()), - Some(&cache), - &global, - ); - let model = surface.normalized.model.expect("model resolved"); - - // Live model wins as primary. - assert_eq!(model.value.as_deref(), Some("model-y")); - assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); - // Global model surfaces as secondary, tagged GlobalDefault. - assert_eq!( - model.overridden_value.as_deref(), - Some("global-model"), - "global model must be the override baseline secondary" - ); - assert_eq!( - model.overridden_origin, - Some(ConfigOrigin::GlobalDefault), - "override baseline origin must be GlobalDefault, not PersonaDefault or BuzzExplicit" - ); - } - - // ── get_baked_build_env / is_secret_key tests ────────────────────────── - - /// Build a `BakedEnvEntry` vec from a synthetic map, mirroring what - /// `get_baked_build_env()` does. Used to test masking without relying on - /// compile-time `option_env!` vars (OSS builds have empty `baked_build_env`). - fn baked_env_from_map(map: &[(&str, &str)]) -> Vec { - map.iter() - .filter(|(_, v)| !v.is_empty()) - .map(|(k, v)| { - let masked = !super::is_safe_to_reveal(k); - BakedEnvEntry { - key: k.to_string(), - value: if masked { - "\u{2022}\u{2022}\u{2022}\u{2022}\u{2022}\u{2022}".to_string() - } else { - v.to_string() - }, - masked, - } - }) - .collect() - } - - #[test] - fn baked_env_non_secret_key_shows_real_value() { - let entries = baked_env_from_map(&[("BUZZ_AGENT_PROVIDER", "databricks_v2")]); - assert_eq!(entries.len(), 1); - assert_eq!(entries[0].key, "BUZZ_AGENT_PROVIDER"); - assert_eq!(entries[0].value, "databricks_v2"); - assert!(!entries[0].masked); - } - - #[test] - fn baked_env_api_key_is_masked() { - let entries = baked_env_from_map(&[("ANTHROPIC_API_KEY", "sk-secret")]); - assert_eq!(entries.len(), 1); - assert_eq!(entries[0].value, "••••••"); - assert!(entries[0].masked); - } - - #[test] - fn baked_env_token_key_is_masked() { - let entries = baked_env_from_map(&[("GITHUB_TOKEN", "ghp_secret")]); - assert_eq!(entries.len(), 1); - assert!(entries[0].masked); - } - - #[test] - fn baked_env_secret_key_is_masked() { - let entries = baked_env_from_map(&[("MY_DB_SECRET", "s3cr3t")]); - assert_eq!(entries.len(), 1); - assert!(entries[0].masked); - } - - #[test] - fn baked_env_password_key_is_masked() { - let entries = baked_env_from_map(&[("DB_PASSWORD", "hunter2")]); - assert_eq!(entries.len(), 1); - assert!(entries[0].masked); - } - - #[test] - fn baked_env_empty_value_filtered_out() { - let entries = baked_env_from_map(&[("BUZZ_AGENT_PROVIDER", "")]); - assert!(entries.is_empty()); - } - - #[test] - fn baked_env_mixed_keys_correct_masking() { - let entries = baked_env_from_map(&[ - ("BUZZ_AGENT_PROVIDER", "databricks_v2"), - ("BUZZ_AGENT_MODEL", "goose-claude-opus-4-8"), - ("DATABRICKS_HOST", "https://example.com"), - ("DATABRICKS_TOKEN", "dapi-secret"), - ]); - assert_eq!(entries.len(), 4); - - let provider = entries - .iter() - .find(|e| e.key == "BUZZ_AGENT_PROVIDER") - .unwrap(); - assert_eq!(provider.value, "databricks_v2"); - assert!(!provider.masked); - - let model = entries - .iter() - .find(|e| e.key == "BUZZ_AGENT_MODEL") - .unwrap(); - assert_eq!(model.value, "goose-claude-opus-4-8"); - assert!(!model.masked); - - let host = entries.iter().find(|e| e.key == "DATABRICKS_HOST").unwrap(); - assert_eq!(host.value, "https://example.com"); - assert!(!host.masked); - - let token = entries - .iter() - .find(|e| e.key == "DATABRICKS_TOKEN") - .unwrap(); - assert_eq!(token.value, "••••••"); - assert!(token.masked); - } - - #[test] - fn baked_env_thinking_effort_is_unmasked() { - // BUZZ_AGENT_THINKING_EFFORT is a non-secret enum — must not be masked. - let entries = baked_env_from_map(&[("BUZZ_AGENT_THINKING_EFFORT", "medium")]); - assert_eq!(entries.len(), 1); - let effort = entries - .iter() - .find(|e| e.key == "BUZZ_AGENT_THINKING_EFFORT") - .unwrap(); - assert_eq!(effort.value, "medium"); - assert!(!effort.masked); - } - - #[test] - fn baked_env_allowlist_is_case_insensitive() { - // Known-safe keys — case-insensitive match must allow them. - assert!(super::is_safe_to_reveal("buzz_agent_provider")); - assert!(super::is_safe_to_reveal("BUZZ_AGENT_PROVIDER")); - assert!(super::is_safe_to_reveal("buzz_agent_model")); - assert!(super::is_safe_to_reveal("BUZZ_AGENT_MODEL")); - assert!(super::is_safe_to_reveal("buzz_agent_thinking_effort")); - assert!(super::is_safe_to_reveal("BUZZ_AGENT_THINKING_EFFORT")); - assert!(super::is_safe_to_reveal("databricks_host")); - assert!(super::is_safe_to_reveal("DATABRICKS_HOST")); - assert!(super::is_safe_to_reveal("databricks_model")); - assert!(super::is_safe_to_reveal("DATABRICKS_MODEL")); - // Keys NOT in the allowlist — masked regardless of naming pattern. - assert!(!super::is_safe_to_reveal("my_api_key")); - assert!(!super::is_safe_to_reveal("GITHUB_TOKEN")); - assert!(!super::is_safe_to_reveal("DB_SECRET")); - assert!(!super::is_safe_to_reveal("DB_PASSWORD")); - // Bare names that old heuristic (contains("_TOKEN") etc.) would have missed. - assert!(!super::is_safe_to_reveal("APIKEY")); - assert!(!super::is_safe_to_reveal("TOKEN")); - assert!(!super::is_safe_to_reveal("SECRET")); - assert!(!super::is_safe_to_reveal("PASSWORD")); - assert!(!super::is_safe_to_reveal("PRIVATE_KEY")); - // Unknown key → masked by default. - assert!(!super::is_safe_to_reveal("SOME_UNKNOWN_KEY")); - } -} +#[path = "agent_config_tests.rs"] +mod tests; diff --git a/desktop/src-tauri/src/commands/agent_config_tests.rs b/desktop/src-tauri/src/commands/agent_config_tests.rs new file mode 100644 index 0000000000..f3667cff45 --- /dev/null +++ b/desktop/src-tauri/src/commands/agent_config_tests.rs @@ -0,0 +1,651 @@ +//! Unit tests for `commands/agent_config.rs` (split to keep `agent_config.rs` +//! under the 1000-line file-size ratchet). +//! +//! Included via `#[path = "agent_config_tests.rs"] mod tests;` at the bottom of +//! `agent_config.rs`, so `use super::*` gives access to all items in that module. + +use super::*; +use crate::managed_agents::config_bridge::types::ConfigOrigin; +use crate::managed_agents::{BackendKind, RespondTo}; + +use std::sync::Mutex; + +static GOOSE_PATH_ROOT_LOCK: Mutex<()> = Mutex::new(()); + +/// Run a test body with GOOSE_PATH_ROOT set to a non-existent path so that the +/// goose config file read returns `None`. Restores the prior value on exit. +fn with_no_goose_config(body: impl FnOnce() -> T) -> T { + let _guard = GOOSE_PATH_ROOT_LOCK + .lock() + .unwrap_or_else(|err| err.into_inner()); + let prior = std::env::var_os("GOOSE_PATH_ROOT"); + std::env::set_var("GOOSE_PATH_ROOT", "/nonexistent-buzz-test-path"); + let output = body(); + match prior { + Some(value) => std::env::set_var("GOOSE_PATH_ROOT", value), + None => std::env::remove_var("GOOSE_PATH_ROOT"), + } + output +} + +fn goose_runtime() -> &'static KnownAcpRuntime { + &KnownAcpRuntime { + id: "goose", + label: "Goose", + commands: &["goose"], + aliases: &[], + avatar_url: "", + mcp_command: None, + mcp_hooks: false, + underlying_cli: None, + cli_install_commands: &[], + cli_install_commands_windows: &[], + adapter_install_commands: &[], + cli_install_instructions_url: "", + adapter_install_instructions_url: "", + cli_install_hint: "", + adapter_install_hint: "", + skill_dir: None, + supports_acp_model_switching: false, + model_env_var: Some("GOOSE_MODEL"), + provider_env_var: Some("GOOSE_PROVIDER"), + provider_locked: false, + default_env: &[], + config_file_path: Some("~/.config/goose/config.yaml"), + config_file_format: Some("yaml"), + supports_acp_native_config: true, + thinking_env_var: Some("GOOSE_THINKING_EFFORT"), + max_tokens_env_var: Some("GOOSE_MAX_TOKENS"), + context_limit_env_var: Some("GOOSE_CONTEXT_LIMIT"), + required_normalized_fields: &["model", "provider"], + login_hint: None, + auth_probe_args: None, + } +} + +fn agent_record() -> ManagedAgentRecord { + ManagedAgentRecord { + pubkey: "agent".to_string(), + name: "Agent".to_string(), + persona_id: Some("persona-1".to_string()), + private_key_nsec: "".to_string(), + auth_tag: None, + relay_url: "ws://localhost:3000".to_string(), + avatar_url: None, + acp_command: "buzz-acp".to_string(), + agent_command: "goose".to_string(), + agent_args: vec![], + mcp_command: "".to_string(), + turn_timeout_seconds: 300, + idle_timeout_seconds: None, + max_turn_duration_seconds: None, + parallelism: 1, + system_prompt: None, + model: None, + env_vars: Default::default(), + start_on_app_launch: false, + auto_restart_on_config_change: true, + runtime_pid: None, + backend: BackendKind::Local, + backend_agent_id: None, + provider_binary_path: None, + team_id: None, + persona_team_dir: None, + persona_name_in_team: None, + created_at: "".to_string(), + updated_at: "".to_string(), + last_started_at: None, + last_stopped_at: None, + last_exit_code: None, + last_error: None, + last_error_code: None, + respond_to: RespondTo::OwnerOnly, + respond_to_allowlist: vec![], + display_name: None, + slug: None, + runtime: None, + name_pool: Vec::new(), + is_builtin: false, + is_active: true, + shared: false, + source_team: None, + source_team_persona_slug: None, + catalog_source: None, + definition_respond_to: None, + definition_respond_to_allowlist: Vec::new(), + definition_parallelism: None, + relay_mesh: None, + agent_command_override: None, + persona_source_version: None, + provider: None, + } +} + +fn persona_with_model(model: &str) -> AgentDefinition { + AgentDefinition { + id: "persona-1".to_string(), + display_name: "Persona".to_string(), + avatar_url: None, + system_prompt: "You are a persona.".to_string(), + runtime: None, + model: Some(model.to_string()), + provider: None, + name_pool: Vec::new(), + is_builtin: false, + is_active: true, + shared: false, + source_team: None, + source_team_persona_slug: None, + catalog_source: None, + env_vars: Default::default(), + respond_to: None, + respond_to_allowlist: Vec::new(), + parallelism: None, + created_at: "".to_string(), + updated_at: "".to_string(), + } +} + +/// A post-spawn session cache whose live model is `current_model` and whose +/// `model_overridden` flag records whether a `SwitchModel` control signal set +/// it (the live-switch signal). +fn session_cache(current_model: &str, model_overridden: bool) -> SessionConfigCache { + SessionConfigCache { + config_options: vec![], + available_modes: vec![], + available_models: vec![], + current_model: Some(current_model.to_string()), + model_overridden, + goose_native_config: None, + captured_at: "".to_string(), + } +} + +/// Definition-authoritative: a stale materialized `record.model` on a +/// linked instance must never outrank (or even be consulted against) the +/// linked persona's model. `update_managed_agent` already blocks writing +/// model/provider/prompt for linked instances, so a non-`None` value here +/// can only be leftover snapshot bytes from before a persona edit — the +/// panel must report the persona's current model, tagged `PersonaDefault`, +/// not the stale byte as `BuzzExplicit`. +#[test] +fn linked_stale_record_model_never_outranks_persona_model() { + let mut record = agent_record(); + record.model = Some("stale-explicit-model".to_string()); + let personas = vec![persona_with_model("persona-model")]; + + let surface = resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + None, + &Default::default(), + ); + + let model = surface.normalized.model.as_ref().expect("model resolved"); + assert_eq!(model.value.as_deref(), Some("persona-model")); + assert_eq!(model.origin, ConfigOrigin::PersonaDefault); +} + +/// Definition-authoritative, blank-definition case: a linked instance +/// whose persona has no model of its own must fall through to the global +/// default, tagged `GlobalDefault` — mirroring +/// `effective_config::resolve_linked`'s `None => global` arm. A stale +/// materialized record model must not shadow this fallthrough either. +#[test] +fn linked_blank_definition_model_falls_through_to_global_default() { + let mut record = agent_record(); + record.model = Some("stale-explicit-model".to_string()); + let mut persona = persona_with_model("unused"); + persona.model = None; + let personas = vec![persona]; + let global = crate::managed_agents::GlobalAgentConfig { + model: Some("global-model".to_string()), + ..Default::default() + }; + + let surface = resolve_config_surface(record, &personas, Some(goose_runtime()), None, &global); + + let model = surface.normalized.model.as_ref().expect("model resolved"); + assert_eq!(model.value.as_deref(), Some("global-model")); + assert_eq!(model.origin, ConfigOrigin::GlobalDefault); +} + +/// A definition-less (no `persona_id`) instance's own explicit model IS +/// authoritative — the stale-record clearing above is scoped to linked +/// instances only. +#[test] +fn definition_less_explicit_record_model_keeps_buzz_explicit_origin() { + let mut record = agent_record(); + record.persona_id = None; + record.model = Some("explicit-model".to_string()); + let personas = vec![persona_with_model("persona-model")]; + + let surface = resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + None, + &Default::default(), + ); + + let model = surface.normalized.model.as_ref().expect("model resolved"); + assert_eq!(model.value.as_deref(), Some("explicit-model")); + assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); +} + +/// Part A — pending-pick: a genuine-explicit pick X with a divergent live +/// model Y but `model_overridden == false` (the live switch is not yet +/// applied — a restart is pending) must keep X as the primary and must NOT +/// surface Y as an override row. The live `acp_model` does not win. This +/// FAILS against a let-live-acp-win variant (one that dropped the +/// `model_overridden` gate), so it is not vacuous. +#[test] +fn pending_pick_keeps_explicit_x_and_does_not_surface_live_y() { + let mut record = agent_record(); + record.persona_id = None; + record.model = Some("model-x".to_string()); + let personas: Vec = vec![]; + let cache = session_cache("model-y", false); + + let surface = resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + Some(&cache), + &Default::default(), + ); + let model = surface.normalized.model.expect("model resolved"); + + assert_eq!(model.value.as_deref(), Some("model-x")); + assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); + assert_ne!(model.origin, ConfigOrigin::RuntimeOverride); + assert_ne!(model.overridden_value.as_deref(), Some("model-y")); +} + +/// W2 — genuine-explicit live switch: record.model = X, no persona, +/// `model_overridden == true`, live model = Y. The live Y must render as the +/// primary with a `RuntimeOverride` origin and X as the secondary tagged +/// `BuzzExplicit` (its true source — NOT `PersonaDefault`). FAILS against the +/// shipped no-persona early-return, which left X as primary and Y struck. +#[test] +fn genuine_explicit_live_switch_renders_y_over_x_buzz_explicit_secondary() { + let mut record = agent_record(); + record.persona_id = None; + record.model = Some("model-x".to_string()); + let personas: Vec = vec![]; + let cache = session_cache("model-y", true); + + let surface = resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + Some(&cache), + &Default::default(), + ); + let model = surface.normalized.model.expect("model resolved"); + + assert_eq!(model.value.as_deref(), Some("model-y")); + assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); + assert_eq!(model.overridden_value.as_deref(), Some("model-x")); + assert_eq!(model.overridden_origin, Some(ConfigOrigin::BuzzExplicit)); +} + +/// Y==X collision: a genuine-explicit agent live-switches to the SAME value +/// it already had. There is no real divergence, so the field must be a clean +/// single value with NO secondary row and origin matching the baseline (not +/// RuntimeOverride). FAILS against a naive `return base` that would leak the +/// `AcpConfigOption` row `build_model_field` populates, and against the +/// prior implementation that stamped `RuntimeOverride` on the equal-value arm. +/// +/// `with_no_goose_config` suppresses the goose config file read so that the +/// fall-through to normal resolution cannot pick up a local `~/.config/goose/config.yaml` +/// model as a spurious secondary — the test is about tier precedence, not the +/// local developer's goose install. +#[test] +fn genuine_explicit_live_switch_to_same_model_yields_clean_field() { + let mut record = agent_record(); + record.persona_id = None; + record.model = Some("model-x".to_string()); + let personas: Vec = vec![]; + let cache = session_cache("model-x", true); + + let surface = with_no_goose_config(|| { + resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + Some(&cache), + &Default::default(), + ) + }); + let model = surface.normalized.model.expect("model resolved"); + + assert_eq!(model.value.as_deref(), Some("model-x")); + // Equal-value switch must NOT stamp RuntimeOverride — baseline origin wins. + assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); + assert_ne!(model.origin, ConfigOrigin::RuntimeOverride); + assert_eq!(model.overridden_value, None); + assert_eq!(model.overridden_origin, None); +} + +/// Persona parity (regression): a persona-linked agent with no explicit +/// record model that live-switches still renders the persona model as the +/// secondary tagged `PersonaDefault` — the typed-baseline change must NOT +/// regress the persona arm to a different origin. +#[test] +fn persona_linked_live_switch_keeps_persona_default_secondary() { + let record = agent_record(); + let personas = vec![persona_with_model("persona-model")]; + let cache = session_cache("model-y", true); + + let surface = resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + Some(&cache), + &Default::default(), + ); + let model = surface.normalized.model.expect("model resolved"); + + assert_eq!(model.value.as_deref(), Some("model-y")); + assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); + assert_eq!(model.overridden_value.as_deref(), Some("persona-model")); + assert_eq!(model.overridden_origin, Some(ConfigOrigin::PersonaDefault)); +} + +/// Fix 2 regression: a global-default-only agent (no record model, no +/// persona model, but global has a model) that live-switches mid-session +/// must render the global model as the secondary tagged `GlobalDefault`. +/// Before the fix, `baseline` was `None` in the `!had_model` arm when +/// persona has no model, so `read_config_surface` had no secondary to +/// surface. Fails against pre-fix code where the baseline arm returned +/// `None` when `!had_model && persona_model.is_none() && model_overridden`. +#[test] +fn global_default_live_switch_renders_global_model_as_secondary_global_default() { + // Record has no model, no persona, global provides the model. + let mut record = agent_record(); + record.persona_id = None; + // record.model = None (set by agent_record()) + let personas: Vec = vec![]; + let cache = session_cache("model-y", true); + let global = crate::managed_agents::GlobalAgentConfig { + model: Some("global-model".to_string()), + ..Default::default() + }; + + let surface = resolve_config_surface( + record, + &personas, + Some(goose_runtime()), + Some(&cache), + &global, + ); + let model = surface.normalized.model.expect("model resolved"); + + // Live model wins as primary. + assert_eq!(model.value.as_deref(), Some("model-y")); + assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); + // Global model surfaces as secondary, tagged GlobalDefault. + assert_eq!( + model.overridden_value.as_deref(), + Some("global-model"), + "global model must be the override baseline secondary" + ); + assert_eq!( + model.overridden_origin, + Some(ConfigOrigin::GlobalDefault), + "override baseline origin must be GlobalDefault, not PersonaDefault or BuzzExplicit" + ); +} + +// ── Snapshot constructor tests (build_inherited_tiers) ────────────────────── +// +// These test the sanitized snapshot constructor — the command-boundary +// function that builds InheritedConfigTiers from raw persona/global data. + +/// Orphaned persona link: a record whose persona_id references a non-existent +/// persona should produce empty persona tiers (not a panic), and the panel +/// still renders from the record and global tiers. +#[test] +fn orphaned_persona_link_yields_empty_persona_tiers() { + let mut record = agent_record(); + record.persona_id = Some("missing-persona".to_string()); + // No personas in the list — dangling link. + let personas: Vec = vec![]; + let global = crate::managed_agents::GlobalAgentConfig { + model: Some("global-model".to_string()), + ..Default::default() + }; + + let tiers = build_inherited_tiers(record.persona_id.as_deref(), None, &personas, &global); + + // Persona tier is empty — the orphan yields no persona inheritance. + assert!(tiers.persona_env.is_empty()); + assert!(tiers.persona_model.is_none()); + assert!(tiers.persona_provider.is_none()); + assert!(tiers.persona_prompt.is_none()); + // Global tiers are unaffected. + assert_eq!(tiers.global_model.as_deref(), Some("global-model")); +} + +/// Reserved key in persona env is stripped by sanitization — it must never +/// reach the reader or the display surface. +#[test] +fn reserved_key_in_inherited_persona_env_is_stripped() { + let mut persona = persona_with_model("model"); + // BUZZ_PRIVATE_KEY is a reserved key — must be stripped. + persona + .env_vars + .insert("BUZZ_PRIVATE_KEY".to_string(), "nsec-secret".to_string()); + // A safe key — must survive. + persona + .env_vars + .insert("GOOSE_MODEL".to_string(), "persona-model".to_string()); + let personas = vec![persona]; + let global = crate::managed_agents::GlobalAgentConfig::default(); + + let tiers = build_inherited_tiers(Some("persona-1"), None, &personas, &global); + + assert!( + !tiers.persona_env.contains_key("BUZZ_PRIVATE_KEY"), + "reserved key must be stripped from persona env tier" + ); + assert!( + tiers.persona_env.contains_key("GOOSE_MODEL"), + "safe key must survive sanitization" + ); +} + +/// `sanitize_inherited_env` strips reserved keys from a definition-env-shaped +/// map. This pins the shared sanitization contract for definition_env — the +/// same function is applied to all three env tiers (persona, global, definition) +/// at the command boundary. +#[test] +fn reserved_key_in_definition_env_shaped_map_is_stripped_by_sanitize() { + // Exercise sanitize_inherited_env directly with a definition-env-shaped map. + let mut raw = std::collections::BTreeMap::new(); + raw.insert("BUZZ_PRIVATE_KEY".to_string(), "nsec-secret".to_string()); + raw.insert("GOOSE_MODEL".to_string(), "harness-model".to_string()); + + let sanitized = sanitize_inherited_env(&raw); + + assert!( + !sanitized.contains_key("BUZZ_PRIVATE_KEY"), + "reserved key must be stripped by sanitize_inherited_env" + ); + assert!( + sanitized.contains_key("GOOSE_MODEL"), + "safe key must survive sanitize_inherited_env" + ); +} + +/// Malformed key in global env is stripped by sanitization — keys must be +/// POSIX-shaped (`[A-Za-z_][A-Za-z0-9_]*`). +#[test] +fn malformed_key_in_inherited_global_env_is_stripped() { + let mut global = crate::managed_agents::GlobalAgentConfig::default(); + // Key with an `=` — would bypass env-var security if passed to spawn. + global + .env_vars + .insert("BAD=KEY".to_string(), "value".to_string()); + // A valid key — must survive. + global + .env_vars + .insert("GOOSE_PROVIDER".to_string(), "anthropic".to_string()); + + let tiers = build_inherited_tiers(None, None, &[], &global); + + assert!( + !tiers.global_env.contains_key("BAD=KEY"), + "malformed key must be stripped from global env tier" + ); + assert!( + tiers.global_env.contains_key("GOOSE_PROVIDER"), + "valid key must survive sanitization" + ); +} + +// ── get_baked_build_env / is_secret_key tests ────────────────────────── + +/// Build a `BakedEnvEntry` vec from a synthetic map, mirroring what +/// `get_baked_build_env()` does. Used to test masking without relying on +/// compile-time `option_env!` vars (OSS builds have empty `baked_build_env`). +fn baked_env_from_map(map: &[(&str, &str)]) -> Vec { + map.iter() + .filter(|(_, v)| !v.is_empty()) + .map(|(k, v)| { + let masked = !super::is_safe_to_reveal(k); + BakedEnvEntry { + key: k.to_string(), + value: if masked { + "\u{2022}\u{2022}\u{2022}\u{2022}\u{2022}\u{2022}".to_string() + } else { + v.to_string() + }, + masked, + } + }) + .collect() +} + +#[test] +fn baked_env_non_secret_key_shows_real_value() { + let entries = baked_env_from_map(&[("BUZZ_AGENT_PROVIDER", "databricks_v2")]); + assert_eq!(entries.len(), 1); + assert_eq!(entries[0].key, "BUZZ_AGENT_PROVIDER"); + assert_eq!(entries[0].value, "databricks_v2"); + assert!(!entries[0].masked); +} + +#[test] +fn baked_env_api_key_is_masked() { + let entries = baked_env_from_map(&[("ANTHROPIC_API_KEY", "sk-secret")]); + assert_eq!(entries.len(), 1); + assert_eq!(entries[0].value, "••••••"); + assert!(entries[0].masked); +} + +#[test] +fn baked_env_token_key_is_masked() { + let entries = baked_env_from_map(&[("GITHUB_TOKEN", "ghp_secret")]); + assert_eq!(entries.len(), 1); + assert!(entries[0].masked); +} + +#[test] +fn baked_env_secret_key_is_masked() { + let entries = baked_env_from_map(&[("MY_DB_SECRET", "s3cr3t")]); + assert_eq!(entries.len(), 1); + assert!(entries[0].masked); +} + +#[test] +fn baked_env_password_key_is_masked() { + let entries = baked_env_from_map(&[("DB_PASSWORD", "hunter2")]); + assert_eq!(entries.len(), 1); + assert!(entries[0].masked); +} + +#[test] +fn baked_env_empty_value_filtered_out() { + let entries = baked_env_from_map(&[("BUZZ_AGENT_PROVIDER", "")]); + assert!(entries.is_empty()); +} + +#[test] +fn baked_env_mixed_keys_correct_masking() { + let entries = baked_env_from_map(&[ + ("BUZZ_AGENT_PROVIDER", "databricks_v2"), + ("BUZZ_AGENT_MODEL", "goose-claude-opus-4-8"), + ("DATABRICKS_HOST", "https://example.com"), + ("DATABRICKS_TOKEN", "dapi-secret"), + ]); + assert_eq!(entries.len(), 4); + + let provider = entries + .iter() + .find(|e| e.key == "BUZZ_AGENT_PROVIDER") + .unwrap(); + assert_eq!(provider.value, "databricks_v2"); + assert!(!provider.masked); + + let model = entries + .iter() + .find(|e| e.key == "BUZZ_AGENT_MODEL") + .unwrap(); + assert_eq!(model.value, "goose-claude-opus-4-8"); + assert!(!model.masked); + + let host = entries.iter().find(|e| e.key == "DATABRICKS_HOST").unwrap(); + assert_eq!(host.value, "https://example.com"); + assert!(!host.masked); + + let token = entries + .iter() + .find(|e| e.key == "DATABRICKS_TOKEN") + .unwrap(); + assert_eq!(token.value, "••••••"); + assert!(token.masked); +} + +#[test] +fn baked_env_thinking_effort_is_unmasked() { + // BUZZ_AGENT_THINKING_EFFORT is a non-secret enum — must not be masked. + let entries = baked_env_from_map(&[("BUZZ_AGENT_THINKING_EFFORT", "medium")]); + assert_eq!(entries.len(), 1); + let effort = entries + .iter() + .find(|e| e.key == "BUZZ_AGENT_THINKING_EFFORT") + .unwrap(); + assert_eq!(effort.value, "medium"); + assert!(!effort.masked); +} + +#[test] +fn baked_env_allowlist_is_case_insensitive() { + // Known-safe keys — case-insensitive match must allow them. + assert!(super::is_safe_to_reveal("buzz_agent_provider")); + assert!(super::is_safe_to_reveal("BUZZ_AGENT_PROVIDER")); + assert!(super::is_safe_to_reveal("buzz_agent_model")); + assert!(super::is_safe_to_reveal("BUZZ_AGENT_MODEL")); + assert!(super::is_safe_to_reveal("buzz_agent_thinking_effort")); + assert!(super::is_safe_to_reveal("BUZZ_AGENT_THINKING_EFFORT")); + assert!(super::is_safe_to_reveal("databricks_host")); + assert!(super::is_safe_to_reveal("DATABRICKS_HOST")); + assert!(super::is_safe_to_reveal("databricks_model")); + assert!(super::is_safe_to_reveal("DATABRICKS_MODEL")); + // Keys NOT in the allowlist — masked regardless of naming pattern. + assert!(!super::is_safe_to_reveal("my_api_key")); + assert!(!super::is_safe_to_reveal("GITHUB_TOKEN")); + assert!(!super::is_safe_to_reveal("DB_SECRET")); + assert!(!super::is_safe_to_reveal("DB_PASSWORD")); + // Bare names that old heuristic (contains("_TOKEN") etc.) would have missed. + assert!(!super::is_safe_to_reveal("APIKEY")); + assert!(!super::is_safe_to_reveal("TOKEN")); + assert!(!super::is_safe_to_reveal("SECRET")); + assert!(!super::is_safe_to_reveal("PASSWORD")); + assert!(!super::is_safe_to_reveal("PRIVATE_KEY")); + // Unknown key → masked by default. + assert!(!super::is_safe_to_reveal("SOME_UNKNOWN_KEY")); +} diff --git a/desktop/src-tauri/src/commands/agent_discovery.rs b/desktop/src-tauri/src/commands/agent_discovery.rs index cbbf4ce351..4abf53ee91 100644 --- a/desktop/src-tauri/src/commands/agent_discovery.rs +++ b/desktop/src-tauri/src/commands/agent_discovery.rs @@ -333,10 +333,7 @@ fn install_acp_runtime_blocking( // For the codex runtime, "found" is not enough — the resolved binary must also // pass the 1.x version gate. An outdated 0.16.x adapter must be overwritten by // the new npm install so the CODEX_CONFIG spawn contract works correctly. - let adapter_path = runtime - .commands - .iter() - .find_map(|cmd| crate::managed_agents::resolve_command(cmd)); + let adapter_path = resolve_adapter_path(runtime.commands, runtime.adapter_install_commands); let adapter_probe_path = crate::managed_agents::readiness::cli_probe::augmented_path(); if let Some(cmds) = plan_adapter_install( runtime_id, @@ -1020,7 +1017,7 @@ use install_report::InstallReporter; mod managed_node; use managed_node::{ ensure_managed_node_runtime_blocking, managed_node_runtime_supported, managed_npm_command, - npm_eacces_hint, + npm_eacces_hint, resolve_adapter_path, }; #[tauri::command] @@ -1741,7 +1738,7 @@ mod tests { #[test] fn test_powershell_command_argv_exact() { // Catalog format: body wrapped in one outer double-quote pair (Bash-layer serialization). - let body = "irm https://chatgpt.com/codex/install.ps1 | iex"; + let body = "$ErrorActionPreference='Stop'; $installer=Join-Path $env:TEMP 'buzz-install-codex.ps1'; Invoke-RestMethod https://chatgpt.com/codex/install.ps1 -OutFile $installer; & $installer; exit $LASTEXITCODE"; let cmd = super::install_powershell_command(&format!( r#"powershell.exe -NoProfile -ExecutionPolicy Bypass -Command "{body}""# )); @@ -1771,12 +1768,12 @@ mod tests { ); } - /// Claude Code catalog command (discovery.rs:107) must dequote to the bare pipeline. + /// Claude Code catalog command must dequote to the two-step download-then-execute body. #[cfg(windows)] #[test] fn test_powershell_command_claude_catalog_dequoted() { let cmd = super::install_powershell_command( - r#"powershell.exe -NoProfile -ExecutionPolicy Bypass -Command "irm https://claude.ai/install.ps1 | iex""#, + r#"powershell.exe -NoProfile -ExecutionPolicy Bypass -Command "$ErrorActionPreference='Stop'; $installer=Join-Path $env:TEMP 'buzz-install-claude.ps1'; Invoke-RestMethod https://claude.ai/install.ps1 -OutFile $installer; & $installer; exit $LASTEXITCODE""#, ); assert_eq!( cmd.get_args() @@ -1787,22 +1784,22 @@ mod tests { "-ExecutionPolicy", "Bypass", "-Command", - "irm https://claude.ai/install.ps1 | iex", + "$ErrorActionPreference='Stop'; $installer=Join-Path $env:TEMP 'buzz-install-claude.ps1'; Invoke-RestMethod https://claude.ai/install.ps1 -OutFile $installer; & $installer; exit $LASTEXITCODE", ], "Claude catalog command must be dequoted correctly" ); } - /// Goose Windows catalog command (discovery.rs:78) must dequote to a bare pipeline - /// with a literal `$env:` prefix — no backslash before the dollar sign. - /// This proves the `\$` → `$` escape fix: post-#2750 the spawn is native and + /// Goose Windows catalog command must dequote to the two-step download-then-execute body + /// with the `$env:CONFIGURE` prefix intact — no backslash before the dollar sign. + /// This proves the `\$` → `$` contract: post-#2750 the spawn is native and /// PowerShell receives the body verbatim, so a residual `\` would produce /// `\$env:CONFIGURE='false'` which is a malformed statement. #[cfg(windows)] #[test] fn test_powershell_command_goose_catalog_dequoted() { let cmd = super::install_powershell_command( - r#"powershell.exe -NoProfile -ExecutionPolicy Bypass -Command "$env:CONFIGURE='false'; irm https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1 | iex""#, + r#"powershell.exe -NoProfile -ExecutionPolicy Bypass -Command "$env:CONFIGURE='false'; $ErrorActionPreference='Stop'; $installer=Join-Path $env:TEMP 'buzz-install-goose.ps1'; Invoke-RestMethod https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1 -OutFile $installer; & $installer; exit $LASTEXITCODE""#, ); assert_eq!( cmd.get_args() @@ -1813,7 +1810,7 @@ mod tests { "-ExecutionPolicy", "Bypass", "-Command", - "$env:CONFIGURE='false'; irm https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1 | iex", + "$env:CONFIGURE='false'; $ErrorActionPreference='Stop'; $installer=Join-Path $env:TEMP 'buzz-install-goose.ps1'; Invoke-RestMethod https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1 -OutFile $installer; & $installer; exit $LASTEXITCODE", ], "Goose catalog command must dequote with bare $env: (no backslash before $)" ); diff --git a/desktop/src-tauri/src/commands/agent_discovery/managed_node.rs b/desktop/src-tauri/src/commands/agent_discovery/managed_node.rs index 72108f0291..fbfb068c0e 100644 --- a/desktop/src-tauri/src/commands/agent_discovery/managed_node.rs +++ b/desktop/src-tauri/src/commands/agent_discovery/managed_node.rs @@ -102,25 +102,155 @@ fn managed_node_failed_step(stderr: String) -> InstallStepResult { } } -fn managed_node_runtime_ready() -> bool { +pub(super) fn managed_node_runtime_ready() -> bool { let Some(node) = crate::managed_agents::buzz_managed_node_bin_path() else { return false; }; if !node.is_file() { return false; } - let mut cmd = std::process::Command::new(&node); + probe_node(&node, MANAGED_NODE_VERSION, Duration::from_secs(3)) +} + +/// Run `executable --version` with a bounded deadline and return `true` only +/// when it exits 0 and its trimmed stdout equals `expected_version`. +/// +/// Transport: stdout is redirected to a temp file so no exit path can block on +/// an inherited handle (a descendant retaining a pipe write-end would otherwise +/// prevent EOF indefinitely). +/// +/// Cleanup: the child runs in its own process group on Unix (`process_group(0)`) +/// so an unconditional group SIGKILL on every exit path terminates all +/// descendants. On Windows, `terminate_process` issues `taskkill /T /F` for +/// tree-wide cleanup. SIGKILL to an already-dead group returns ESRCH (no-op). +pub(super) fn probe_node( + executable: &std::path::Path, + expected_version: &str, + timeout: Duration, +) -> bool { + let tmp = match tempfile::NamedTempFile::new() { + Ok(f) => f, + Err(_) => return false, + }; + let out_file = match tmp.reopen() { + Ok(f) => f, + Err(_) => return false, + }; + + let mut cmd = std::process::Command::new(executable); cmd.arg("--version") .stdin(std::process::Stdio::null()) - .stdout(std::process::Stdio::piped()) + .stdout(std::process::Stdio::from(out_file)) .stderr(std::process::Stdio::null()); crate::util::configure_no_window(&mut cmd); - let output = cmd.output(); - output - .ok() - .filter(|output| output.status.success()) - .map(|output| String::from_utf8_lossy(&output.stdout).trim() == MANAGED_NODE_VERSION) - .unwrap_or(false) + #[cfg(unix)] + { + use std::os::unix::process::CommandExt; + cmd.process_group(0); + } + let Ok(mut child) = cmd.spawn() else { + return false; + }; + + let deadline = std::time::Instant::now() + timeout; + let exit_status = loop { + match child.try_wait() { + Ok(Some(status)) => break status, + Ok(None) => { + if std::time::Instant::now() >= deadline { + kill_probe_group(child.id()); + let _ = child.wait(); + return false; + } + std::thread::sleep(Duration::from_millis(50)); + } + Err(_) => { + kill_probe_group(child.id()); + let _ = child.wait(); + return false; + } + } + }; + + // Group-kill unconditionally: SIGKILL to a dead group is ESRCH (no-op). + kill_probe_group(child.id()); + + if !exit_status.success() { + return false; + } + + let mut output = String::new(); + if std::io::Read::read_to_string(&mut tmp.as_file(), &mut output).is_err() { + return false; + } + output.trim() == expected_version +} + +/// Kill the probe's process group/tree unconditionally (no TERM grace — this +/// is a probe, not an agent session). ESRCH on a dead group is fine. +fn kill_probe_group(pid: u32) { + #[cfg(unix)] + unsafe { + libc::kill(-(pid as i32), libc::SIGKILL); + } + #[cfg(windows)] + { + let _ = crate::managed_agents::terminate_process(pid); + } + #[cfg(not(any(unix, windows)))] + { + let _ = pid; + } +} + +/// Returns `true` when the managed Node runtime is absent or no longer executes — +/// meaning any existing npm adapter shims are broken and must be reinstalled. +/// +/// This fires when the pinned Node version changes (e.g. v24.11.0 → v24.18.0): +/// the old dir stays on disk, shims appear installed, but they fail at run time +/// because the Node binary they reference is gone. Treating the adapter as +/// missing forces `ensure_managed_node_runtime_blocking` to re-download Node and +/// npm to reinstall the shims. +pub(super) fn managed_node_orphaned() -> bool { + managed_node_runtime_supported() && !managed_node_runtime_ready() +} + +/// Returns `true` when an adapter at `resolved` should be invalidated. +/// +/// Only a Buzz-managed shim (path under `managed_prefix`) with an orphaned +/// runtime is invalidated; external adapters are always preserved. +pub(super) fn should_invalidate_adapter( + resolved: &std::path::Path, + managed_prefix: &std::path::Path, + orphaned: bool, +) -> bool { + orphaned && resolved.starts_with(managed_prefix) +} + +/// Resolve the adapter binary path, accounting for the Node-orphan case. +/// Resolves first; invalidates only managed-prefix shims when Node is orphaned. +pub(super) fn resolve_adapter_path( + commands: &[&str], + adapter_install_commands: &[&str], +) -> Option { + let resolved = commands + .iter() + .find_map(|cmd| crate::managed_agents::resolve_command(cmd)); + + let needs_managed_npm = adapter_install_commands + .iter() + .any(|cmd| is_npm_global_install(cmd)); + if needs_managed_npm { + if let (Some(ref path), Some(ref managed_bin)) = + (&resolved, crate::managed_agents::buzz_managed_npm_bin_dir()) + { + if should_invalidate_adapter(path, managed_bin, managed_node_orphaned()) { + return None; + } + } + } + + resolved } fn managed_node_install_lock() -> &'static Mutex<()> { @@ -538,211 +668,5 @@ pub(super) fn npm_eacces_hint(stderr: &str, _command: &str) -> Option { // ── end managed npm adapter installs ────────────────────────────────────────── #[cfg(test)] -mod tests { - use super::*; - - #[test] - fn test_npm_eacces_hint_guidance_mentions_buzz_private_dir() { - let hint = npm_eacces_hint("EACCES: permission denied", "npm install -g foo").unwrap(); - assert!( - hint.contains("Buzz's private Node tools directory"), - "hint: {hint}" - ); - } - - #[test] - fn test_rewrite_npm_install_uses_private_prefix() { - assert_eq!( - rewrite_npm_global_install( - "npm install -g @agentclientprotocol/codex-acp", - "'/tmp/Buzz Node'" - ), - "npm install --global --prefix '/tmp/Buzz Node' @agentclientprotocol/codex-acp" - ); - } - - #[test] - fn test_rewrite_npm_i_uses_private_prefix() { - assert_eq!( - rewrite_npm_global_install("npm i -g some-package", "'/tmp/buzz'"), - "npm i --global --prefix '/tmp/buzz' some-package" - ); - } - - #[test] - fn test_rewrite_npm_uninstall_uses_private_prefix() { - assert_eq!( - rewrite_npm_global_install("npm uninstall -g @zed-industries/codex-acp", "'/tmp/buzz'"), - "npm uninstall --global --prefix '/tmp/buzz' @zed-industries/codex-acp" - ); - } - - #[test] - fn test_rewrite_ignores_non_global_command() { - assert_eq!( - rewrite_npm_global_install("npm install foo", "'/tmp/buzz'"), - "npm install foo" - ); - } - - #[test] - fn test_shell_quote_escapes_single_quotes() { - assert_eq!( - shell_quote(std::path::Path::new("/tmp/Buzz's Node")), - "'/tmp/Buzz'\\''s Node'" - ); - } - - // ── zip validation tests ────────────────────────────────────────────────── - - /// Build an in-memory zip archive with the supplied entry names and return - /// a temporary file containing it (zip::ZipArchive requires Seek). - fn make_zip_with_entries(entry_names: &[&str]) -> tempfile::NamedTempFile { - let mut buf: Vec = Vec::new(); - { - let mut writer = zip::ZipWriter::new(std::io::Cursor::new(&mut buf)); - let opts = zip::write::SimpleFileOptions::default(); - for name in entry_names { - writer.start_file(*name, opts).unwrap(); - } - writer.finish().unwrap(); - } - let mut tmp = tempfile::NamedTempFile::new().unwrap(); - std::io::Write::write_all(&mut tmp, &buf).unwrap(); - tmp - } - - #[test] - fn test_validate_zip_accepts_normal_entries() { - let tmp = make_zip_with_entries(&[ - "node-v24.18.0-win-x64/node.exe", - "node-v24.18.0-win-x64/npm.cmd", - "node-v24.18.0-win-x64/npm", - ]); - let file = std::fs::File::open(tmp.path()).unwrap(); - let archive = zip::ZipArchive::new(file).unwrap(); - assert!(validate_managed_node_zip_entries(&archive).is_ok()); - } - - #[test] - fn test_validate_zip_rejects_absolute_path() { - let tmp = make_zip_with_entries(&["/etc/passwd"]); - let file = std::fs::File::open(tmp.path()).unwrap(); - let archive = zip::ZipArchive::new(file).unwrap(); - let err = validate_managed_node_zip_entries(&archive).unwrap_err(); - assert!( - err.contains("absolute path"), - "expected 'absolute path' in: {err}" - ); - } - - #[test] - fn test_validate_zip_rejects_path_traversal() { - let tmp = make_zip_with_entries(&["../../../etc/passwd"]); - let file = std::fs::File::open(tmp.path()).unwrap(); - let archive = zip::ZipArchive::new(file).unwrap(); - let err = validate_managed_node_zip_entries(&archive).unwrap_err(); - assert!( - err.contains("path traversal"), - "expected 'path traversal' in: {err}" - ); - } - - #[test] - fn test_validate_zip_rejects_backslash_rooted() { - // Windows-style absolute path using backslash — must reject on every host. - let tmp = make_zip_with_entries(&["\\Windows\\system32\\evil.dll"]); - let file = std::fs::File::open(tmp.path()).unwrap(); - let archive = zip::ZipArchive::new(file).unwrap(); - let err = validate_managed_node_zip_entries(&archive).unwrap_err(); - assert!( - err.contains("absolute path"), - "expected 'absolute path' in: {err}" - ); - } - - #[test] - fn test_validate_zip_rejects_drive_prefix() { - // Windows drive-letter absolute path — must reject on every host. - let tmp = make_zip_with_entries(&["C:\\evil\\payload.exe"]); - let file = std::fs::File::open(tmp.path()).unwrap(); - let archive = zip::ZipArchive::new(file).unwrap(); - let err = validate_managed_node_zip_entries(&archive).unwrap_err(); - assert!( - err.contains("absolute path"), - "expected 'absolute path' in: {err}" - ); - } - - #[test] - fn test_validate_zip_rejects_backslash_traversal() { - // Path traversal using Windows separator — must reject on every host. - let tmp = make_zip_with_entries(&["node-v24.18.0-win-x64\\..\\..\\evil"]); - let file = std::fs::File::open(tmp.path()).unwrap(); - let archive = zip::ZipArchive::new(file).unwrap(); - let err = validate_managed_node_zip_entries(&archive).unwrap_err(); - assert!( - err.contains("path traversal"), - "expected 'path traversal' in: {err}" - ); - } - - // ── verify_node_tree layout tests ───────────────────────────────────────── - - #[test] - fn test_verify_node_tree_unix_layout_passes() { - let tmp = tempfile::TempDir::new().unwrap(); - let bin = tmp.path().join("bin"); - std::fs::create_dir_all(&bin).unwrap(); - std::fs::write(bin.join("node"), b"").unwrap(); - std::fs::write(bin.join("npm"), b"").unwrap(); - // On non-Windows the unix branch is active — this must pass. - #[cfg(not(windows))] - assert!(verify_node_tree(tmp.path()).is_ok()); - // On Windows the windows branch is active — unix layout must fail. - #[cfg(windows)] - assert!(verify_node_tree(tmp.path()).is_err()); - } - - #[test] - fn test_verify_node_tree_unix_layout_missing_npm_fails() { - let tmp = tempfile::TempDir::new().unwrap(); - let bin = tmp.path().join("bin"); - std::fs::create_dir_all(&bin).unwrap(); - std::fs::write(bin.join("node"), b"").unwrap(); - // npm intentionally absent - #[cfg(not(windows))] - { - let err = verify_node_tree(tmp.path()).unwrap_err(); - assert!(err.contains("bin/npm"), "err: {err}"); - } - } - - #[test] - fn test_verify_node_tree_windows_layout_passes() { - let tmp = tempfile::TempDir::new().unwrap(); - std::fs::write(tmp.path().join("node.exe"), b"").unwrap(); - std::fs::write(tmp.path().join("npm.cmd"), b"").unwrap(); - std::fs::write(tmp.path().join("npm"), b"").unwrap(); - // On Windows the windows branch is active — this must pass. - #[cfg(windows)] - assert!(verify_node_tree(tmp.path()).is_ok()); - // On non-Windows the unix branch is active — windows-layout root files - // don't satisfy bin/node + bin/npm, so this must fail. - #[cfg(not(windows))] - assert!(verify_node_tree(tmp.path()).is_err()); - } - - #[test] - fn test_verify_node_tree_windows_layout_missing_npm_shim_fails() { - let tmp = tempfile::TempDir::new().unwrap(); - std::fs::write(tmp.path().join("node.exe"), b"").unwrap(); - std::fs::write(tmp.path().join("npm.cmd"), b"").unwrap(); - // npm POSIX shim intentionally absent - #[cfg(windows)] - { - let err = verify_node_tree(tmp.path()).unwrap_err(); - assert!(err.contains("npm"), "err: {err}"); - } - } -} +#[path = "managed_node_tests.rs"] +mod tests; diff --git a/desktop/src-tauri/src/commands/agent_discovery/managed_node_tests.rs b/desktop/src-tauri/src/commands/agent_discovery/managed_node_tests.rs new file mode 100644 index 0000000000..a8e1d7f4c8 --- /dev/null +++ b/desktop/src-tauri/src/commands/agent_discovery/managed_node_tests.rs @@ -0,0 +1,481 @@ +use super::*; + +#[test] +fn test_npm_eacces_hint_guidance_mentions_buzz_private_dir() { + let hint = npm_eacces_hint("EACCES: permission denied", "npm install -g foo").unwrap(); + assert!( + hint.contains("Buzz's private Node tools directory"), + "hint: {hint}" + ); +} + +#[test] +fn test_rewrite_npm_install_uses_private_prefix() { + assert_eq!( + rewrite_npm_global_install( + "npm install -g @agentclientprotocol/codex-acp", + "'/tmp/Buzz Node'" + ), + "npm install --global --prefix '/tmp/Buzz Node' @agentclientprotocol/codex-acp" + ); +} + +#[test] +fn test_rewrite_npm_i_uses_private_prefix() { + assert_eq!( + rewrite_npm_global_install("npm i -g some-package", "'/tmp/buzz'"), + "npm i --global --prefix '/tmp/buzz' some-package" + ); +} + +#[test] +fn test_rewrite_npm_uninstall_uses_private_prefix() { + assert_eq!( + rewrite_npm_global_install("npm uninstall -g @zed-industries/codex-acp", "'/tmp/buzz'"), + "npm uninstall --global --prefix '/tmp/buzz' @zed-industries/codex-acp" + ); +} + +#[test] +fn test_rewrite_ignores_non_global_command() { + assert_eq!( + rewrite_npm_global_install("npm install foo", "'/tmp/buzz'"), + "npm install foo" + ); +} + +#[test] +fn test_shell_quote_escapes_single_quotes() { + assert_eq!( + shell_quote(std::path::Path::new("/tmp/Buzz's Node")), + "'/tmp/Buzz'\\''s Node'" + ); +} + +// ── zip validation tests ────────────────────────────────────────────────────── + +/// Build an in-memory zip archive with the supplied entry names and return +/// a temporary file containing it (zip::ZipArchive requires Seek). +fn make_zip_with_entries(entry_names: &[&str]) -> tempfile::NamedTempFile { + let mut buf: Vec = Vec::new(); + { + let mut writer = zip::ZipWriter::new(std::io::Cursor::new(&mut buf)); + let opts = zip::write::SimpleFileOptions::default(); + for name in entry_names { + writer.start_file(*name, opts).unwrap(); + } + writer.finish().unwrap(); + } + let mut tmp = tempfile::NamedTempFile::new().unwrap(); + std::io::Write::write_all(&mut tmp, &buf).unwrap(); + tmp +} + +#[test] +fn test_validate_zip_accepts_normal_entries() { + let tmp = make_zip_with_entries(&[ + "node-v24.18.0-win-x64/node.exe", + "node-v24.18.0-win-x64/npm.cmd", + "node-v24.18.0-win-x64/npm", + ]); + let file = std::fs::File::open(tmp.path()).unwrap(); + let archive = zip::ZipArchive::new(file).unwrap(); + assert!(validate_managed_node_zip_entries(&archive).is_ok()); +} + +#[test] +fn test_validate_zip_rejects_absolute_path() { + let tmp = make_zip_with_entries(&["/etc/passwd"]); + let file = std::fs::File::open(tmp.path()).unwrap(); + let archive = zip::ZipArchive::new(file).unwrap(); + let err = validate_managed_node_zip_entries(&archive).unwrap_err(); + assert!( + err.contains("absolute path"), + "expected 'absolute path' in: {err}" + ); +} + +#[test] +fn test_validate_zip_rejects_path_traversal() { + let tmp = make_zip_with_entries(&["../../../etc/passwd"]); + let file = std::fs::File::open(tmp.path()).unwrap(); + let archive = zip::ZipArchive::new(file).unwrap(); + let err = validate_managed_node_zip_entries(&archive).unwrap_err(); + assert!( + err.contains("path traversal"), + "expected 'path traversal' in: {err}" + ); +} + +#[test] +fn test_validate_zip_rejects_backslash_rooted() { + // Windows-style absolute path using backslash — must reject on every host. + let tmp = make_zip_with_entries(&["\\Windows\\system32\\evil.dll"]); + let file = std::fs::File::open(tmp.path()).unwrap(); + let archive = zip::ZipArchive::new(file).unwrap(); + let err = validate_managed_node_zip_entries(&archive).unwrap_err(); + assert!( + err.contains("absolute path"), + "expected 'absolute path' in: {err}" + ); +} + +#[test] +fn test_validate_zip_rejects_drive_prefix() { + // Windows drive-letter absolute path — must reject on every host. + let tmp = make_zip_with_entries(&["C:\\evil\\payload.exe"]); + let file = std::fs::File::open(tmp.path()).unwrap(); + let archive = zip::ZipArchive::new(file).unwrap(); + let err = validate_managed_node_zip_entries(&archive).unwrap_err(); + assert!( + err.contains("absolute path"), + "expected 'absolute path' in: {err}" + ); +} + +#[test] +fn test_validate_zip_rejects_backslash_traversal() { + // Path traversal using Windows separator — must reject on every host. + let tmp = make_zip_with_entries(&["node-v24.18.0-win-x64\\..\\..\\evil"]); + let file = std::fs::File::open(tmp.path()).unwrap(); + let archive = zip::ZipArchive::new(file).unwrap(); + let err = validate_managed_node_zip_entries(&archive).unwrap_err(); + assert!( + err.contains("path traversal"), + "expected 'path traversal' in: {err}" + ); +} + +// ── verify_node_tree layout tests ───────────────────────────────────────────── + +#[test] +fn test_verify_node_tree_unix_layout_passes() { + let tmp = tempfile::TempDir::new().unwrap(); + let bin = tmp.path().join("bin"); + std::fs::create_dir_all(&bin).unwrap(); + std::fs::write(bin.join("node"), b"").unwrap(); + std::fs::write(bin.join("npm"), b"").unwrap(); + // On non-Windows the unix branch is active — this must pass. + #[cfg(not(windows))] + assert!(verify_node_tree(tmp.path()).is_ok()); + // On Windows the windows branch is active — unix layout must fail. + #[cfg(windows)] + assert!(verify_node_tree(tmp.path()).is_err()); +} + +#[test] +fn test_verify_node_tree_unix_layout_missing_npm_fails() { + let tmp = tempfile::TempDir::new().unwrap(); + let bin = tmp.path().join("bin"); + std::fs::create_dir_all(&bin).unwrap(); + std::fs::write(bin.join("node"), b"").unwrap(); + // npm intentionally absent + #[cfg(not(windows))] + { + let err = verify_node_tree(tmp.path()).unwrap_err(); + assert!(err.contains("bin/npm"), "err: {err}"); + } +} + +#[test] +fn test_verify_node_tree_windows_layout_passes() { + let tmp = tempfile::TempDir::new().unwrap(); + std::fs::write(tmp.path().join("node.exe"), b"").unwrap(); + std::fs::write(tmp.path().join("npm.cmd"), b"").unwrap(); + std::fs::write(tmp.path().join("npm"), b"").unwrap(); + // On Windows the windows branch is active — this must pass. + #[cfg(windows)] + assert!(verify_node_tree(tmp.path()).is_ok()); + // On non-Windows the unix branch is active — windows-layout root files + // don't satisfy bin/node + bin/npm, so this must fail. + #[cfg(not(windows))] + assert!(verify_node_tree(tmp.path()).is_err()); +} + +#[test] +fn test_verify_node_tree_windows_layout_missing_npm_shim_fails() { + let tmp = tempfile::TempDir::new().unwrap(); + std::fs::write(tmp.path().join("node.exe"), b"").unwrap(); + std::fs::write(tmp.path().join("npm.cmd"), b"").unwrap(); + // npm POSIX shim intentionally absent + #[cfg(windows)] + { + let err = verify_node_tree(tmp.path()).unwrap_err(); + assert!(err.contains("npm"), "err: {err}"); + } +} + +// ── should_invalidate_adapter / orphan policy pure unit tests ───────────────── + +#[test] +fn test_should_invalidate_adapter_invalidates_managed_shim_when_orphaned() { + let prefix = std::path::Path::new("/managed/npm/bin"); + let shim = prefix.join("codex-acp"); + assert!( + should_invalidate_adapter(&shim, prefix, true), + "managed shim + orphaned runtime must be invalidated" + ); +} + +#[test] +fn test_should_invalidate_adapter_keeps_external_adapter_when_orphaned() { + let prefix = std::path::Path::new("/managed/npm/bin"); + let external = std::path::Path::new("/usr/local/bin/codex-acp"); + assert!( + !should_invalidate_adapter(external, prefix, true), + "external adapter must not be invalidated even when Node is orphaned" + ); +} + +#[test] +fn test_should_invalidate_adapter_keeps_managed_shim_when_node_healthy() { + let prefix = std::path::Path::new("/managed/npm/bin"); + let shim = prefix.join("codex-acp"); + assert!( + !should_invalidate_adapter(&shim, prefix, false), + "managed shim must not be invalidated when Node is healthy" + ); +} + +#[test] +fn test_resolve_adapter_path_returns_none_when_binary_absent() { + let commands: &[&str] = &["nonexistent-buzz-test-binary-xyz"]; + let adapter_install_commands: &[&str] = &["curl -fsSL https://example.com | bash"]; + assert!( + resolve_adapter_path(commands, adapter_install_commands).is_none(), + "must return None when the command is not on PATH" + ); +} + +// ── probe_node seam regressions ─────────────────────────────────────────────── +// +// All four scenarios drive probe_node() directly — the same +// tempfile/deadline/cleanup/status/version path used by managed_node_runtime_ready. +// Each test CAN fail if production: +// - drops process_group(0) → descendant-holds-stdout assertion (d) fails +// (tempfile transport still returns promptly; +// the sleep survives and kill($!,0) returns 0) +// - skips group-kill on a path → hung-binary test exceeds margin +// - ignores exit_status.success() → nonzero-exit test returns true +// - skips version comparison → wrong-version test returns true +// +// Script files are written into a TempDir (no open write fd at spawn time) +// to avoid ETXTBSY on Linux. + +/// Scenario 1 — descendant holds stdout write-end, direct child exits immediately. +/// +/// The script backgrounds a 60-second sleep (inheriting stdout), records both the +/// script's own PID (`$$`) and the sleep's PID (`$!`) to sidecar files, then exits +/// with the expected version string. Four assertions: +/// (a) bounded return — would hang ~60 s if tempfile transport regressed to pipe; +/// (b) correct result; +/// (c) process group dead after return — catches skipped-cleanup-on-success: if +/// the group-kill is absent but `process_group(0)` is still present, a live +/// member in the group is detectable; +/// (d) descendant PID dead after return — catches dropped `process_group(0)`: if +/// the call is removed the sleep stays in the runner's group (not the probe's), +/// `kill(-pgid,0)` is vacuously ESRCH, but `kill(desc_pid,0)` returns 0 and +/// this assertion fails. This is the canonical mutation for (c). +#[cfg(unix)] +#[test] +fn test_probe_node_descendant_holds_stdout_returns_promptly_and_kills_group() { + use std::os::unix::fs::PermissionsExt; + let tmp_dir = tempfile::TempDir::new().unwrap(); + let script = tmp_dir.path().join("probe.sh"); + let pgid_file = tmp_dir.path().join("pgid"); + let desc_pid_file = tmp_dir.path().join("desc_pid"); + let pgid_file_path = pgid_file.to_str().unwrap().to_owned(); + let desc_pid_file_path = desc_pid_file.to_str().unwrap().to_owned(); + // Line 1 of script: record the script's own PID (= PGID after process_group(0)). + // Line 2: background the sleep and record its PID. + // Line 3: emit the expected version and exit so the direct child exits promptly. + let script_content = format!( + "#!/bin/sh\necho $$ > {pgid_file_path}\n/bin/sleep 60 &\necho $! > {desc_pid_file_path}\necho v24.18.0\nexit 0\n" + ); + std::fs::write(&script, script_content.as_bytes()).unwrap(); + std::fs::set_permissions(&script, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let probe_timeout = std::time::Duration::from_secs(3); + let t = std::time::Instant::now(); + let result = probe_node(&script, "v24.18.0", probe_timeout); + let elapsed = t.elapsed(); + + // Give the group-kill a moment to propagate before checking liveness. + std::thread::sleep(std::time::Duration::from_millis(200)); + + // (a) bounded return — tempfile transport must not hang on the descendant's + // retained pipe write-end. + assert!( + elapsed < probe_timeout + std::time::Duration::from_secs(2), + "probe_node hung — likely descendant retained pipe write-end: elapsed {elapsed:?}" + ); + // (b) correct result + assert!(result, "probe_node must return true for matching version"); + + // (c) process group dead — catches skipped-cleanup: if the group-kill on the + // success path is removed while process_group(0) is still present, the + // sleep remains in the probe's group and kill(-pgid,0) returns 0. + let pgid_str = std::fs::read_to_string(&pgid_file) + .expect("script must have written its PID to the pgid sidecar"); + let pgid: i32 = pgid_str + .trim() + .parse() + .expect("pgid sidecar must contain a numeric PID"); + let group_alive = unsafe { libc::kill(-pgid, 0) } == 0; + assert!( + !group_alive, + "process group {pgid} must be dead after probe_node" + ); + + // (d) descendant PID dead — catches dropped process_group(0): without that + // call the sleep is never in the probe's group, so kill(-pgid,0) is + // vacuously ESRCH while the sleep survives. Asserting the descendant's + // own PID is dead proves the sleep was actually killed. + let desc_pid_str = std::fs::read_to_string(&desc_pid_file) + .expect("script must have written the sleep PID to the desc_pid sidecar"); + let desc_pid: libc::pid_t = desc_pid_str + .trim() + .parse() + .expect("desc_pid sidecar must contain a numeric PID"); + // Pre-assert cleanup: if the descendant is somehow still alive, kill it so + // a failing test does not leave a 60-second sleep in the runner's process group. + let desc_alive = unsafe { libc::kill(desc_pid, 0) } == 0; + if desc_alive { + unsafe { libc::kill(desc_pid, libc::SIGKILL) }; + } + assert!( + !desc_alive, + "descendant PID {desc_pid} must be dead after probe_node — \ + if process_group(0) is dropped, the sleep escapes into the runner's group \ + and is never killed by the group-kill" + ); +} + +/// Scenario 2 — direct hang: probe_node must traverse the real try_wait +/// deadline and return false. Does NOT call kill_probe_group directly. +/// +/// This test FAILS if the deadline loop in probe_node is broken or if the +/// timeout/kill path is not exercised (e.g., missing group-kill exits the +/// loop early via a different mechanism). +#[cfg(unix)] +#[test] +fn test_probe_node_times_out_on_hung_binary() { + use std::os::unix::fs::PermissionsExt; + let tmp_dir = tempfile::TempDir::new().unwrap(); + let script = tmp_dir.path().join("hung.sh"); + std::fs::write(&script, b"#!/bin/sh\n/bin/sleep 30\n").unwrap(); + std::fs::set_permissions(&script, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let probe_timeout = std::time::Duration::from_secs(3); + let t = std::time::Instant::now(); + let result = probe_node(&script, "v24.18.0", probe_timeout); + let elapsed = t.elapsed(); + + assert!(!result, "probe_node must return false for a hung binary"); + // Must have traversed the deadline (not returned early via a bug). + assert!( + elapsed >= probe_timeout, + "probe_node returned before deadline: {elapsed:?} < {probe_timeout:?}" + ); + // Must not hang past the deadline by more than the poll interval + margin. + assert!( + elapsed < probe_timeout + std::time::Duration::from_secs(3), + "probe_node exceeded deadline by too much: {elapsed:?}" + ); +} + +/// Scenario 3 — non-zero exit: probe_node must return false even when stdout +/// contains the expected version string. +/// +/// This test FAILS if probe_node skips or inverts the exit_status.success() check. +#[cfg(unix)] +#[test] +fn test_probe_node_returns_false_on_nonzero_exit() { + use std::os::unix::fs::PermissionsExt; + let tmp_dir = tempfile::TempDir::new().unwrap(); + let script = tmp_dir.path().join("fail.sh"); + // Prints the expected version string but exits non-zero. + std::fs::write(&script, b"#!/bin/sh\necho v24.18.0\nexit 1\n").unwrap(); + std::fs::set_permissions(&script, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let result = probe_node(&script, "v24.18.0", std::time::Duration::from_secs(3)); + assert!( + !result, + "probe_node must return false when the process exits non-zero" + ); +} + +/// Scenario 4 — wrong version output: probe_node must return false when stdout +/// does not match expected_version. +/// +/// This test FAILS if probe_node skips or incorrectly performs the version comparison. +#[cfg(unix)] +#[test] +fn test_probe_node_returns_false_on_wrong_version_output() { + use std::os::unix::fs::PermissionsExt; + let tmp_dir = tempfile::TempDir::new().unwrap(); + let script = tmp_dir.path().join("wrongver.sh"); + std::fs::write(&script, b"#!/bin/sh\necho v99.0.0\nexit 0\n").unwrap(); + std::fs::set_permissions(&script, std::fs::Permissions::from_mode(0o755)).unwrap(); + + let result = probe_node(&script, "v24.18.0", std::time::Duration::from_secs(3)); + assert!( + !result, + "probe_node must return false when stdout version does not match expected" + ); +} + +/// Windows-shaped seam: non-zero exit via .bat file. +/// +/// Drives the same probe_node path on Windows (terminate_process / taskkill /T /F). +/// This test FAILS if probe_node ignores exit_status.success() on Windows. +#[cfg(windows)] +#[test] +fn test_probe_node_windows_returns_false_on_nonzero_exit() { + let tmp_dir = tempfile::TempDir::new().unwrap(); + let bat = tmp_dir.path().join("fail.bat"); + // Prints the expected version but exits non-zero — must still fail. + std::fs::write(&bat, b"@echo off\r\necho v24.18.0\r\nexit /b 1\r\n").unwrap(); + + let result = probe_node(&bat, "v24.18.0", std::time::Duration::from_secs(3)); + assert!( + !result, + "probe_node must return false when the .bat exits non-zero (Windows)" + ); +} + +/// Windows-shaped seam: wrong version output via .bat file. +/// +/// This test FAILS if probe_node skips the version comparison on Windows. +#[cfg(windows)] +#[test] +fn test_probe_node_windows_returns_false_on_wrong_version_output() { + let tmp_dir = tempfile::TempDir::new().unwrap(); + let bat = tmp_dir.path().join("wrongver.bat"); + std::fs::write(&bat, b"@echo off\r\necho v99.0.0\r\nexit /b 0\r\n").unwrap(); + + let result = probe_node(&bat, "v24.18.0", std::time::Duration::from_secs(3)); + assert!( + !result, + "probe_node must return false when stdout version does not match (Windows)" + ); +} + +/// Returns false when the node binary path does not exist (fast path, no spawn). +#[test] +fn test_managed_node_runtime_ready_returns_false_when_binary_absent() { + let Some(node) = crate::managed_agents::buzz_managed_node_bin_path() else { + assert!( + !managed_node_runtime_ready(), + "managed_node_runtime_ready must return false when no path resolves" + ); + return; + }; + if node.is_file() { + return; + } + assert!( + !managed_node_runtime_ready(), + "managed_node_runtime_ready must return false when the binary file does not exist" + ); +} diff --git a/desktop/src-tauri/src/commands/agents.rs b/desktop/src-tauri/src/commands/agents.rs index 0758fc3aac..3b114b0474 100644 --- a/desktop/src-tauri/src/commands/agents.rs +++ b/desktop/src-tauri/src/commands/agents.rs @@ -1362,7 +1362,7 @@ use deploy::build_deploy_payload; #[cfg(test)] use deploy::deploy_payload_json; #[cfg(test)] -pub(crate) use deploy::resolve_deploy_model_provider; +use deploy::{ensure_remote_provider_supported, resolve_deploy_model_provider}; #[path = "agents_profile.rs"] mod profile; diff --git a/desktop/src-tauri/src/commands/agents_deploy.rs b/desktop/src-tauri/src/commands/agents_deploy.rs index af785711d5..9bb0f6230d 100644 --- a/desktop/src-tauri/src/commands/agents_deploy.rs +++ b/desktop/src-tauri/src/commands/agents_deploy.rs @@ -1,6 +1,8 @@ //! Provider deploy payload construction, split from `agents.rs` (file-size -//! guard). `build_deploy_payload` gathers live state; `deploy_payload_json` -//! is the pure serialization half so payload completeness stays testable. +//! guard). The launch block is derived from the same effective descriptor and +//! policy helpers as local spawn so remote execution does not reimplement them. + +use std::collections::BTreeMap; use tauri::AppHandle; @@ -13,17 +15,6 @@ use crate::{ }; /// Resolve the deploy-specific structured model/provider for a managed agent. -/// -/// Delegates to the single effective-config resolver which enforces -/// definition-authoritative semantics for linked instances: -/// - **Linked:** definition → global. Stale record bytes are never consulted. -/// - **Definition-less:** instance → global. -/// - **Orphaned:** returns `(None, None)` — spawn is blocked elsewhere. -/// -/// Both local spawn and deploy now use the same resolver, so they can never -/// disagree on what model/provider an agent runs with. -/// -/// Exported `pub(crate)` for unit testing. #[cfg(test)] pub(crate) fn resolve_deploy_model_provider( record: &ManagedAgentRecord, @@ -36,58 +27,116 @@ pub(crate) fn resolve_deploy_model_provider( .unwrap_or((None, None)) } +/// Serialize the portable launch contract shared with provider-backed agents. +/// +/// `descriptor.env` is the authoritative six-layer environment. Policy values +/// are deliberately separate because providers apply them below that layered +/// environment, preserving the local spawn's power-user override semantics. +pub(super) fn build_launch_block( + record: &ManagedAgentRecord, + descriptor: &crate::managed_agents::readiness::EffectiveHarnessDescriptor, + teams: &[crate::managed_agents::TeamRecord], + effective_prompt: Option<&str>, + effective_model: Option<&str>, + owner_pubkey: &str, +) -> serde_json::Value { + use crate::managed_agents::{known_acp_runtime, resolve_session_title, SESSION_TITLE_ENV_VAR}; + + let runtime = known_acp_runtime(&descriptor.command); + let mut policy_env = BTreeMap::new(); + + if let Some(runtime) = runtime { + policy_env.extend( + runtime + .default_env + .iter() + .map(|(key, value)| ((*key).to_string(), (*value).to_string())), + ); + if runtime.mcp_hooks { + policy_env.insert("MCP_HOOK_SERVERS".into(), "*".into()); + } + } + policy_env.insert("BUZZ_ACP_RELAY_OBSERVER".into(), "true".into()); + policy_env.insert("BUZZ_ACP_LAZY_POOL".into(), "true".into()); + policy_env.insert("BUZZ_ACP_AGENTS".into(), record.parallelism.to_string()); + + if let Some(value) = effective_prompt { + policy_env.insert("BUZZ_ACP_SYSTEM_PROMPT".into(), value.to_string()); + } + if let Some(value) = effective_model { + policy_env.insert("BUZZ_ACP_MODEL".into(), value.to_string()); + } + if let Some(value) = record.idle_timeout_seconds { + policy_env.insert("BUZZ_ACP_IDLE_TIMEOUT".into(), value.to_string()); + } + if let Some(value) = record.max_turn_duration_seconds { + policy_env.insert("BUZZ_ACP_MAX_TURN_DURATION".into(), value.to_string()); + } + if let Some(value) = resolve_session_title(record.display_name.as_deref(), &record.name) { + policy_env.insert(SESSION_TITLE_ENV_VAR.into(), value); + } + if let Some(value) = + crate::managed_agents::spawn_hash::effective_team_instructions(record, teams) + { + policy_env.insert("BUZZ_ACP_TEAM_INSTRUCTIONS".into(), value); + } + + serde_json::json!({ + "command": descriptor.command, + "args": descriptor.args, + "env": descriptor.env, + "policy_env": policy_env, + "owner_pubkey": owner_pubkey, + }) +} + +pub(super) fn ensure_remote_provider_supported(provider: Option<&str>) -> Result<(), String> { + if provider.map(str::trim) == Some(crate::managed_agents::RELAY_MESH_PROVIDER_ID) { + return Err( + "shared-compute agents cannot be deployed remotely because the mesh endpoint is local to the desktop" + .to_string(), + ); + } + Ok(()) +} + /// Build the standard agent JSON payload for provider deploy calls. -/// -/// Like local spawn, provider deploy re-reads live persona env vars and -/// structured model/provider so remote agents receive current credentials -/// and the same authoritative values that local spawn derives from -/// `runtime_metadata_env_vars`. The only field still pinned is -/// `agent_command`/`agent_args` — those were captured at create time. -/// The only read-time resolution is `relay_url`: a blank pin resolves to -/// the active workspace relay here, matching the create-path contract. -/// -/// Fails closed when the private key is unavailable (keyring outage leaves -/// it empty after hydration): without this guard a provider deploy would -/// serialize `"private_key_nsec": ""` and launch the agent with no -/// identity — the same hazard the local spawn path refuses via -/// `spawn_key_refusal`. pub(super) fn build_deploy_payload( app: &AppHandle, state: &AppState, record: &ManagedAgentRecord, ) -> Result { - // Fails closed when the private key is unavailable — same guard as local - // spawn. Without this, a keyring outage would serialize `"private_key_nsec": ""` - // and launch the agent with no identity. if let Some(err) = crate::managed_agents::spawn_key_refusal(record) { return Err(err); } - // Merge global + persona + agent env_vars for provider deploy — the same - // live-persona-under-overrides semantics as local spawn. Global env vars - // are the lowest user-settable layer: global < persona < agent (last-wins - // on key collision). Without this, provider-backed agents wouldn't receive - // credentials saved on the persona or the agent itself. - let global_config = crate::managed_agents::load_global_agent_config(app).unwrap_or_default(); - let global_env = global_config.env_vars.clone(); - let persona_env = - crate::managed_agents::resolve_persona_env(app, record.persona_id.as_deref())?; - // Merge: global < persona (persona wins over global). - let global_persona_merged = crate::managed_agents::merged_user_env(&global_env, &persona_env); - // Merge: global+persona < agent (agent wins over everything). - let merged_env = - crate::managed_agents::merged_user_env(&global_persona_merged, &record.env_vars); - + let global = crate::managed_agents::load_global_agent_config(app).unwrap_or_default(); let personas = load_personas(app).unwrap_or_default(); - let cfg = crate::managed_agents::effective_config::resolve_effective_config( - record, - &personas, - &global_config, + let teams = crate::managed_agents::load_teams(app).unwrap_or_default(); + let persona_env = + crate::managed_agents::live_persona_env(&personas, record.persona_id.as_deref()); + let global_persona_env = crate::managed_agents::merged_user_env(&global.env_vars, &persona_env); + let merged_user_env = + crate::managed_agents::merged_user_env(&global_persona_env, &record.env_vars); + let effective = crate::managed_agents::effective_config::resolve_effective_config( + record, &personas, &global, ) .require_resolved()?; - let effective_model = cfg.model.value; - let effective_provider = cfg.provider.value; - let effective_prompt = cfg.system_prompt.value; + + ensure_remote_provider_supported(effective.provider.value.as_deref())?; + + let descriptor = + crate::managed_agents::resolve_effective_harness_descriptor(record, &personas, &global) + .map_err(|error| crate::managed_agents::user_facing_harness_error(&error))?; + let owner_pubkey = super::workspace_owner_hex(state)?; + let launch = build_launch_block( + record, + &descriptor, + &teams, + effective.system_prompt.value.as_deref(), + effective.model.value.as_deref(), + &owner_pubkey, + ); Ok(deploy_payload_json( record, @@ -95,23 +144,24 @@ pub(super) fn build_deploy_payload( &record.relay_url, &relay_ws_url_with_override(state), ), - effective_model, - effective_provider, - effective_prompt, - merged_env, + effective.model.value, + effective.provider.value, + effective.system_prompt.value, + merged_user_env, + launch, )) } -/// Pure serialization half of [`build_deploy_payload`] — every field the -/// provider harness receives is deliberately listed here, so payload -/// completeness is testable without an `AppHandle`. +/// Pure serialization half of [`build_deploy_payload`]. Legacy top-level fields +/// remain for display/bookkeeping; providers execute the resolved `launch` block. pub(super) fn deploy_payload_json( record: &ManagedAgentRecord, relay_url: String, effective_model: Option, effective_provider: Option, effective_prompt: Option, - merged_env: std::collections::BTreeMap, + merged_env: BTreeMap, + launch: serde_json::Value, ) -> serde_json::Value { serde_json::json!({ "name": &record.name, @@ -130,5 +180,81 @@ pub(super) fn deploy_payload_json( "respond_to": record.respond_to, "respond_to_allowlist": &record.respond_to_allowlist, "env_vars": merged_env, + "launch": launch, }) } + +#[cfg(test)] +mod tests { + use super::*; + use crate::managed_agents::{readiness::EffectiveHarnessDescriptor, RespondTo, TeamRecord}; + + fn record() -> ManagedAgentRecord { + serde_json::from_value(serde_json::json!({ + "pubkey": "abcd1234", + "name": "agent-handle", + "display_name": "Agent\u{0000} Name", + "private_key_nsec": "nsec1fake", + "relay_url": "wss://relay.example", + "acp_command": "buzz-acp", + "agent_command": "goose", + "agent_args": [], + "mcp_command": "", + "turn_timeout_seconds": 320, + "idle_timeout_seconds": 17, + "max_turn_duration_seconds": 23, + "parallelism": 4, + "respond_to": RespondTo::OwnerOnly, + "respond_to_allowlist": [], + "team_id": "team-1", + "created_at": "2026-01-01T00:00:00Z", + "updated_at": "2026-01-01T00:00:00Z" + })) + .unwrap() + } + + #[test] + fn launch_block_preserves_descriptor_and_spawn_policy() { + let record = record(); + let descriptor = EffectiveHarnessDescriptor { + command: "goose".into(), + args: vec!["acp".into()], + env: BTreeMap::from([ + ("GOOSE_MODE".into(), "custom".into()), + ("SECRET_FROM_PERSONA".into(), "secret".into()), + ]), + }; + let teams: Vec = serde_json::from_value(serde_json::json!([{ + "id": "team-1", "name": "Team", "instructions": "Coordinate", "persona_ids": [], "created_at": "2026-01-01T00:00:00Z", "updated_at": "2026-01-01T00:00:00Z" + }])).unwrap(); + + let launch = build_launch_block( + &record, + &descriptor, + &teams, + Some("prompt"), + Some("model"), + "owner-hex", + ); + + assert_eq!(launch["command"], "goose"); + assert_eq!(launch["args"], serde_json::json!(["acp"])); + assert_eq!(launch["env"]["GOOSE_MODE"], "custom"); + // policy_env is applied first, so this default remains separate from + // the descriptor value that wins in launch.env. + assert_eq!(launch["policy_env"]["GOOSE_MODE"], "auto"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_LAZY_POOL"], "true"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_RELAY_OBSERVER"], "true"); + assert_eq!( + launch["policy_env"]["BUZZ_ACP_TEAM_INSTRUCTIONS"], + "Coordinate" + ); + assert_eq!(launch["policy_env"]["BUZZ_ACP_SESSION_TITLE"], "Agent Name"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_SYSTEM_PROMPT"], "prompt"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_MODEL"], "model"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_IDLE_TIMEOUT"], "17"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_MAX_TURN_DURATION"], "23"); + assert_eq!(launch["policy_env"]["BUZZ_ACP_AGENTS"], "4"); + assert_eq!(launch["owner_pubkey"], "owner-hex"); + } +} diff --git a/desktop/src-tauri/src/commands/agents_tests.rs b/desktop/src-tauri/src/commands/agents_tests.rs index 03389d1d18..df135298c4 100644 --- a/desktop/src-tauri/src/commands/agents_tests.rs +++ b/desktop/src-tauri/src/commands/agents_tests.rs @@ -263,6 +263,19 @@ fn normalize_relay_mesh_trims_and_preserves_valid_config() { ); } +#[test] +fn deploy_refuses_resolved_relay_mesh_provider_with_padding() { + let record = bare_agent_record(Some("p1"), None, None); + let personas = vec![persona_record("p1", None, Some(" relay-mesh "))]; + let global = crate::managed_agents::GlobalAgentConfig::default(); + + let (_, provider) = resolve_deploy_model_provider(&record, &personas, &global); + let error = ensure_remote_provider_supported(provider.as_deref()) + .expect_err("resolved shared-compute provider must not deploy remotely"); + + assert!(error.contains("cannot be deployed remotely"), "{error}"); +} + #[test] fn created_avatar_prefers_explicit_input() { let resolved = resolve_created_avatar_url( @@ -398,50 +411,93 @@ fn legacy_avatar_empty_when_nothing_resolves() { // ── Provider deploy payload completeness ───────────────────────────────────── -/// Regression (PR #1667 review, Thufir): the provider deploy payload must -/// carry every behavioral field the local spawn path applies — a field -/// missing here silently strips it from provider-backed agents. +/// The shared provider fixture is the contract arbiter: it must be the exact +/// richest deploy request produced by the real desktop serializers. #[test] -fn deploy_payload_carries_the_full_behavioral_quad() { - let allow = "a".repeat(64); - let record: ManagedAgentRecord = serde_json::from_str(&format!( - r#"{{ - "pubkey": "abcd1234", - "name": "test-agent", - "private_key_nsec": "nsec1fake", - "relay_url": "wss://localhost:3000", - "acp_command": "buzz-acp", - "agent_command": "goose", - "agent_args": [], - "mcp_command": "", - "turn_timeout_seconds": 320, - "system_prompt": null, - "parallelism": 4, - "respond_to": "allowlist", - "respond_to_allowlist": ["{allow}"], - "created_at": "2026-01-01T00:00:00Z", - "updated_at": "2026-01-01T00:00:00Z", - "last_started_at": null, - "last_stopped_at": null, - "last_exit_code": null, - "last_error": null - }}"# - )) - .expect("sample record"); - - let payload = deploy_payload_json( +fn deploy_payload_matches_the_shared_full_launch_fixture() { + let fixture_path = std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join( + "../../crates/buzz-backend-kubernetes/tests/fixtures/provider-wire/deploy-full-launch.request.json", + ); + let fixture: serde_json::Value = serde_json::from_str( + &std::fs::read_to_string(&fixture_path) + .unwrap_or_else(|error| panic!("read {}: {error}", fixture_path.display())), + ) + .expect("parse shared provider fixture"); + let record: ManagedAgentRecord = serde_json::from_value(serde_json::json!({ + "pubkey": "abcd1234", + "name": "worker", + "private_key_nsec": "nsec1vl029mgpspedva04g90vltkh6fvh240zqtv9k0t9af8935ke9laqsnlfe5", + "relay_url": "wss://localhost:3000", + "auth_tag": "tag-1", + "acp_command": "buzz-acp", + "agent_command": "goose", + "runtime": "goose", + "model": "gpt-5", + "provider": "openai", + "env_vars": {"USER_KEY": "user-value"}, + "agent_args": [], + "mcp_command": "", + "turn_timeout_seconds": 300, + "system_prompt": null, + "idle_timeout_seconds": null, + "max_turn_duration_seconds": null, + "parallelism": 10, + "respond_to": "allowlist", + "respond_to_allowlist": ["aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"], + "created_at": "2026-01-01T00:00:00Z", + "updated_at": "2026-01-01T00:00:00Z" + })) + .expect("fixture source record"); + let descriptor = crate::managed_agents::resolve_effective_harness_descriptor( &record, - "wss://relay.example".to_string(), - Some("gpt-x".to_string()), - Some("openai".to_string()), + &[], + &crate::managed_agents::GlobalAgentConfig::default(), + ) + .expect("resolve fixture source record descriptor"); + let launch = super::deploy::build_launch_block( + &record, + &descriptor, + &[], None, - std::collections::BTreeMap::new(), + Some("gpt-5"), + "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", + ); + let agent = deploy_payload_json( + &record, + "wss://relay.example".into(), + Some("gpt-5".into()), + Some("openai".into()), + None, + std::collections::BTreeMap::from([("USER_KEY".into(), "user-value".into())]), + launch, ); - assert_eq!(payload["parallelism"], 4); - assert_eq!(payload["respond_to"], "allowlist"); - assert_eq!(payload["respond_to_allowlist"][0], "a".repeat(64)); - assert_eq!(payload["model"], "gpt-x"); - assert_eq!(payload["provider"], "openai"); - assert_eq!(payload["relay_url"], "wss://relay.example"); + assert_eq!( + agent, fixture["agent"], + "desktop payload drifted from the shared provider fixture" + ); +} + +#[test] +fn tauri_platform_configs_bundle_kubernetes_only_on_supported_hosts() { + use tauri_utils::{config::parse::read_from, platform::Target}; + + let config_root = std::path::Path::new(env!("CARGO_MANIFEST_DIR")); + for (target, expected) in [ + (Target::MacOS, true), + (Target::Linux, true), + (Target::Windows, false), + ] { + let (config, paths) = read_from(target, config_root).expect("read Tauri config"); + let external_bins = config["bundle"]["externalBin"] + .as_array() + .expect("bundle.externalBin array"); + let has_kubernetes = external_bins + .iter() + .any(|value| value == "binaries/buzz-backend-kubernetes"); + assert_eq!( + has_kubernetes, expected, + "unexpected Kubernetes externalBin for {target}; merged {paths:?}" + ); + } } diff --git a/desktop/src-tauri/src/commands/export_util.rs b/desktop/src-tauri/src/commands/export_util.rs index ded14679c1..e12cbd19e1 100644 --- a/desktop/src-tauri/src/commands/export_util.rs +++ b/desktop/src-tauri/src/commands/export_util.rs @@ -35,8 +35,8 @@ pub async fn pick_save_path( /// user cancelled the dialog. /// /// NOT for secrets: the write is plain `std::fs::write` (no atomic commit, no -/// 0o600). Secret exports go through `pick_save_path` + -/// `key_backup::write_backup_file`. +/// 0o600). Secret exports go through `pick_save_path` and a dedicated +/// secret-file writer such as `key_backup::write_portable_backup_file`. pub async fn save_bytes_with_dialog( app: &AppHandle, suggested_filename: &str, diff --git a/desktop/src-tauri/src/commands/identity.rs b/desktop/src-tauri/src/commands/identity.rs index 33ecf3cfca..bddf2e725a 100644 --- a/desktop/src-tauri/src/commands/identity.rs +++ b/desktop/src-tauri/src/commands/identity.rs @@ -297,9 +297,10 @@ pub async fn verify_ncryptsec_backup( /// Save a portable copy of an `ncryptsec1…` backup to a user-chosen path. /// /// The input must parse as a structurally valid NIP-49 payload. The dialog is -/// selection-only; the write uses secret-file semantics (atomic + 0o600). -/// Never mutates canonical app state. Returns the chosen path, or `None` when -/// the user cancelled. +/// selection-only; the write uses the exact save-panel-authorized path with +/// owner-only permissions, sync, and reread verification. Existing files are +/// preserved rather than truncated. Never mutates canonical app state. Returns +/// the chosen path, or `None` when the user cancelled. #[tauri::command] pub async fn save_ncryptsec_copy( ncryptsec: String, @@ -324,7 +325,7 @@ pub async fn save_ncryptsec_copy( let dest_for_write = dest.clone(); tokio::task::spawn_blocking(move || { - crate::key_backup::write_backup_file(&dest_for_write, &normalized) + crate::key_backup::write_portable_backup_file(&dest_for_write, &normalized) }) .await .map_err(|e| format!("spawn_blocking failed: {e}"))??; diff --git a/desktop/src-tauri/src/commands/media_download.rs b/desktop/src-tauri/src/commands/media_download.rs index d3b1a9499d..7bc94da25d 100644 --- a/desktop/src-tauri/src/commands/media_download.rs +++ b/desktop/src-tauri/src/commands/media_download.rs @@ -8,7 +8,8 @@ use crate::commands::export_util::save_bytes_with_dialog; use crate::commands::media::{detect_and_validate_mime, mint_media_get_auth, sanitize_filename}; use crate::commands::{ personas::{ - decode_snapshot_from_bytes, MAX_SNAPSHOT_JSON_BYTES, MAX_SNAPSHOT_PNG_BYTES, PNG_MAGIC, + parse_snapshot_payload_from_bytes, MAX_SNAPSHOT_JSON_BYTES, MAX_SNAPSHOT_PNG_BYTES, + PNG_MAGIC, }, team_snapshot::{ decode_team_snapshot_from_bytes, MAX_TEAM_SNAPSHOT_JSON_BYTES, MAX_TEAM_SNAPSHOT_PNG_BYTES, @@ -505,10 +506,12 @@ pub async fn fetch_snapshot_bytes( // 4. Bytes must parse as the snapshot type selected by the filename. // Team parsing rejects retired flat JSON and persona-pack ZIP inputs - // before anything reaches the frontend. + // before anything reaches the frontend. Agent kinds accept both plain + // manifests and structurally valid locked (encrypted) card envelopes — + // transit validation never decrypts; unlock happens at import time. match kind { SnapshotFileKind::AgentJson | SnapshotFileKind::AgentPng => { - decode_snapshot_from_bytes(&bytes) + parse_snapshot_payload_from_bytes(&bytes) .map_err(|e| format!("invalid agent snapshot: {e}"))?; } SnapshotFileKind::TeamJson | SnapshotFileKind::TeamPng => { diff --git a/desktop/src-tauri/src/commands/mesh_llm.rs b/desktop/src-tauri/src/commands/mesh_llm.rs index 998bc6e7d2..528ca38767 100644 --- a/desktop/src-tauri/src/commands/mesh_llm.rs +++ b/desktop/src-tauri/src/commands/mesh_llm.rs @@ -364,8 +364,7 @@ pub(crate) async fn restore_mesh_sharing(app: &AppHandle, state: &AppState) -> C } // This is restoration of a previously inference-ready serving node. Keep // the enabled checkpoint armed while restoring so a transient startup - // failure does not silently turn Share Compute off. New starts remain - // disarmed in `mesh_start_node` until their first inference probe passes. + // failure does not silently turn Share Compute off. let request = mesh_llm::StartMeshNodeRequest { mode: mesh_llm::MeshNodeMode::Serve, model_id: Some(config.model_id.clone()), @@ -378,20 +377,26 @@ pub(crate) async fn restore_mesh_sharing(app: &AppHandle, state: &AppState) -> C let started = mesh_llm::DesktopMeshRuntime::start(request) .await .map_err(|error| format!("failed to restore Share Compute: {error:#}"))?; - if let Err(error) = wait_for_mesh_inference(&config.model_id).await { - let cleanup = started.stop().await; - if let Err(cleanup_error) = cleanup { - eprintln!( - "buzz-mesh: restored node failed inference readiness and cleanup was incomplete: {cleanup_error:#}" - ); - } - return Err(format!("failed to restore Share Compute: {error}")); - } + // Install the restored runtime immediately: it is tracked by AppState from + // here on, so it can never be orphaned. Restoring a previously + // inference-ready node still has to load ~tens of GB of weights and may + // download package layers after the ports bind, and the readiness probe + // itself serializes behind any first inference. None of that is a failed + // restore — stopping the node and reporting failure (the old behaviour) + // tore down a node that was simply still warming up. The checkpoint stays + // armed (`enabled`), so a genuinely broken restore is retried next launch + // rather than silently turning Share Compute off. *runtime = Some(started); config.enabled = true; config.start_on_next_launch = false; save_mesh_sharing_config(app, &config)?; drop(runtime); + if let Err(error) = wait_for_mesh_inference(&config.model_id).await { + eprintln!( + "buzz-mesh: restored node is not inference-ready yet ({error}); \ + leaving it to warm up without tearing it down" + ); + } mesh_llm::publish_current_status_once(app, "restore").await; Ok(()) } @@ -467,9 +472,11 @@ pub async fn mesh_start_node( } if let Some(config) = sharing_config.as_ref() { - // Do not arm launch restoration until the exact inference path used by - // agents succeeds. Mesh may bind its ports after primary weights load - // while package layers are still downloading. + // Persist a DISARMED checkpoint to cover the window of the potentially + // long `start()` below: if Buzz exits before the runtime is installed + // and tracked, the next launch stays stopped rather than trying to + // restore a node that never came up. The enabled config is armed right + // after install succeeds. save_mesh_sharing_config(&app, &pending_new_start_checkpoint(config))?; } @@ -496,25 +503,28 @@ pub async fn mesh_start_node( )); } }; - if let Some(config) = sharing_config.as_ref() { - if let Err(error) = wait_for_mesh_inference(&config.model_id).await { - let cleanup = started.stop().await; - if let Err(cleanup_error) = &cleanup { - eprintln!( - "buzz-mesh: started node failed inference readiness and cleanup was incomplete: {cleanup_error:#}" - ); - } - drop(runtime); - app.request_restart(); - return Err(format!( - "mesh node started but inference never became ready: {error}; Buzz is restarting to guarantee cleanup" - )); - } - } + // Install (track) the runtime BEFORE probing readiness so it can never be + // orphaned. A readiness timeout is not death: mesh binds its ports before + // weights finish loading / layers finish downloading, and serializes all + // ingress HTTP (this probe included) behind any in-flight turn — a cold + // start can take minutes. The old code stopped the node and restarted the + // app on that timeout, turning startup latency into a restart loop. *runtime = Some(started); drop(runtime); if let Some(config) = sharing_config.as_ref() { + // Installed + tracked == Share Compute is on, so persist the enabled + // config now (mirroring restore), not gated on the probe. Gating it + // meant a slow first start served fine but came back OFF next launch. + // Safe: neither the watchdog (evicts only a closed port) nor restore + // (leaves a warming node alone) can loop a slow-but-alive node, and an + // unstartable config fails earlier in `start()`. Probe is informational. save_mesh_sharing_config(&app, config)?; + if let Err(error) = wait_for_mesh_inference(&config.model_id).await { + eprintln!( + "buzz-mesh: node started but inference is not ready yet ({error}); \ + leaving it to warm up (Share Compute stays armed for next launch)" + ); + } } mesh_llm::publish_current_status_once(&app, "start").await; Ok(status) diff --git a/desktop/src-tauri/src/commands/personas/card.rs b/desktop/src-tauri/src/commands/personas/card.rs new file mode 100644 index 0000000000..29a5c35e6a --- /dev/null +++ b/desktop/src-tauri/src/commands/personas/card.rs @@ -0,0 +1,999 @@ +//! `mint_agent_card` / `save_agent_card` Tauri commands — Agent Trading Cards. +//! +//! Mints a collectible trading-card PNG for an agent via one OpenAI Responses +//! API call (designer model + native `image_generation` tool), then embeds the +//! agent's `buzz_agent_snapshot` manifest through the existing snapshot +//! encoder so the card IS an importable `.agent.png`. +//! +//! Boundary rules (agreed with Wren, buzz-agent-trading-cards thread): +//! - Snapshot construction/injection reuses `agent_snapshot.rs` — cards +//! inherit manifest-v1 behavior, exclusions, and size checks. No card-only +//! wire format exists. +//! - Memory inclusion is opt-in and shares the export flow's semantics: the +//! same three levels (`none`/`core`/`everything`), the same owner-gated +//! `get_agent_memory` fetch, and a memory source DERIVED from the resolved +//! instance (never caller-supplied), so cross-agent memory pairing is +//! structurally impossible. The default is `none`; the encoder still +//! rejects `none` + entries. +//! - The 10 MiB `.agent.png` ceiling is enforced on the FINAL bytes (after +//! resize + chunk injection) via `validate_snapshot_encode_size`. +//! - Round-trip verification decodes the final bytes and compares the logical +//! manifest before anything is returned to the frontend. +//! - The API key is resolved through the same env layering the agent runtime +//! uses (global config < persona < agent record) and never leaves Rust. +//! It is never logged. + +use base64::{engine::general_purpose::STANDARD, Engine as _}; +use serde::{Deserialize, Serialize}; +use tauri::{AppHandle, State}; + +use super::super::export_util::save_bytes_with_dialog; +use super::snapshot::{ + memory_entries_from_listing, parse_memory_level, resolve_from_lists, + validate_snapshot_encode_size, +}; +use crate::{ + app_state::AppState, + commands::engrams::get_agent_memory, + managed_agents::{ + agent_snapshot::{ + build_snapshot, decode_avatar_data_url, decode_snapshot_png, encode_snapshot_png, + extract_chunk_payload_png, MemoryLevel, + }, + agent_snapshot_envelope::{ + decrypt_envelope, encode_locked_snapshot_png, parse_chunk_payload, ChunkPayload, + }, + load_agent_definitions, load_global_agent_config, load_managed_agents, load_personas, + save_global_agent_config, validate_global_config, + }, +}; + +/// The Buzz card frame template — Tyler's gold-honeycomb base. Generation +/// input only: it never participates in the snapshot manifest, PNG chunk, +/// import decoder, or attachment validation. Embedded at compile time for +/// deterministic packaging (see `card_template_decodes` test). +const CARD_TEMPLATE_PNG: &[u8] = include_bytes!("../../../assets/card_template.png"); + +/// Designer model driving copy + art direction. +const DESIGNER_MODEL: &str = "gpt-5.6-sol"; +/// Image model invoked natively via the Responses `image_generation` tool. +const IMAGE_MODEL: &str = "gpt-image-2"; +/// Final card width in pixels (2:3 portrait → 1500x2250). +const CARD_WIDTH: u32 = 1500; +/// Longest edge for the real avatar inlined into an unlocked card's manifest. +/// Kind:0 pictures render small; 512px keeps the doubly-base64-encoded +/// manifest chunk modest next to the 1500-wide card body. +const MANIFEST_AVATAR_MAX_DIM: u32 = 512; +/// Upper bound for a fetched avatar (pre-resize input to the model). +const MAX_AVATAR_FETCH_BYTES: usize = 10 * 1024 * 1024; +/// One mint is a single long API call (~2–3 minutes observed). +const MINT_TIMEOUT_SECS: u64 = 600; + +/// Error prefix the frontend matches to route the user to provider settings +/// instead of showing a raw failure. +pub(crate) const NO_KEY_ERROR_PREFIX: &str = "NO_OPENAI_KEY:"; + +/// Wire shape returned by `mint_agent_card`. +#[derive(Debug, Serialize)] +#[serde(rename_all = "camelCase")] +pub struct MintedCard { + /// Final `.agent.png` bytes (chunk-injected, round-trip verified), + /// base64-encoded for the IPC boundary. + pub card_png_base64: String, + /// Suggested filename, e.g. `eva.agent.png`. + pub file_name: String, + /// Designer commentary emitted alongside the image (may be empty). + pub designer_notes: String, + /// True when the embedded snapshot is NIP-44-encrypted to the + /// (owner, agent) pair — only their nsecs can import this card. + pub locked: bool, + /// How much memory is embedded in the card's snapshot ("none"/"core"/ + /// "everything"). The viewer's import disclosure depends on this. + pub memory_level: MemoryLevel, +} + +// ── Card archive ────────────────────────────────────────────────────────────── + +/// Sidecar metadata for one archived card PNG. Stored as `.json` next +/// to `.agent.png` in the cards dir — two plain files per mint, no +/// shared index to corrupt. Listing scans sidecars; a card whose PNG is +/// missing is skipped rather than failing the whole list. +#[derive(Debug, Clone, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub struct ArchivedCardMeta { + /// Unique on-disk PNG file name within the cards dir. + pub stored_file_name: String, + /// Suggested save-as name, e.g. `eva.agent.png`. + pub file_name: String, + /// The id the card was minted for (instance pubkey or definition slug). + pub agent_id: String, + pub agent_name: String, + pub designer_notes: String, + pub locked: bool, + /// Memory embedded in this card's snapshot. Defaults to `None` when the + /// sidecar predates the field — every pre-field mint was minted with + /// `MemoryLevel::None` (it was structural), so the default is honest. + #[serde(default)] + pub memory_level: MemoryLevel, + /// ISO-8601 mint timestamp. + pub minted_at: String, + /// Small JPEG preview for gallery grids, base64. Populated by + /// `list_agent_cards` from the sidecar thumb file — never stored in the + /// JSON sidecar itself. + #[serde(default, skip_deserializing)] + pub thumb_jpeg_base64: Option, +} + +fn cards_dir(app: &AppHandle) -> Result { + let dir = crate::managed_agents::managed_agents_base_dir(app)?.join("cards"); + std::fs::create_dir_all(&dir).map_err(|e| format!("failed to create cards dir: {e}"))?; + Ok(dir) +} + +/// Persist a freshly minted card to the archive. Failures are surfaced to the +/// caller (which logs and continues) — an archive write must never fail a +/// mint the user already paid for. +fn archive_minted_card( + app: &AppHandle, + agent_id: &str, + agent_name: &str, + card: &MintedCard, + bytes: &[u8], +) -> Result { + let dir = cards_dir(app)?; + let stem = format!( + "{}-{}", + crate::util::slugify(agent_name, "agent", 50), + uuid::Uuid::new_v4() + ); + let stored_file_name = format!("{stem}.agent.png"); + let meta = ArchivedCardMeta { + stored_file_name: stored_file_name.clone(), + file_name: card.file_name.clone(), + agent_id: agent_id.to_string(), + agent_name: agent_name.to_string(), + designer_notes: card.designer_notes.clone(), + locked: card.locked, + memory_level: card.memory_level, + minted_at: crate::util::now_iso(), + thumb_jpeg_base64: None, + }; + // PNG first, sidecar second: a crash between the two leaves an orphaned + // PNG (invisible to the list), never a sidecar pointing at nothing. + std::fs::write(dir.join(&stored_file_name), bytes) + .map_err(|e| format!("failed to write archived card: {e}"))?; + let meta_json = serde_json::to_string_pretty(&meta) + .map_err(|e| format!("failed to serialize card metadata: {e}"))?; + std::fs::write(dir.join(format!("{stem}.json")), meta_json) + .map_err(|e| format!("failed to write card metadata: {e}"))?; + // Thumb last and best-effort: the gallery grid falls back to lazy + // full-card loading for a card whose thumb is missing. + if let Ok(thumb) = encode_card_thumb(bytes) { + let _ = std::fs::write(dir.join(format!("{stem}.thumb.jpg")), thumb); + } + Ok(meta) +} + +/// Downscale card PNG bytes to a small JPEG for gallery grids. The full card +/// is ~1500x2250 PNG (megabytes); shipping that per card over IPC just to +/// draw a grid tile is waste. +fn encode_card_thumb(bytes: &[u8]) -> Result, String> { + const THUMB_WIDTH: u32 = 300; + let img = image::load_from_memory(bytes).map_err(|e| format!("thumb decode: {e}"))?; + let scale = THUMB_WIDTH as f64 / img.width() as f64; + let thumb = img.resize( + THUMB_WIDTH, + (img.height() as f64 * scale).round().max(1.0) as u32, + image::imageops::FilterType::Triangle, + ); + let mut out = Vec::new(); + // JPEG has no alpha; cards are opaque, so flatten unconditionally. + let rgb = image::DynamicImage::ImageRgb8(thumb.to_rgb8()); + rgb.write_to( + &mut std::io::Cursor::new(&mut out), + image::ImageFormat::Jpeg, + ) + .map_err(|e| format!("thumb encode: {e}"))?; + Ok(out) +} + +/// Reject any archive file name that could escape the cards dir or name a +/// non-archive file. Archive names are generated by `archive_minted_card` +/// (slug + UUID), so a strict shape check loses nothing legitimate. +fn validate_archive_file_name(stored_file_name: &str) -> Result<(), String> { + let valid = stored_file_name.ends_with(".agent.png") + && !stored_file_name.contains(['/', '\\']) + && !stored_file_name.contains(".."); + if !valid { + return Err("Invalid archived card file name.".to_string()); + } + Ok(()) +} + +/// List all archived cards, newest first. +#[tauri::command] +pub fn list_agent_cards(app: AppHandle) -> Result, String> { + let dir = cards_dir(&app)?; + let entries = std::fs::read_dir(&dir).map_err(|e| format!("failed to read cards dir: {e}"))?; + let mut cards = Vec::new(); + for entry in entries.flatten() { + let path = entry.path(); + if path.extension().and_then(|e| e.to_str()) != Some("json") { + continue; + } + let Ok(content) = std::fs::read_to_string(&path) else { + continue; + }; + let Ok(meta) = serde_json::from_str::(&content) else { + // A malformed sidecar hides one card, never the archive. + eprintln!( + "buzz-desktop: card-archive: skipping malformed sidecar {}", + path.display() + ); + continue; + }; + let mut meta = meta; + if validate_archive_file_name(&meta.stored_file_name).is_ok() + && dir.join(&meta.stored_file_name).is_file() + { + // Attach the pre-rendered grid thumb when present (best-effort). + if let Some(stem) = path.file_stem().and_then(|s| s.to_str()) { + meta.thumb_jpeg_base64 = std::fs::read(dir.join(format!("{stem}.thumb.jpg"))) + .ok() + .map(|b| STANDARD.encode(&b)); + } + cards.push(meta); + } + } + // ISO-8601 sorts lexicographically; newest first. + cards.sort_by(|a, b| b.minted_at.cmp(&a.minted_at)); + Ok(cards) +} + +/// Load one archived card's PNG bytes as base64, keyed by its stored file +/// name (as returned by `list_agent_cards`). +#[tauri::command] +pub fn load_agent_card(stored_file_name: String, app: AppHandle) -> Result { + validate_archive_file_name(&stored_file_name)?; + let bytes = std::fs::read(cards_dir(&app)?.join(&stored_file_name)) + .map_err(|e| format!("failed to read archived card: {e}"))?; + Ok(STANDARD.encode(&bytes)) +} + +// ── Key resolution ──────────────────────────────────────────────────────────── + +/// Pure layering: global env < persona env < agent record env, then the +/// process environment as a development fallback. Returns the first +/// non-empty value for `key`. +pub(crate) fn resolve_env_from_layers( + key: &str, + global_env: &std::collections::BTreeMap, + persona_env: &std::collections::BTreeMap, + record_env: &std::collections::BTreeMap, + process_value: Option, +) -> Option { + for layer in [record_env, persona_env, global_env] { + if let Some(v) = layer.get(key) { + let v = v.trim(); + if !v.is_empty() { + return Some(v.to_string()); + } + } + } + process_value.filter(|k| !k.trim().is_empty()) +} + +/// Pure classification: same four env inputs as `resolve_env_from_layers`, +/// returns which layer supplies `OPENAI_API_KEY` (agent > persona > global > +/// process > none). +pub(crate) fn resolve_key_layer( + global_env: &std::collections::BTreeMap, + persona_env: &std::collections::BTreeMap, + record_env: &std::collections::BTreeMap, + process_value: Option, +) -> &'static str { + let key = "OPENAI_API_KEY"; + let nonempty = |m: &std::collections::BTreeMap| { + m.get(key).is_some_and(|v| !v.trim().is_empty()) + }; + if nonempty(record_env) { + return "agent"; + } + if nonempty(persona_env) { + return "persona"; + } + if nonempty(global_env) { + return "global"; + } + let proc = process_value.as_deref().unwrap_or(""); + if !proc.trim().is_empty() { + return "process"; + } + "none" +} + +/// The Responses endpoint to post mints to. `OPENAI_BASE_URL` (same env +/// layering as the key) overrides the default host, supporting endpoints and +/// proxies that speak the OpenAI Responses shape with Bearer auth. Azure +/// OpenAI is NOT covered by this override alone — it uses its own URL scheme +/// and `api-key` auth header, which would need a real driver. +pub(crate) fn responses_url(base_url: Option) -> String { + let base = base_url.unwrap_or_else(|| "https://api.openai.com/v1".to_string()); + format!("{}/responses", base.trim_end_matches('/')) +} + +// ── Prompt construction ─────────────────────────────────────────────────────── + +/// Build the designer instructions. Pure so tests can pin the contract: +/// style-match-the-avatar is DEFAULT behavior; owner directions (art AND +/// card text) take primacy over those style defaults, but never over the +/// fixed contract (frame identity, geometry, text fidelity). +pub(crate) fn build_card_instructions( + agent_name: &str, + persona_notes: &str, + style_notes: &str, +) -> String { + let owner_directions = if style_notes.trim().is_empty() { + String::new() + } else { + format!( + "\nOWNER'S DIRECTIONS — these override the default art-style and copy guidance \ + below wherever they conflict (they cannot change the frame, layout, or \ + text-fidelity requirements). The owner may direct the art, the card text \ + (type line, ability, flavor), or both:\n{style_notes}\n" + ) + }; + format!( + r#"You are designing one premium collectible trading card for the Buzz agent "{agent_name}". + +Input image 1 is the official Buzz card frame template (gold honeycomb border, dark interior, name banner top, hex badge top-right, text box lower third). Input image 2 is the agent's avatar — study its exact art style: medium, pixel grid if any, palette, shading, background motifs. + +Persona notes for the card copy: +{persona_notes} +{owner_directions} +First, write professional trading-card copy at Magic: The Gathering editorial quality: +- a type line (e.g. "Legendary Agent — Team Lead"), +- ONE keyworded ability: short bolded ability name + one sentence of crisp rules text written like real MTG rules (present tense, precise, no fluff), +- ONE italic flavor-text line, evocative and short, the kind that gets quoted. +Where the owner's directions specify card text, use their wording within the 220-character text-box limit below (edited only for spelling; if their text exceeds the limit, condense it minimally while keeping their words and intent); invent copy only for the parts they left open. +Keep total text-box copy under 220 characters so it renders cleanly. + +Then generate the finished card with the image tool, exactly 1024x1536 portrait: +- The frame must follow input image 1 faithfully: same gold honeycomb border, same layout, honey drip detail. +- Default art style: match input image 2's art style EXACTLY — same medium, same pixel density if pixel art, same palette, same background honeycomb-lattice sky. It must look like the same artist drew a larger scene: the character in a confident pose, conjuring glowing golden hexagons. The owner's directions above override any of this default styling where they conflict. +- Name banner: "{agent_name}" plus the type line beneath it in smaller type. +- Text box: the ability name in bold, rules text in regular, then the flavor line in italics, cleanly typeset like a real MTG card — professional kerning, no misspellings, hyphenate nothing. +- Top-right hex badge: one small emblem of your choice, no text. +Render all text with perfect fidelity."# + ) +} + +/// Encode raw image bytes as a `data:image/png;base64,` URL, downscaling to +/// `max_dim` on the longest edge so request payloads stay small. +fn image_data_url(bytes: &[u8], max_dim: u32) -> Result { + Ok(format!( + "data:image/png;base64,{}", + STANDARD.encode(png_bytes_resized(bytes, max_dim)?) + )) +} + +/// Re-encode an image as PNG, downscaling so neither side exceeds `max_dim`. +fn png_bytes_resized(bytes: &[u8], max_dim: u32) -> Result, String> { + let img = image::load_from_memory(bytes).map_err(|e| format!("Failed to decode image: {e}"))?; + let img = if img.width().max(img.height()) > max_dim { + img.resize(max_dim, max_dim, image::imageops::FilterType::Lanczos3) + } else { + img + }; + let mut png = Vec::new(); + img.write_to(&mut std::io::Cursor::new(&mut png), image::ImageFormat::Png) + .map_err(|e| format!("Failed to encode image: {e}"))?; + Ok(png) +} + +// ── Response parsing ────────────────────────────────────────────────────────── + +/// Extract the generated image (base64) and any designer text from a +/// Responses API payload. Pure for testability. +pub(crate) fn extract_card_output(resp: &serde_json::Value) -> Result<(String, String), String> { + let output = resp + .get("output") + .and_then(|o| o.as_array()) + .ok_or_else(|| "Responses payload has no output array".to_string())?; + + let mut image_b64 = None; + let mut notes = Vec::new(); + for item in output { + match item.get("type").and_then(|t| t.as_str()) { + Some("image_generation_call") => { + if let Some(result) = item.get("result").and_then(|r| r.as_str()) { + image_b64 = Some(result.to_string()); + } + } + Some("message") => { + if let Some(content) = item.get("content").and_then(|c| c.as_array()) { + for c in content { + if c.get("type").and_then(|t| t.as_str()) == Some("output_text") { + if let Some(text) = c.get("text").and_then(|t| t.as_str()) { + notes.push(text.to_string()); + } + } + } + } + } + _ => {} + } + } + + let image_b64 = image_b64.ok_or_else(|| { + let types: Vec<&str> = output + .iter() + .filter_map(|i| i.get("type").and_then(|t| t.as_str())) + .collect(); + format!("No image in Responses output (item types: {types:?})") + })?; + Ok((image_b64, notes.join("\n"))) +} + +// ── Commands ────────────────────────────────────────────────────────────────── + +/// Save an `OPENAI_API_KEY` into the global Agent Defaults env for card +/// minting — a narrow seam with deliberately different semantics from the +/// general `set_global_agent_config`: +/// +/// - **No agent restarts.** The general command stops/restarts every running +/// local agent whose effective env changes, because agent env is baked at +/// spawn time. The mint command re-reads the config from disk on every +/// mint, so minting needs no restart — and a card setup must never disrupt +/// running agents as a side effect. Agents pick the key up naturally on +/// their next (re)start. +/// - **Read-modify-write of the latest on-disk config.** The config is +/// re-read immediately before the single-key insert + write (under the +/// managed-agents store lock, which serializes it against the other card +/// and agent-store commands), so a settings save that landed after this +/// dialog opened is not clobbered with a stale dialog-open snapshot. +/// (The general settings editor performs its own whole-config write; as +/// today, the last writer wins between the two surfaces.) +/// +/// Standard global-config validation still applies (POSIX key shape, +/// reserved-key reject, size caps) — this is not a validation bypass. +#[tauri::command] +pub fn card_mint_save_openai_key( + key: String, + app: AppHandle, + state: State<'_, AppState>, +) -> Result<(), String> { + let key = key.trim().to_string(); + if key.is_empty() { + return Err("API key cannot be empty.".to_string()); + } + + let _store_guard = state + .managed_agents_store_lock + .lock() + .map_err(|e| e.to_string())?; + + let mut config = load_global_agent_config(&app)?; + config.env_vars.insert("OPENAI_API_KEY".to_string(), key); + validate_global_config(&config)?; + save_global_agent_config(&app, &config) +} + +/// Report which env layer resolves the OpenAI key for a card mint of agent +/// `id` — same layering as `mint_agent_card`. Delegates to `resolve_key_layer` +/// for the classification; see that helper for the return-value contract. +#[tauri::command] +pub fn card_mint_key_status( + id: String, + app: AppHandle, + state: State<'_, AppState>, +) -> Result { + let _store_guard = state + .managed_agents_store_lock + .lock() + .map_err(|e| e.to_string())?; + + let instances = load_managed_agents(&app)?; + let definitions = load_agent_definitions(&app)?; + let (record, _) = resolve_from_lists(&id, &instances, &definitions)?; + + let global = load_global_agent_config(&app).unwrap_or_default(); + let personas = load_personas(&app).unwrap_or_default(); + let persona_env = record + .persona_id + .as_deref() + .and_then(|pid| personas.iter().find(|p| p.id == pid)) + .map(|p| p.env_vars.clone()) + .unwrap_or_default(); + + Ok(resolve_key_layer( + &global.env_vars, + &persona_env, + &record.env_vars, + std::env::var("OPENAI_API_KEY").ok(), + ) + .to_string()) +} + +/// Mint a trading card for the agent identified by `id` (instance pubkey, +/// instance slug, or definition slug — same resolution as snapshot export). +/// +/// When `lock` is true the embedded manifest is NIP-44-encrypted to the +/// (owner, agent) pair per the locked-envelope contract — this requires a +/// linked agent instance (the second key endpoint); bare definitions cannot +/// be locked. +/// +/// When `memory_level` is `"core"` or `"everything"`, the owner's decrypted +/// memory for the agent is embedded in the manifest — same levels and fetch +/// as snapshot export. The memory source is always the resolved instance +/// itself (derived, never caller-supplied), so it requires a linked instance; +/// bare definitions can only mint `"none"` (the default). +/// +/// Returns the final, chunk-injected, round-trip-verified `.agent.png` bytes. +/// Reroll = call again; the command holds no session state. +#[tauri::command] +pub async fn mint_agent_card( + id: String, + style_notes: Option, + lock: Option, + memory_level: Option, + app: AppHandle, + state: State<'_, AppState>, +) -> Result { + let lock = lock.unwrap_or(false); + let memory_level = parse_memory_level(memory_level.as_deref().unwrap_or(""))?; + // ── Resolve the record + API key under lock ────────────────────────────── + let (mut record, is_definition, api_key, base_url) = { + let _store_guard = state + .managed_agents_store_lock + .lock() + .map_err(|e| e.to_string())?; + + let instances = load_managed_agents(&app)?; + let definitions = load_agent_definitions(&app)?; + let (record, is_definition) = + resolve_from_lists(&id, &instances, &definitions).map(|(r, d)| (r.clone(), d))?; + + let global = load_global_agent_config(&app).unwrap_or_default(); + let personas = load_personas(&app).unwrap_or_default(); + let persona_env = record + .persona_id + .as_deref() + .and_then(|pid| personas.iter().find(|p| p.id == pid)) + .map(|p| p.env_vars.clone()) + .unwrap_or_default(); + + let api_key = resolve_env_from_layers( + "OPENAI_API_KEY", + &global.env_vars, + &persona_env, + &record.env_vars, + std::env::var("OPENAI_API_KEY").ok(), + ) + .ok_or_else(|| { + format!( + "{NO_KEY_ERROR_PREFIX} No OPENAI_API_KEY found. Add one in the agent's \ + environment variables or global agent settings to mint cards." + ) + })?; + let base_url = resolve_env_from_layers( + "OPENAI_BASE_URL", + &global.env_vars, + &persona_env, + &record.env_vars, + std::env::var("OPENAI_BASE_URL").ok(), + ); + + (record, is_definition, api_key, base_url) + }; + + // ── Locking needs its two exact key endpoints up front, BEFORE the + // API spend: the owner identity secret and the agent instance pubkey. + let lock_keys = if lock { + if is_definition { + return Err( + "Locked cards need a linked agent instance — this persona has never been \ + started, so there is no agent key to lock to." + .to_string(), + ); + } + let owner_keys = state.signing_keys()?; + // Same canonical check the envelope decoder enforces (incl. curve + // validation) — a non-point record pubkey must fail BEFORE the API + // spend, not at post-mint encryption. + let agent_pubkey = crate::managed_agents::agent_snapshot_envelope::parse_canonical_pubkey( + "agentPubkey", + &record.pubkey, + ) + .map_err(|_| { + "Agent record has an invalid pubkey (not a canonical x-only key).".to_string() + })?; + if owner_keys.public_key() == agent_pubkey { + return Err("Cannot lock a card to itself: owner and agent keys match.".to_string()); + } + Some((owner_keys, agent_pubkey)) + } else { + None + }; + + // ── Memory needs a keyed instance, resolved up front BEFORE the API + // spend — the memory source is always the resolved instance itself + // (derived, never caller-supplied), so cross-agent pairing cannot be + // expressed. A failed fetch fails the mint here, not after payment. + let memory_entries = if memory_level == MemoryLevel::None { + Vec::new() + } else { + if is_definition { + return Err( + "Cards with memory need a linked agent instance — this persona has never \ + been started, so there is no agent memory to include." + .to_string(), + ); + } + let listing = get_agent_memory(record.pubkey.clone(), app.clone(), state.clone()).await?; + memory_entries_from_listing(listing, memory_level) + }; + + let display_name = record + .display_name + .clone() + .unwrap_or_else(|| record.name.clone()); + + // ── Prefer the agent's own kind:0 profile picture ──────────────────────── + // The record's `avatar_url` is a stale presentation snapshot: with + // agent-managed profiles the agent updates its own kind:0 `picture` and + // desktop reconciliation is disabled (`agent_settings.rs`), so the relay + // profile — not the local record — is the live source of truth for how the + // agent looks. Definitions have no keypair and thus no kind:0; they keep + // the record's avatar. A relay error fails the mint here, BEFORE the API + // spend (same fail-early rule as the key/memory guards above) — minting + // with the wrong face wastes the spend it was supposed to protect. + if !is_definition { + let relay_url = crate::relay::effective_agent_relay_url( + &record.relay_url, + &crate::relay::relay_ws_url_with_override(&state), + ); + let profile = crate::relay::query_agent_profile(&state, &relay_url, &record.pubkey) + .await + .map_err(|e| format!("Could not read the agent's profile for its avatar: {e}"))?; + record.avatar_url = preferred_avatar_url( + profile.and_then(|info| info.picture), + record.avatar_url.take(), + ); + } + + // ── Resolve avatar bytes (data URL, else fetch) ────────────────────────── + let avatar_bytes = match record.avatar_url.as_deref() { + Some(url) if url.starts_with("data:") => decode_avatar_data_url(url) + .ok_or_else(|| "Agent avatar data URL could not be decoded.".to_string())?, + Some(url) if url.starts_with("http://") || url.starts_with("https://") => { + // Relay-hosted avatars (kind:0 pictures under the relay's /media/) + // may require Blossom get-auth (`require_media_get_auth`). Mint the + // header ONLY for same-origin URLs so the token never leaves the + // relay (same contract as `media_download.rs`). + let relay_base = crate::relay::relay_api_base_url_with_override(&state); + let auth = is_same_origin(url, &relay_base) + .then(|| crate::commands::media::mint_media_get_auth(&state, &relay_base)) + .flatten(); + fetch_avatar(url, auth.as_deref()).await? + } + _ => { + return Err( + "Agent has no avatar image. Set an avatar before minting a card.".to_string(), + ) + } + }; + + // ── Build the manifest now (with any requested memory) so a broken agent + // fails before we spend minutes on the API call. ─────────────────────── + let manifest_avatar = manifest_avatar_bytes( + lock_keys.is_some(), + &avatar_bytes, + record.avatar_url.as_deref(), + )?; + let snapshot = build_snapshot( + &record, + memory_level, + memory_entries, + manifest_avatar.as_deref(), + ); + + // ── One Responses API call ─────────────────────────────────────────────── + // For locked mints, prove the manifest (including any embedded memory) + // fits the NIP-44 plaintext cap BEFORE spending minutes on the API call + // (same fail-early rule as the memory guard above). + if lock_keys.is_some() { + let json_len = + crate::managed_agents::agent_snapshot::encode_snapshot_json(&snapshot)?.len(); + if json_len > buzz_core_pkg::engram::NIP44_PLAINTEXT_MAX { + let hint = if memory_level == MemoryLevel::None { + "Reduce the avatar size or mint an unlocked card." + } else { + "Include less memory, reduce the avatar size, or mint an unlocked card." + }; + return Err(format!( + "Agent manifest is too large to lock ({json_len} bytes; the encrypted \ + format caps at {}). {hint}", + buzz_core_pkg::engram::NIP44_PLAINTEXT_MAX + )); + } + } + let instructions = build_card_instructions( + &display_name, + snapshot.definition.system_prompt.as_deref().unwrap_or(""), + style_notes.as_deref().unwrap_or(""), + ); + let body = serde_json::json!({ + "model": DESIGNER_MODEL, + "reasoning": {"effort": "high"}, + "instructions": "You are a senior TCG card designer and MTG rules editor.", + "input": [{ + "role": "user", + "content": [ + {"type": "input_text", "text": instructions}, + {"type": "input_image", "image_url": image_data_url(CARD_TEMPLATE_PNG, 1024)?}, + {"type": "input_image", "image_url": image_data_url(&avatar_bytes, 1024)?}, + ], + }], + "tools": [{ + "type": "image_generation", + "model": IMAGE_MODEL, + "quality": "high", + "size": "1024x1536", + "output_format": "png", + }], + "tool_choice": "required", + }); + + let client = reqwest::Client::builder() + .timeout(std::time::Duration::from_secs(MINT_TIMEOUT_SECS)) + .build() + .map_err(|e| format!("Failed to build HTTP client: {e}"))?; + let resp = client + .post(responses_url(base_url)) + .bearer_auth(&api_key) + .json(&body) + .send() + .await + .map_err(|e| format!("Card mint request failed: {e}"))?; + + let status = resp.status(); + let payload: serde_json::Value = resp + .json() + .await + .map_err(|e| format!("Card mint response was not JSON: {e}"))?; + if !status.is_success() { + // Never echo the request (it embeds nothing secret, but keep the + // failure surface small); the OpenAI error body is safe to surface. + let detail = payload + .get("error") + .and_then(|e| e.get("message")) + .and_then(|m| m.as_str()) + .unwrap_or("unknown error"); + return Err(format!("Card mint failed (HTTP {status}): {detail}")); + } + + let (image_b64, designer_notes) = extract_card_output(&payload)?; + let raw_card = STANDARD + .decode(image_b64.as_bytes()) + .map_err(|e| format!("Generated image was not valid base64: {e}"))?; + + // ── Resize to 1500-wide, inject chunk via the existing encoder ────────── + let card_img = image::load_from_memory(&raw_card) + .map_err(|e| format!("Generated image could not be decoded: {e}"))?; + let scale = CARD_WIDTH as f64 / card_img.width() as f64; + let card_img = card_img.resize( + CARD_WIDTH, + (card_img.height() as f64 * scale).round() as u32, + image::imageops::FilterType::Lanczos3, + ); + let mut card_png = Vec::new(); + card_img + .write_to( + &mut std::io::Cursor::new(&mut card_png), + image::ImageFormat::Png, + ) + .map_err(|e| format!("Failed to encode card PNG: {e}"))?; + + let final_bytes = match &lock_keys { + None => encode_snapshot_png(&snapshot, Some(&card_png)) + .map_err(|e| format!("Failed to embed agent snapshot in card: {e}"))?, + Some((owner_keys, agent_pubkey)) => { + encode_locked_snapshot_png(&snapshot, owner_keys, agent_pubkey, Some(&card_png)) + .map_err(|e| format!("Failed to embed locked agent snapshot in card: {e}"))? + } + }; + + // ── Verify: size ceiling + round-trip on the FINAL bytes ──────────────── + // Locked cards: extract the actual chunk, parse the envelope, decrypt + // with the owner key, then compare the logical manifest (ciphertext is + // nondeterministic — never compare bytes). + validate_snapshot_encode_size(final_bytes.len(), true)?; + let decoded = match &lock_keys { + None => decode_snapshot_png(&final_bytes) + .map_err(|e| format!("Card failed round-trip verification: {e}"))?, + Some((owner_keys, _)) => { + let payload = extract_chunk_payload_png(&final_bytes) + .map_err(|e| format!("Card failed round-trip verification: {e}"))?; + match parse_chunk_payload(&payload) + .map_err(|e| format!("Card failed round-trip verification: {e}"))? + { + ChunkPayload::Locked(envelope) => { + decrypt_envelope(&envelope, owner_keys.secret_key()) + .map_err(|e| format!("Card failed round-trip verification: {e}"))? + } + ChunkPayload::Plain(_) => { + return Err( + "Card round-trip verification failed: expected a locked envelope." + .to_string(), + ) + } + } + } + }; + if decoded != snapshot { + return Err("Card round-trip verification failed: manifest mismatch.".to_string()); + } + + let slug = crate::util::slugify(&display_name, "agent", 50); + let minted = MintedCard { + card_png_base64: STANDARD.encode(&final_bytes), + file_name: format!("{slug}.agent.png"), + designer_notes, + locked: lock_keys.is_some(), + memory_level, + }; + + // Archive best-effort: the mint is already paid for and verified, so a + // failed archive write logs and continues — it never fails the mint. + if let Err(e) = archive_minted_card(&app, &id, &display_name, &minted, &final_bytes) { + eprintln!("buzz-desktop: card-archive: failed to archive minted card: {e}"); + } + + Ok(minted) +} + +/// The avatar the mint should use: the agent's kind:0 `picture` when one is +/// published and non-blank, else the local record's `avatar_url`. +/// +/// Pure so the precedence is unit-testable without a relay: a blank or +/// whitespace-only `picture` must NOT shadow a real record avatar. +fn preferred_avatar_url( + kind0_picture: Option, + record_avatar_url: Option, +) -> Option { + kind0_picture + .filter(|p| !p.trim().is_empty()) + .or(record_avatar_url) +} + +/// The avatar bytes the card manifest should inline. +/// +/// Unlocked cards must carry the agent's REAL avatar inline: the PNG body is +/// the generated card artwork, and the importer only adopts the body as the +/// avatar when the manifest carries no inline bytes (`import.rs`) — without +/// these bytes an imported agent would wear the card as its face. Downscaled +/// to [`MANIFEST_AVATAR_MAX_DIM`] so the manifest tEXt chunk stays small. +/// +/// Locked cards keep the data-URL-only behavior: the whole manifest must fit +/// the NIP-44 plaintext cap (65 KB), which cannot carry inline pixels, and a +/// locked envelope never reaches the import body override anyway. +fn manifest_avatar_bytes( + locked: bool, + avatar_bytes: &[u8], + record_avatar_url: Option<&str>, +) -> Result>, String> { + if locked { + return Ok(decode_avatar_data_url(record_avatar_url.unwrap_or(""))); + } + png_bytes_resized(avatar_bytes, MANIFEST_AVATAR_MAX_DIM) + .map(Some) + .map_err(|e| format!("Failed to inline the agent avatar into the card manifest: {e}")) +} + +/// True when `url` shares an origin (scheme, host, port) with `relay_base`. +/// +/// Gate for attaching the minted media get-auth header — the token must never +/// be sent to a non-relay origin (same contract as `validate_download_url` in +/// `media_download.rs`, but non-fatal: a foreign origin just fetches +/// unauthenticated instead of failing the mint). +fn is_same_origin(url: &str, relay_base: &str) -> bool { + match (url::Url::parse(url), url::Url::parse(relay_base)) { + (Ok(u), Ok(b)) => u.origin() == b.origin(), + _ => false, + } +} + +/// Fetch an avatar over HTTP with a hard size cap. +/// +/// `auth` is an optional pre-minted Blossom get-auth header value, attached +/// verbatim — the caller is responsible for only supplying it for +/// relay-origin URLs. Redirects are not followed when auth is present +/// (redirect-hop guard, same rule as `media_download.rs`). +/// +/// The cap bounds network and memory, not just the final buffer: the +/// Content-Length header is checked before any body bytes are read, and the +/// body is streamed with a running count so a missing or dishonest header +/// still cannot exceed the cap (same contract as `media_download.rs`). +async fn fetch_avatar(url: &str, auth: Option<&str>) -> Result, String> { + use futures_util::StreamExt; + + let mut builder = reqwest::Client::builder().timeout(std::time::Duration::from_secs(30)); + if auth.is_some() { + // Never let a relay 3xx forward the auth header across origins. + builder = builder.redirect(reqwest::redirect::Policy::none()); + } + let client = builder + .build() + .map_err(|e| format!("Failed to build HTTP client: {e}"))?; + let mut req = client.get(url); + if let Some(auth) = auth { + req = req.header("authorization", auth); + } + let resp = req + .send() + .await + .map_err(|e| format!("Failed to fetch agent avatar: {e}"))?; + if !resp.status().is_success() { + return Err(format!("Avatar fetch failed: HTTP {}", resp.status())); + } + + if let Some(content_length) = resp.content_length() { + if content_length > MAX_AVATAR_FETCH_BYTES as u64 { + return Err("Agent avatar is too large to use as card input.".to_string()); + } + } + + let mut bytes = Vec::new(); + let mut stream = resp.bytes_stream(); + while let Some(chunk) = stream.next().await { + let chunk = chunk.map_err(|e| format!("Failed to read avatar bytes: {e}"))?; + append_within_avatar_cap(&mut bytes, &chunk)?; + } + Ok(bytes) +} + +/// Append a body chunk to the avatar buffer, rejecting before the append if +/// the total would cross `MAX_AVATAR_FETCH_BYTES`. Split out so the cap +/// boundary is unit-testable without an HTTP server. +fn append_within_avatar_cap(buf: &mut Vec, chunk: &[u8]) -> Result<(), String> { + if buf.len() + chunk.len() > MAX_AVATAR_FETCH_BYTES { + return Err("Agent avatar is too large to use as card input.".to_string()); + } + buf.extend_from_slice(chunk); + Ok(()) +} + +/// Save previously minted card bytes to disk via the OS save dialog. +/// +/// Re-validates the bytes (chunk parses as a plain manifest or a +/// structurally valid locked envelope, size within the import ceiling) so a +/// corrupted preview can never be written as a `.agent.png`. No decryption +/// happens here — the mint already round-trip-verified with the real key. +#[tauri::command] +pub async fn save_agent_card( + card_png_base64: String, + file_name: String, + app: AppHandle, +) -> Result { + let bytes = STANDARD + .decode(card_png_base64.as_bytes()) + .map_err(|e| format!("Card bytes were not valid base64: {e}"))?; + validate_snapshot_encode_size(bytes.len(), true)?; + let payload = extract_chunk_payload_png(&bytes) + .map_err(|e| format!("Refusing to save: card failed snapshot validation: {e}"))?; + parse_chunk_payload(&payload) + .map_err(|e| format!("Refusing to save: card failed snapshot validation: {e}"))?; + + let safe_name = if file_name.ends_with(".agent.png") && !file_name.contains(['/', '\\']) { + file_name + } else { + "card.agent.png".to_string() + }; + save_bytes_with_dialog(&app, &safe_name, "Agent card", &["png"], &bytes).await +} + +// ── Tests ───────────────────────────────────────────────────────────────────── + +#[cfg(test)] +mod tests; diff --git a/desktop/src-tauri/src/commands/personas/card/tests.rs b/desktop/src-tauri/src/commands/personas/card/tests.rs new file mode 100644 index 0000000000..407ab44974 --- /dev/null +++ b/desktop/src-tauri/src/commands/personas/card/tests.rs @@ -0,0 +1,399 @@ +//! Unit tests for `card.rs` — split into a child module file so the parent +//! stays under the 1000-line gate (same layout as `snapshot/tests.rs`). + +use super::*; +use std::collections::BTreeMap; + +#[test] +fn archive_file_name_validation_rejects_escapes() { + assert!(validate_archive_file_name("eva-1234.agent.png").is_ok()); + for bad in [ + "../escape.agent.png", + "sub/dir.agent.png", + "sub\\dir.agent.png", + "not-a-card.png", + "plain.json", + "", + ] { + assert!( + validate_archive_file_name(bad).is_err(), + "expected rejection: {bad:?}" + ); + } +} + +#[test] +fn card_template_decodes_with_expected_shape() { + // The embedded template is generation input only, but a corrupt or + // accidentally swapped asset should fail the build's test gate, not a + // user's first mint. + let img = image::load_from_memory(CARD_TEMPLATE_PNG).expect("template must decode"); + // 2:3-ish portrait frame. + assert!(img.height() > img.width(), "template must be portrait"); + assert!(img.width() >= 512, "template unexpectedly small"); +} + +#[test] +fn key_resolution_layering_record_wins() { + let mut global = BTreeMap::new(); + global.insert("OPENAI_API_KEY".to_string(), "global".to_string()); + let mut persona = BTreeMap::new(); + persona.insert("OPENAI_API_KEY".to_string(), "persona".to_string()); + let mut record = BTreeMap::new(); + record.insert("OPENAI_API_KEY".to_string(), "record".to_string()); + + assert_eq!( + resolve_env_from_layers("OPENAI_API_KEY", &global, &persona, &record, None).as_deref(), + Some("record") + ); + record.clear(); + assert_eq!( + resolve_env_from_layers("OPENAI_API_KEY", &global, &persona, &record, None).as_deref(), + Some("persona") + ); + persona.clear(); + assert_eq!( + resolve_env_from_layers("OPENAI_API_KEY", &global, &persona, &record, None).as_deref(), + Some("global") + ); + global.clear(); + assert_eq!( + resolve_env_from_layers( + "OPENAI_API_KEY", + &global, + &persona, + &record, + Some("process".to_string()) + ) + .as_deref(), + Some("process") + ); + assert!(resolve_env_from_layers("OPENAI_API_KEY", &global, &persona, &record, None).is_none()); +} + +/// Prove that `resolve_key_layer` classifies layers in the same precedence +/// order that `mint_agent_card`/`resolve_env_from_layers` uses, so the dialog +/// update path is only offered when writing global will actually win. +#[test] +fn key_status_layer_matches_mint_resolution_priority() { + let key = "OPENAI_API_KEY"; + let mut global = BTreeMap::new(); + let mut persona = BTreeMap::new(); + let mut record = BTreeMap::new(); + + // No key anywhere → "none" + assert_eq!(resolve_key_layer(&global, &persona, &record, None), "none"); + + // Only global → "global" (the only writable layer) + global.insert(key.to_string(), "sk-global".to_string()); + assert_eq!( + resolve_key_layer(&global, &persona, &record, None), + "global" + ); + // mint resolution also picks global when record and persona are empty + assert_eq!( + resolve_env_from_layers(key, &global, &persona, &record, None).as_deref(), + Some("sk-global") + ); + + // Persona overrides global → status must report "persona", NOT "global" + persona.insert(key.to_string(), "sk-persona".to_string()); + assert_eq!( + resolve_key_layer(&global, &persona, &record, None), + "persona" + ); + // mint would use the persona key + assert_eq!( + resolve_env_from_layers(key, &global, &persona, &record, None).as_deref(), + Some("sk-persona") + ); + // Writing to global would NOT change what mint resolves — status correctly + // returns "persona" so the dialog shows a read-only redirect instead. + let mut global_updated = global.clone(); + global_updated.insert(key.to_string(), "sk-new-global".to_string()); + assert_eq!( + resolve_env_from_layers(key, &global_updated, &persona, &record, None).as_deref(), + Some("sk-persona"), + "writing global must not change resolution when persona key exists" + ); + + // Agent record overrides both → status must report "agent" + record.insert(key.to_string(), "sk-agent".to_string()); + assert_eq!(resolve_key_layer(&global, &persona, &record, None), "agent"); + assert_eq!( + resolve_env_from_layers(key, &global, &persona, &record, None).as_deref(), + Some("sk-agent") + ); + + // Process env is last resort (only when all map layers are empty) + let empty = BTreeMap::new(); + assert_eq!( + resolve_key_layer(&empty, &empty, &empty, Some("sk-process".to_string())), + "process" + ); + + // Blank values are skipped — process wins over a whitespace global + let mut blank_global = BTreeMap::new(); + blank_global.insert(key.to_string(), " ".to_string()); + assert_eq!( + resolve_key_layer( + &blank_global, + &empty, + &empty, + Some("sk-process".to_string()) + ), + "process" + ); +} + +#[test] +fn key_resolution_skips_blank_values() { + let mut record = BTreeMap::new(); + record.insert("OPENAI_API_KEY".to_string(), " ".to_string()); + let mut persona = BTreeMap::new(); + persona.insert("OPENAI_API_KEY".to_string(), "persona".to_string()); + assert_eq!( + resolve_env_from_layers("OPENAI_API_KEY", &BTreeMap::new(), &persona, &record, None) + .as_deref(), + Some("persona") + ); +} + +#[test] +fn responses_url_default_and_override() { + assert_eq!(responses_url(None), "https://api.openai.com/v1/responses"); + // Trailing slashes must not produce a double-slash path. + assert_eq!( + responses_url(Some("https://proxy.example/v1/".to_string())), + "https://proxy.example/v1/responses" + ); + assert_eq!( + responses_url(Some("https://proxy.example/v1".to_string())), + "https://proxy.example/v1/responses" + ); +} + +#[test] +fn instructions_pin_style_match_default_and_owner_primacy() { + let base = build_card_instructions("Eva", "leads the team", ""); + assert!(base.contains("match input image 2's art style EXACTLY")); + assert!(base.contains("\"Eva\"")); + assert!(!base.contains("OWNER'S DIRECTIONS")); + + let directed = build_card_instructions("Eva", "leads the team", "make it stormy"); + // Owner directions take primacy over style defaults... + assert!(directed.contains("OWNER'S DIRECTIONS")); + assert!(directed.contains("make it stormy")); + assert!(directed.contains("override the default art-style and copy guidance")); + // ...but the fixed contract survives: frame, style anchor (as an + // overridable default), and text-fidelity requirements stay present. + assert!(directed.contains("match input image 2's art style EXACTLY")); + assert!(directed.contains("cannot change the frame, layout, or")); + assert!(directed.contains("Render all text with perfect fidelity")); + // Card-text direction is an explicitly named capability, and the + // owner-wording rule acknowledges the fixed 220-char text-box limit + // (no mutually impossible "verbatim" vs "under 220 chars" pair). + assert!(directed.contains("card text")); + assert!(directed.contains("use their wording within the 220-character text-box limit")); +} + +#[test] +fn extract_card_output_happy_path_and_missing_image() { + let ok = serde_json::json!({ + "output": [ + {"type": "reasoning"}, + {"type": "image_generation_call", "result": "aW1n"}, + {"type": "message", "content": [ + {"type": "output_text", "text": "notes here"} + ]} + ] + }); + let (img, notes) = extract_card_output(&ok).unwrap(); + assert_eq!(img, "aW1n"); + assert_eq!(notes, "notes here"); + + let missing = serde_json::json!({"output": [{"type": "message", "content": []}]}); + let err = extract_card_output(&missing).unwrap_err(); + assert!(err.contains("No image"), "{err}"); + + let no_output = serde_json::json!({}); + assert!(extract_card_output(&no_output).is_err()); +} + +#[test] +fn kind0_picture_wins_over_record_avatar_unless_blank() { + let some = |s: &str| Some(s.to_string()); + // Published picture wins. + assert_eq!( + preferred_avatar_url(some("https://relay/media/k0.png"), some("data:image/png;x")), + some("https://relay/media/k0.png") + ); + // No profile / no picture: record avatar survives. + assert_eq!( + preferred_avatar_url(None, some("data:image/png;x")), + some("data:image/png;x") + ); + // Blank or whitespace picture must not shadow a real avatar. + assert_eq!( + preferred_avatar_url(some(""), some("data:image/png;x")), + some("data:image/png;x") + ); + assert_eq!( + preferred_avatar_url(some(" "), some("data:image/png;x")), + some("data:image/png;x") + ); + // Nothing anywhere: None (caller surfaces the "no avatar" error). + assert_eq!(preferred_avatar_url(None, None), None); +} + +#[test] +fn unlocked_manifest_inlines_real_avatar_bytes_downscaled() { + // 700px source (over MANIFEST_AVATAR_MAX_DIM) in a solid color. + let avatar = image::DynamicImage::ImageRgba8(image::RgbaImage::from_pixel( + 700, + 700, + image::Rgba([9, 120, 33, 255]), + )); + let mut avatar_png = std::io::Cursor::new(Vec::new()); + avatar + .write_to(&mut avatar_png, image::ImageFormat::Png) + .unwrap(); + + let inlined = manifest_avatar_bytes(false, avatar_png.get_ref(), None) + .unwrap() + .expect("unlocked mints must inline the real avatar"); + let img = image::load_from_memory(&inlined).unwrap(); + assert_eq!( + (img.width(), img.height()), + (MANIFEST_AVATAR_MAX_DIM, MANIFEST_AVATAR_MAX_DIM) + ); + assert_eq!(img.to_rgba8().get_pixel(0, 0).0, [9, 120, 33, 255]); + + // Undecodable avatar bytes fail the mint (pre-spend), never silently + // produce a card whose import would wear the artwork as a face. + assert!(manifest_avatar_bytes(false, b"not a png", None).is_err()); +} + +#[test] +fn locked_manifest_keeps_data_url_only_avatar() { + // Locked mints must not inline fetched bytes (NIP-44 cap): only a record + // data URL carries over, exactly as before. + let unused = [0u8; 4]; + assert_eq!( + manifest_avatar_bytes(true, &unused, Some("data:image/png;base64,aGk=")) + .unwrap() + .as_deref(), + Some(b"hi".as_slice()) + ); + assert_eq!( + manifest_avatar_bytes(true, &unused, Some("https://relay/media/a.png")).unwrap(), + None + ); + assert_eq!(manifest_avatar_bytes(true, &unused, None).unwrap(), None); +} + +#[test] +fn media_get_auth_gate_is_same_origin_only() { + // The minted Blossom get-auth header may only travel to the relay's own + // origin — scheme, host, and port all count (same contract as + // `validate_download_url` in `media_download.rs`). + let relay = "https://relay.example.com"; + assert!(is_same_origin( + "https://relay.example.com/media/abc.png", + relay + )); + // Different host, scheme, or port: no auth. + assert!(!is_same_origin( + "https://evil.example.com/media/abc.png", + relay + )); + assert!(!is_same_origin( + "http://relay.example.com/media/abc.png", + relay + )); + assert!(!is_same_origin( + "https://relay.example.com:8443/media/abc.png", + relay + )); + // Unparseable inputs fail closed. + assert!(!is_same_origin("not a url", relay)); + assert!(!is_same_origin( + "https://relay.example.com/x", + "also not a url" + )); + // Explicit port on both sides matches. + assert!(is_same_origin( + "http://localhost:3100/media/abc.png", + "http://localhost:3100" + )); +} + +#[test] +fn avatar_cap_rejects_before_appending_crossing_chunk() { + // The streaming accumulator must reject a chunk that would cross the + // cap BEFORE buffering it — this is what bounds memory when + // Content-Length is missing or dishonest. + let mut buf = vec![0u8; MAX_AVATAR_FETCH_BYTES - 1]; + assert!(append_within_avatar_cap(&mut buf, &[0u8]).is_ok()); + assert_eq!(buf.len(), MAX_AVATAR_FETCH_BYTES); + // Exactly at the cap: one more byte must fail and not grow the buffer. + assert!(append_within_avatar_cap(&mut buf, &[0u8]).is_err()); + assert_eq!(buf.len(), MAX_AVATAR_FETCH_BYTES); + + // A single oversized chunk is rejected outright. + let mut fresh = Vec::new(); + let oversized = vec![0u8; MAX_AVATAR_FETCH_BYTES + 1]; + assert!(append_within_avatar_cap(&mut fresh, &oversized).is_err()); + assert!(fresh.is_empty()); +} + +#[test] +fn save_rejects_plain_png_without_snapshot_chunk() { + // A plain PNG (no buzz_agent_snapshot chunk) must not be saveable as + // a card. Exercise the same validation the command runs. + let img = image::DynamicImage::new_rgba8(4, 4); + let mut png = Vec::new(); + img.write_to(&mut std::io::Cursor::new(&mut png), image::ImageFormat::Png) + .unwrap(); + assert!(decode_snapshot_png(&png).is_err()); +} + +#[test] +fn archived_sidecar_without_memory_level_defaults_to_none() { + // Every mint before the memory option existed embedded MemoryLevel::None + // structurally, so old sidecars (no memoryLevel field) must deserialize + // to None — the gallery's disclosure depends on this being honest. + let legacy = r#"{ + "storedFileName": "eva-1234.agent.png", + "fileName": "eva.agent.png", + "agentId": "abc", + "agentName": "Eva", + "designerNotes": "", + "locked": false, + "mintedAt": "2026-07-28T00:00:00Z" + }"#; + let meta: ArchivedCardMeta = serde_json::from_str(legacy).unwrap(); + assert_eq!(meta.memory_level, MemoryLevel::None); + + let with_level = legacy.replace( + "\"locked\": false,", + "\"locked\": false, \"memoryLevel\": \"everything\",", + ); + let meta: ArchivedCardMeta = serde_json::from_str(&with_level).unwrap(); + assert_eq!(meta.memory_level, MemoryLevel::Everything); +} + +#[test] +fn minted_card_serializes_memory_level_snake_case_value() { + // The TS layer narrows on the exact wire strings "none"/"core"/ + // "everything" — pin the serde representation the frontend will see. + let minted = MintedCard { + card_png_base64: String::new(), + file_name: "eva.agent.png".to_string(), + designer_notes: String::new(), + locked: false, + memory_level: MemoryLevel::Core, + }; + let json = serde_json::to_value(&minted).unwrap(); + assert_eq!(json["memoryLevel"], "core"); +} diff --git a/desktop/src-tauri/src/commands/personas/mod.rs b/desktop/src-tauri/src/commands/personas/mod.rs index 66f7296a25..0cd7ad0324 100644 --- a/desktop/src-tauri/src/commands/personas/mod.rs +++ b/desktop/src-tauri/src/commands/personas/mod.rs @@ -306,11 +306,14 @@ pub async fn set_persona_active( } pub(crate) const PNG_MAGIC: [u8; 4] = [0x89, 0x50, 0x4E, 0x47]; +mod card; mod snapshot; -pub use snapshot::encode_agent_snapshot_for_send; -pub use snapshot::export_agent_snapshot; +pub use card::*; +#[cfg(test)] +pub(crate) use snapshot::import::decode_snapshot_from_bytes; pub(crate) use snapshot::import::{ - decode_snapshot_from_bytes, resolve_snapshot_import_behavior, MAX_SNAPSHOT_JSON_BYTES, + parse_snapshot_payload_from_bytes, resolve_snapshot_import_behavior, MAX_SNAPSHOT_JSON_BYTES, MAX_SNAPSHOT_PNG_BYTES, }; pub use snapshot::{confirm_agent_snapshot_import, preview_agent_snapshot_import}; +pub use snapshot::{encode_agent_snapshot_for_send, export_agent_snapshot}; diff --git a/desktop/src-tauri/src/commands/personas/snapshot.rs b/desktop/src-tauri/src/commands/personas/snapshot.rs index 583296dac0..e7bd1597e6 100644 --- a/desktop/src-tauri/src/commands/personas/snapshot.rs +++ b/desktop/src-tauri/src/commands/personas/snapshot.rs @@ -142,7 +142,7 @@ pub(crate) fn validate_snapshot_encode_size(bytes_len: usize, is_png: bool) -> R } /// Parse a `memory_level` string to `MemoryLevel`. -fn parse_memory_level(s: &str) -> Result { +pub(crate) fn parse_memory_level(s: &str) -> Result { match s { "none" | "" => Ok(MemoryLevel::None), "core" => Ok(MemoryLevel::Core), @@ -153,6 +153,32 @@ fn parse_memory_level(s: &str) -> Result { } } +/// Flatten an owner-decrypted memory listing into manifest entries for +/// `memory_level`: `Core` takes the core entry only; `Everything` appends all +/// `mem/*` entries after it. Pure so both the export and card-mint paths share +/// (and tests can pin) the level → entries selection. +pub(crate) fn memory_entries_from_listing( + listing: crate::commands::engrams::AgentMemoryListing, + memory_level: MemoryLevel, +) -> Vec { + let mut entries = Vec::new(); + if let Some(core) = listing.core { + entries.push(AgentSnapshotMemoryEntry { + slug: core.slug, + body: core.body, + }); + } + if memory_level == MemoryLevel::Everything { + for mem in listing.memories { + entries.push(AgentSnapshotMemoryEntry { + slug: mem.slug, + body: mem.body, + }); + } + } + entries +} + /// Parse a `format` string to a PNG flag. fn parse_format_is_png(s: &str) -> Result { match s { @@ -267,22 +293,7 @@ pub(crate) async fn materialize_snapshot_bytes( // ── Fetch memory ───────────────────────────────────────────────────────── let memory_entries: Vec = if let Some(pubkey) = memory_pubkey { let listing = get_agent_memory(pubkey, app.clone(), state).await?; - let mut entries = Vec::new(); - if let Some(core) = listing.core { - entries.push(AgentSnapshotMemoryEntry { - slug: core.slug, - body: core.body, - }); - } - if memory_level == MemoryLevel::Everything { - for mem in listing.memories { - entries.push(AgentSnapshotMemoryEntry { - slug: mem.slug, - body: mem.body, - }); - } - } - entries + memory_entries_from_listing(listing, memory_level) } else { Vec::new() }; diff --git a/desktop/src-tauri/src/commands/personas/snapshot/fidelity_tests.rs b/desktop/src-tauri/src/commands/personas/snapshot/fidelity_tests.rs index 00a1457393..b769d74d7b 100644 --- a/desktop/src-tauri/src/commands/personas/snapshot/fidelity_tests.rs +++ b/desktop/src-tauri/src/commands/personas/snapshot/fidelity_tests.rs @@ -208,3 +208,64 @@ fn import_png_placeholder_keeps_manifest_avatar_fallback() { assert!(decoded.profile.avatar_data_url.is_none()); assert_eq!(decoded.profile.avatar_url, snapshot.profile.avatar_url); } + +/// An unlocked trading card imports the agent's REAL avatar, never the card. +/// +/// Mint-shaped input: the PNG body is the generated card artwork, while the +/// manifest inlines the source avatar (`manifest_avatar_bytes` in `card.rs`). +/// The #3578 body-wins override must not fire when the manifest already +/// carries inline avatar bytes — otherwise the imported agent publishes the +/// 1500-wide card as its kind:0 picture. +#[test] +fn import_unlocked_card_uses_manifest_avatar_not_card_artwork() { + use crate::managed_agents::agent_snapshot::{decode_avatar_data_url, encode_snapshot_png}; + use base64::{engine::general_purpose::STANDARD, Engine as _}; + + // The real avatar: 4×3 solid blue, inlined in the manifest at mint time. + let real_avatar = image::DynamicImage::ImageRgba8(image::RgbaImage::from_pixel( + 4, + 3, + image::Rgba([23, 91, 177, 255]), + )); + let mut real_avatar_png = std::io::Cursor::new(Vec::new()); + real_avatar + .write_to(&mut real_avatar_png, image::ImageFormat::Png) + .unwrap(); + + let mut snapshot = make_snapshot(MemoryLevel::None, vec![]); + snapshot.profile.avatar_data_url = Some(format!( + "data:image/png;base64,{}", + STANDARD.encode(real_avatar_png.get_ref()) + )); + snapshot.profile.avatar_url = Some("https://relay.example/media/live-kind0.png".to_string()); + + // The card artwork: a distinct 1500×2250 solid red "trading card" as the + // PNG body — the exact dimensions the minter encodes for unlocked cards. + // Size matters: 2250px exceeds `snapshot_avatar`'s 2048px decode limit, + // so reaching the body override here wouldn't just import the wrong + // face — it would fail the import outright. + let card_art = image::DynamicImage::ImageRgba8(image::RgbaImage::from_pixel( + 1500, + 2250, + image::Rgba([200, 16, 16, 255]), + )); + let mut card_png = std::io::Cursor::new(Vec::new()); + card_art + .write_to(&mut card_png, image::ImageFormat::Png) + .unwrap(); + let file_bytes = encode_snapshot_png(&snapshot, Some(card_png.get_ref())).unwrap(); + + // Production import decode: the effective avatar must be the real one. + let decoded = decode_snapshot_from_bytes(&file_bytes).unwrap(); + let avatar_bytes = + decode_avatar_data_url(decoded.profile.avatar_data_url.as_deref().unwrap()).unwrap(); + let imported = image::load_from_memory(&avatar_bytes).unwrap(); + assert_eq!( + (imported.width(), imported.height()), + (4, 3), + "imported avatar must be the source avatar, not the card artwork" + ); + assert_eq!(imported.to_rgba8().get_pixel(0, 0).0, [23, 91, 177, 255]); + // The live kind:0 URL fallback survives untouched. + assert_eq!(decoded.profile.avatar_url, snapshot.profile.avatar_url); +} diff --git a/desktop/src-tauri/src/commands/personas/snapshot/import.rs b/desktop/src-tauri/src/commands/personas/snapshot/import.rs index eccf8ee601..d7f0323304 100644 --- a/desktop/src-tauri/src/commands/personas/snapshot/import.rs +++ b/desktop/src-tauri/src/commands/personas/snapshot/import.rs @@ -13,7 +13,11 @@ use tauri::{AppHandle, Emitter, State}; use crate::{ app_state::AppState, managed_agents::{ - agent_snapshot::{decode_snapshot_json, decode_snapshot_png, AgentSnapshot, MemoryLevel}, + agent_snapshot::{extract_chunk_payload_png, AgentSnapshot, MemoryLevel}, + agent_snapshot_envelope::{ + decrypt_envelope, parse_chunk_payload, resolve_unlock_secret, ChunkPayload, + LOCKED_CARD_REFUSAL, + }, load_managed_agents, load_personas, save_managed_agents, save_personas, AgentDefinition, ManagedAgentRecord, RespondTo, }, @@ -72,6 +76,16 @@ pub struct AgentSnapshotImportPreview { pub has_source_allowlist: bool, /// Number of source allowlist entries. pub source_allowlist_count: usize, + /// Full source allowlist entries, surfaced before import so hidden access + /// configuration is never reduced to a count. + pub source_allowlist: Vec, + /// Pretty-printed, validated manifest exactly as decoded from the file. + /// The UI makes this available before confirmation for full payload review. + pub manifest_json: String, + /// True when the snapshot came from a locked (encrypted) card that this + /// machine successfully unlocked. Cards that cannot be unlocked never + /// reach a preview — they fail closed with the locked-card refusal. + pub locked: bool, } /// The confirmation request sent from the UI after the user reviews the preview. @@ -210,50 +224,112 @@ const PNG_MAGIC: [u8; 4] = [0x89, 0x50, 0x4e, 0x47]; /// /// **Size cap:** PNG inputs over 10 MiB and JSON inputs over 5 MiB are rejected /// before allocation to avoid avoidable large-input work. -pub(crate) fn decode_snapshot_from_bytes( - file_bytes: &[u8], -) -> Result { - if file_bytes.len() >= 4 && file_bytes[..4] == PNG_MAGIC { +/// +/// **Locked cards:** a structurally valid locked envelope parses successfully +/// as `ChunkPayload::Locked` — no decryption happens here. Callers that can +/// unlock go through [`decode_snapshot_for_import`]; callers that only need +/// transit validation (e.g. `fetch_snapshot_bytes`) accept `Locked` as-is. +pub(crate) fn parse_snapshot_payload_from_bytes(file_bytes: &[u8]) -> Result { + let payload: ChunkPayload = if file_bytes.len() >= 4 && file_bytes[..4] == PNG_MAGIC { if file_bytes.len() > MAX_SNAPSHOT_PNG_BYTES { return Err(format!( "Snapshot file is too large ({} MiB). PNG snapshots must be under 10 MiB.", file_bytes.len() / (1024 * 1024) )); } - let mut snapshot = decode_snapshot_png(file_bytes)?; + let chunk_json = extract_chunk_payload_png(file_bytes)?; + let mut payload = parse_chunk_payload(&chunk_json)?; // The PNG image body is the portable avatar. It deliberately wins over - // manifest avatar fields, whose URL may only be reachable by the - // sender. A 1×1 export placeholder leaves the manifest fallback intact. - if let Some(avatar_data_url) = - crate::managed_agents::snapshot_avatar::snapshot_png_avatar_data_url(file_bytes)? - { - snapshot.profile.avatar_data_url = Some(avatar_data_url); + // a manifest avatar *URL*, which may only be reachable by the sender. + // A 1×1 export placeholder leaves the manifest fallback intact. + // Inline manifest avatar *bytes* are authoritative and never + // overridden: trading cards supply the generated card artwork as the + // PNG body and carry the agent's real avatar inline — adopting the + // body there would import the card as the agent's face. + // Locked envelopes stay opaque here — there is no manifest to override + // until the unlock path decrypts one. + if let ChunkPayload::Plain(snapshot) = &mut payload { + if snapshot.profile.avatar_data_url.is_none() { + if let Some(avatar_data_url) = + crate::managed_agents::snapshot_avatar::snapshot_png_avatar_data_url( + file_bytes, + )? + { + snapshot.profile.avatar_data_url = Some(avatar_data_url); + } + } } - if snapshot.memory.level == MemoryLevel::None && !snapshot.memory.entries.is_empty() { - return Err( - "Snapshot is malformed: memory.level is 'none' but entries are present." - .to_string(), - ); + payload + } else { + // JSON path — apply size cap before serde allocation. + if file_bytes.len() > MAX_SNAPSHOT_JSON_BYTES { + return Err(format!( + "Snapshot file is too large ({} MiB). JSON snapshots must be under 5 MiB.", + file_bytes.len() / (1024 * 1024) + )); } - return Ok(snapshot); - } - // JSON path — apply size cap before serde allocation. - if file_bytes.len() > MAX_SNAPSHOT_JSON_BYTES { - return Err(format!( - "Snapshot file is too large ({} MiB). JSON snapshots must be under 5 MiB.", - file_bytes.len() / (1024 * 1024) - )); - } - let snapshot = decode_snapshot_json(file_bytes)?; + parse_chunk_payload(file_bytes)? + }; // Consistency check: none + non-empty entries is always malformed, - // regardless of format. Mirrors the PNG path above so the rule is - // enforced at decode time for both formats. - if !snapshot.memory.entries.is_empty() && snapshot.memory.level == MemoryLevel::None { + // regardless of enclosing format. Enforced at decode time for plain + // payloads here, and after decryption for locked ones (see + // `enforce_memory_consistency` callers). + if let ChunkPayload::Plain(snapshot) = &payload { + enforce_memory_consistency(snapshot)?; + } + Ok(payload) +} + +/// The shared malformed-memory guard: `memory.level == none` with non-empty +/// entries is always rejected before any write. +fn enforce_memory_consistency( + snapshot: &crate::managed_agents::agent_snapshot::AgentSnapshot, +) -> Result<(), String> { + if snapshot.memory.level == MemoryLevel::None && !snapshot.memory.entries.is_empty() { return Err( "Snapshot is malformed: memory.level is 'none' but entries are present.".to_string(), ); } - Ok(snapshot) + Ok(()) +} + +/// Decode a plain snapshot from raw bytes, refusing locked cards. +/// +/// Test-only convenience: production call sites either unlock through +/// [`decode_snapshot_for_import`] or validate structurally through +/// [`parse_snapshot_payload_from_bytes`]. +#[cfg(test)] +pub(crate) fn decode_snapshot_from_bytes( + file_bytes: &[u8], +) -> Result { + match parse_snapshot_payload_from_bytes(file_bytes)? { + ChunkPayload::Plain(snapshot) => Ok(*snapshot), + ChunkPayload::Locked(_) => Err(LOCKED_CARD_REFUSAL.to_string()), + } +} + +/// Decode a snapshot for import, unlocking locked cards when — and only +/// when — this machine holds one of the envelope's two exact key endpoints +/// (the owner identity or the named local agent record). +/// +/// Returns the decoded manifest and whether it came from a locked envelope. +/// When neither endpoint exists, fails closed with the locked-card refusal — +/// never partial plaintext, never crypto details. +pub(crate) fn decode_snapshot_for_import( + file_bytes: &[u8], + owner_keys: Option<&nostr::Keys>, + records: &[ManagedAgentRecord], +) -> Result<(crate::managed_agents::agent_snapshot::AgentSnapshot, bool), String> { + match parse_snapshot_payload_from_bytes(file_bytes)? { + ChunkPayload::Plain(snapshot) => Ok((*snapshot, false)), + ChunkPayload::Locked(envelope) => { + let secret = resolve_unlock_secret(&envelope, owner_keys, records) + .ok_or_else(|| LOCKED_CARD_REFUSAL.to_string())?; + let snapshot = decrypt_envelope(&envelope, &secret)?; + enforce_memory_consistency(&snapshot)?; + Ok((snapshot, true)) + } + } } async fn materialize_import_avatar( @@ -283,19 +359,38 @@ where /// `.agent.png` file. The format is sniffed from the content, not the /// extension, so an incorrectly-named file is handled correctly. /// +/// Locked cards are unlocked here when this machine holds one of the +/// envelope's two exact key endpoints; a card that cannot be unlocked fails +/// with the locked-card refusal (shown directly to the user), never a +/// partial preview. Identity-recovery mode is tolerated: owner keys are +/// simply unavailable, so only the agent-record endpoint can unlock. +/// /// Returns an `AgentSnapshotImportPreview` or a descriptive error. Errors -/// represent irrecoverable failures (corrupt / unsupported file) and are -/// shown directly to the user. +/// represent irrecoverable failures (corrupt / unsupported / locked-to- +/// someone-else file) and are shown directly to the user. #[tauri::command] pub async fn preview_agent_snapshot_import( file_bytes: Vec, file_name: String, + app: AppHandle, + state: State<'_, AppState>, ) -> Result { + // Key material + records are gathered up front (cheap, lock-scoped) so + // the blocking decode below owns plain data. + let owner_keys = state.signing_keys().ok(); + let records = { + let _store_guard = state + .managed_agents_store_lock + .lock() + .map_err(|e| e.to_string())?; + load_managed_agents(&app)? + }; tokio::task::spawn_blocking(move || { reject_legacy_persona_filename(&file_name)?; - let snapshot = decode_snapshot_from_bytes(&file_bytes)?; + let (snapshot, locked) = + decode_snapshot_for_import(&file_bytes, owner_keys.as_ref(), &records)?; - Ok(build_agent_snapshot_import_preview(&snapshot)) + build_agent_snapshot_import_preview(&snapshot, locked) }) .await .map_err(|e| format!("spawn_blocking failed: {e}"))? @@ -303,7 +398,8 @@ pub async fn preview_agent_snapshot_import( pub(crate) fn build_agent_snapshot_import_preview( snapshot: &AgentSnapshot, -) -> AgentSnapshotImportPreview { + locked: bool, +) -> Result { let memory_level = match snapshot.memory.level { MemoryLevel::None => "none", MemoryLevel::Core => "core", @@ -311,7 +407,11 @@ pub(crate) fn build_agent_snapshot_import_preview( } .to_string(); - AgentSnapshotImportPreview { + let manifest_json = serde_json::to_string_pretty(snapshot) + .map_err(|e| format!("failed to render snapshot manifest: {e}"))?; + let source_allowlist = snapshot.definition.respond_to_allowlist.clone(); + + Ok(AgentSnapshotImportPreview { display_name: snapshot.profile.display_name.clone(), is_builtin: snapshot.definition.source_is_builtin, model: snapshot.definition.model.clone(), @@ -325,9 +425,12 @@ pub(crate) fn build_agent_snapshot_import_preview( .or_else(|| snapshot.profile.avatar_url.clone()), memory_level, memory_entry_count: snapshot.memory.entries.len(), - source_allowlist_count: snapshot.definition.respond_to_allowlist.len(), - has_source_allowlist: !snapshot.definition.respond_to_allowlist.is_empty(), - } + source_allowlist_count: source_allowlist.len(), + has_source_allowlist: !source_allowlist.is_empty(), + source_allowlist, + manifest_json, + locked, + }) } // ── `confirm_agent_snapshot_import` ────────────────────────────────────────── @@ -355,8 +458,20 @@ pub async fn confirm_agent_snapshot_import( app: AppHandle, state: State<'_, AppState>, ) -> Result { - // ── Phase 1: validate (no I/O) ─────────────────────────────────────────── - let snapshot = decode_snapshot_from_bytes(&input.file_bytes)?; + // ── Phase 1: validate (no writes) ──────────────────────────────────────── + // Locked cards unlock only via this machine's exact key endpoints; + // anything else fails closed here, before key generation. + let snapshot = { + let owner_keys = state.signing_keys().ok(); + let records = { + let _store_guard = state + .managed_agents_store_lock + .lock() + .map_err(|e| e.to_string())?; + load_managed_agents(&app)? + }; + decode_snapshot_for_import(&input.file_bytes, owner_keys.as_ref(), &records)?.0 + }; let display_name = snapshot.profile.display_name.trim().to_string(); if display_name.is_empty() { diff --git a/desktop/src-tauri/src/commands/personas/snapshot/tests.rs b/desktop/src-tauri/src/commands/personas/snapshot/tests.rs index 4289310280..c453b09a9d 100644 --- a/desktop/src-tauri/src/commands/personas/snapshot/tests.rs +++ b/desktop/src-tauri/src/commands/personas/snapshot/tests.rs @@ -564,7 +564,7 @@ fn import_preview_includes_exported_definition_metadata() { let bytes = crate::managed_agents::agent_snapshot::encode_snapshot_json(&snapshot).unwrap(); let decoded = decode_snapshot_from_bytes(&bytes).unwrap(); - let preview = build_agent_snapshot_import_preview(&decoded); + let preview = build_agent_snapshot_import_preview(&decoded, false).unwrap(); assert!(preview.is_builtin); assert_eq!(preview.model.as_deref(), Some("claude-opus-4-5")); @@ -949,51 +949,14 @@ fn test_parse_format_is_png_invalid_returns_error() { } // ── Export: validate_snapshot_encode_size ──────────────────────────────────── -// -// Tests call `validate_snapshot_encode_size` directly so they prove the exact -// production guard — not a manual reconstruction. Removing or reversing the -// check in production code will cause these tests to fail. -/// JSON: boundary-1 passes, boundary is the last legal byte count. -#[test] -fn validate_encode_size_json_at_boundary_minus_1_passes() { - assert!(super::validate_snapshot_encode_size(MAX_SNAPSHOT_JSON_BYTES - 1, false).is_ok()); -} +#[path = "tests_memory_entries.rs"] +mod memory_entries; -/// JSON: exactly at the boundary is the last accepted size. -#[test] -fn validate_encode_size_json_at_boundary_passes() { - assert!(super::validate_snapshot_encode_size(MAX_SNAPSHOT_JSON_BYTES, false).is_ok()); -} +#[path = "tests_encode_size.rs"] +mod encode_size; -/// JSON: boundary+1 is rejected. -#[test] -fn validate_encode_size_json_over_boundary_is_rejected() { - let err = super::validate_snapshot_encode_size(MAX_SNAPSHOT_JSON_BYTES + 1, false).unwrap_err(); - assert!( - err.contains("size limit"), - "error must mention size limit, got: {err}" - ); -} +// ── Import: decode_snapshot_for_import (locked cards) ───────────────────── -/// PNG: boundary-1 passes. -#[test] -fn validate_encode_size_png_at_boundary_minus_1_passes() { - assert!(super::validate_snapshot_encode_size(MAX_SNAPSHOT_PNG_BYTES - 1, true).is_ok()); -} - -/// PNG: exactly at the boundary passes. -#[test] -fn validate_encode_size_png_at_boundary_passes() { - assert!(super::validate_snapshot_encode_size(MAX_SNAPSHOT_PNG_BYTES, true).is_ok()); -} - -/// PNG: boundary+1 is rejected. -#[test] -fn validate_encode_size_png_over_boundary_is_rejected() { - let err = super::validate_snapshot_encode_size(MAX_SNAPSHOT_PNG_BYTES + 1, true).unwrap_err(); - assert!( - err.contains("size limit"), - "error must mention size limit, got: {err}" - ); -} +#[path = "tests_locked.rs"] +mod locked_import; diff --git a/desktop/src-tauri/src/commands/personas/snapshot/tests_encode_size.rs b/desktop/src-tauri/src/commands/personas/snapshot/tests_encode_size.rs new file mode 100644 index 0000000000..36eaa99716 --- /dev/null +++ b/desktop/src-tauri/src/commands/personas/snapshot/tests_encode_size.rs @@ -0,0 +1,55 @@ +//! Export-size guard tests for `validate_snapshot_encode_size`. +//! +//! Kept in a sibling file so `snapshot/tests.rs` stays under the +//! 1000-line gate; `#[path]`-included from there as a child module, +//! so `super::*` still resolves to the shared test imports. +//! +//! Tests call `validate_snapshot_encode_size` directly so they prove the +//! exact production guard — not a manual reconstruction. Removing or +//! reversing the check in production code will cause these tests to fail. + +use super::*; + +/// JSON: boundary-1 passes, boundary is the last legal byte count. +#[test] +fn validate_encode_size_json_at_boundary_minus_1_passes() { + assert!(validate_snapshot_encode_size(MAX_SNAPSHOT_JSON_BYTES - 1, false).is_ok()); +} + +/// JSON: exactly at the boundary is the last accepted size. +#[test] +fn validate_encode_size_json_at_boundary_passes() { + assert!(validate_snapshot_encode_size(MAX_SNAPSHOT_JSON_BYTES, false).is_ok()); +} + +/// JSON: boundary+1 is rejected. +#[test] +fn validate_encode_size_json_over_boundary_is_rejected() { + let err = validate_snapshot_encode_size(MAX_SNAPSHOT_JSON_BYTES + 1, false).unwrap_err(); + assert!( + err.contains("size limit"), + "error must mention size limit, got: {err}" + ); +} + +/// PNG: boundary-1 passes. +#[test] +fn validate_encode_size_png_at_boundary_minus_1_passes() { + assert!(validate_snapshot_encode_size(MAX_SNAPSHOT_PNG_BYTES - 1, true).is_ok()); +} + +/// PNG: exactly at the boundary passes. +#[test] +fn validate_encode_size_png_at_boundary_passes() { + assert!(validate_snapshot_encode_size(MAX_SNAPSHOT_PNG_BYTES, true).is_ok()); +} + +/// PNG: boundary+1 is rejected. +#[test] +fn validate_encode_size_png_over_boundary_is_rejected() { + let err = validate_snapshot_encode_size(MAX_SNAPSHOT_PNG_BYTES + 1, true).unwrap_err(); + assert!( + err.contains("size limit"), + "error must mention size limit, got: {err}" + ); +} diff --git a/desktop/src-tauri/src/commands/personas/snapshot/tests_locked.rs b/desktop/src-tauri/src/commands/personas/snapshot/tests_locked.rs new file mode 100644 index 0000000000..296444f78d --- /dev/null +++ b/desktop/src-tauri/src/commands/personas/snapshot/tests_locked.rs @@ -0,0 +1,129 @@ +//! Locked-card import tests for `decode_snapshot_for_import`. +//! +//! Kept in a sibling file so `snapshot/tests.rs` stays under the +//! 1000-line gate; `#[path]`-included from there as a child module, +//! so `super::*` still resolves to the shared test helpers. + +use super::*; +use crate::commands::personas::snapshot::import::{ + decode_snapshot_for_import, parse_snapshot_payload_from_bytes, +}; +use crate::managed_agents::agent_snapshot_envelope::{ + encode_locked_snapshot_png, encrypt_snapshot_envelope, ChunkPayload, LOCKED_CARD_REFUSAL, +}; + +/// Build a keyed instance record holding real key material, so the +/// agent-endpoint unlock path resolves exactly as production does. +fn record_for(agent: &nostr::Keys) -> ManagedAgentRecord { + ManagedAgentRecord { + pubkey: agent.public_key().to_hex(), + slug: None, + persona_id: Some("locked-test".to_string()), + private_key_nsec: nostr::ToBech32::to_bech32(agent.secret_key()).unwrap(), + ..make_definition("") + } +} + +fn locked_png(owner: &nostr::Keys, agent: &nostr::Keys) -> (AgentSnapshot, Vec) { + let snapshot = make_snapshot(MemoryLevel::None, vec![]); + let png = encode_locked_snapshot_png(&snapshot, owner, &agent.public_key(), None).unwrap(); + (snapshot, png) +} + +/// Owner identity key unlocks a locked card; `locked` is reported true. +#[test] +fn owner_endpoint_unlocks_locked_png() { + let (owner, agent) = (nostr::Keys::generate(), nostr::Keys::generate()); + let (snapshot, png) = locked_png(&owner, &agent); + let (decoded, locked) = decode_snapshot_for_import(&png, Some(&owner), &[]).unwrap(); + assert_eq!(decoded, snapshot); + assert!(locked); +} + +/// A local managed-agent record holding the agent nsec unlocks the card +/// even when the owner identity does not match (e.g. re-import on the +/// agent's own machine under a different owner identity). +#[test] +fn agent_record_endpoint_unlocks_locked_png() { + let (owner, agent) = (nostr::Keys::generate(), nostr::Keys::generate()); + let (snapshot, png) = locked_png(&owner, &agent); + let other_identity = nostr::Keys::generate(); + let records = vec![record_for(&agent)]; + let (decoded, locked) = + decode_snapshot_for_import(&png, Some(&other_identity), &records).unwrap(); + assert_eq!(decoded, snapshot); + assert!(locked); +} + +/// No matching endpoint → only the locked-card refusal, nothing else. +#[test] +fn stranger_fails_closed_with_refusal_only() { + let (owner, agent) = (nostr::Keys::generate(), nostr::Keys::generate()); + let (_snapshot, png) = locked_png(&owner, &agent); + let stranger = nostr::Keys::generate(); + let unrelated_record = record_for(&nostr::Keys::generate()); + let err = decode_snapshot_for_import(&png, Some(&stranger), &[unrelated_record]).unwrap_err(); + assert_eq!(err, LOCKED_CARD_REFUSAL); + // And with no key material at all. + let err = decode_snapshot_for_import(&png, None, &[]).unwrap_err(); + assert_eq!(err, LOCKED_CARD_REFUSAL); +} + +/// Plain snapshots pass through unchanged with `locked == false`, with or +/// without key material in scope. +#[test] +fn plain_snapshot_passes_through_unlocked() { + use crate::managed_agents::agent_snapshot::encode_snapshot_png; + let snapshot = make_snapshot(MemoryLevel::None, vec![]); + let png = encode_snapshot_png(&snapshot, None).unwrap(); + let owner = nostr::Keys::generate(); + let (decoded, locked) = decode_snapshot_for_import(&png, Some(&owner), &[]).unwrap(); + assert_eq!(decoded, snapshot); + assert!(!locked); + let (decoded, locked) = decode_snapshot_for_import(&png, None, &[]).unwrap(); + assert_eq!(decoded, snapshot); + assert!(!locked); +} + +/// The memory-consistency guard fires AFTER decryption too: a locked +/// envelope whose plaintext declares level none + non-empty entries is +/// rejected even for a legitimate endpoint. +#[test] +fn decrypted_manifest_memory_consistency_enforced() { + let (owner, agent) = (nostr::Keys::generate(), nostr::Keys::generate()); + let malformed = make_snapshot( + MemoryLevel::None, + vec![AgentSnapshotMemoryEntry { + slug: "core".to_string(), + body: "leaked".to_string(), + }], + ); + // encrypt_snapshot_envelope does not guard memory consistency (the + // PNG encoder does), so this constructs the malicious payload. + let envelope = encrypt_snapshot_envelope(&malformed, &owner, &agent.public_key()).unwrap(); + let json = serde_json::to_vec(&envelope).unwrap(); + let err = decode_snapshot_for_import(&json, Some(&owner), &[]).unwrap_err(); + assert!( + err.contains("'none' but entries are present"), + "post-decrypt consistency guard must fire, got: {err}" + ); +} + +/// Transit validation (`fetch_snapshot_bytes` path) accepts a locked PNG +/// without any key material — structural validation only, no decryption. +#[test] +fn transit_validation_accepts_locked_png_without_keys() { + let (owner, agent) = (nostr::Keys::generate(), nostr::Keys::generate()); + let (_snapshot, png) = locked_png(&owner, &agent); + let payload = parse_snapshot_payload_from_bytes(&png).unwrap(); + assert!(matches!(payload, ChunkPayload::Locked(_))); +} + +/// The keyless plain decoder refuses locked cards with the refusal. +#[test] +fn plain_decoder_refuses_locked_cards() { + let (owner, agent) = (nostr::Keys::generate(), nostr::Keys::generate()); + let (_snapshot, png) = locked_png(&owner, &agent); + let err = decode_snapshot_from_bytes(&png).unwrap_err(); + assert_eq!(err, LOCKED_CARD_REFUSAL); +} diff --git a/desktop/src-tauri/src/commands/personas/snapshot/tests_memory_entries.rs b/desktop/src-tauri/src/commands/personas/snapshot/tests_memory_entries.rs new file mode 100644 index 0000000000..b17efa1ad1 --- /dev/null +++ b/desktop/src-tauri/src/commands/personas/snapshot/tests_memory_entries.rs @@ -0,0 +1,55 @@ +//! Tests for `memory_entries_from_listing` — the shared level → entries +//! selection used by both snapshot export and card minting. Split from +//! `tests.rs` to keep that file under the 1000-line gate; `#[path]`-included +//! from there as a child module, so `super::*` resolves to `tests`'s parent +//! scope re-exports. + +use super::*; + +fn listing_fixture() -> crate::commands::engrams::AgentMemoryListing { + let entry = |slug: &str, body: &str| crate::commands::engrams::EngramEntry { + slug: slug.to_string(), + body: body.to_string(), + event_id: "e".repeat(64), + created_at: 1, + outgoing_refs: vec![], + }; + crate::commands::engrams::AgentMemoryListing { + core: Some(entry("core", "core body")), + memories: vec![entry("mem/a", "a body"), entry("mem/b", "b body")], + truncated: false, + fetched_at: 1, + } +} + +#[test] +fn memory_entries_core_takes_core_only() { + let entries = memory_entries_from_listing(listing_fixture(), MemoryLevel::Core); + assert_eq!(entries.len(), 1); + assert_eq!(entries[0].slug, "core"); + assert_eq!(entries[0].body, "core body"); +} + +#[test] +fn memory_entries_everything_appends_mem_entries_after_core() { + let entries = memory_entries_from_listing(listing_fixture(), MemoryLevel::Everything); + assert_eq!( + entries.iter().map(|e| e.slug.as_str()).collect::>(), + vec!["core", "mem/a", "mem/b"] + ); +} + +#[test] +fn memory_entries_missing_core_still_yields_mem_entries_for_everything() { + let mut listing = listing_fixture(); + listing.core = None; + let entries = memory_entries_from_listing(listing, MemoryLevel::Everything); + assert_eq!( + entries.iter().map(|e| e.slug.as_str()).collect::>(), + vec!["mem/a", "mem/b"] + ); + + let mut core_only = listing_fixture(); + core_only.core = None; + assert!(memory_entries_from_listing(core_only, MemoryLevel::Core).is_empty()); +} diff --git a/desktop/src-tauri/src/key_backup.rs b/desktop/src-tauri/src/key_backup.rs index f97bf95a67..e8fcc8abe4 100644 --- a/desktop/src-tauri/src/key_backup.rs +++ b/desktop/src-tauri/src/key_backup.rs @@ -133,9 +133,14 @@ pub fn backup_file_path(data_dir: &std::path::Path) -> std::path::PathBuf { data_dir.join(BACKUP_FILE_NAME) } -/// Atomically write `ncryptsec` to `path` with owner-only permissions, then -/// reread and byte-compare. Same crash-safety pattern as +/// Atomically write the app-managed `ncryptsec` backup with owner-only +/// permissions, then reread and byte-compare. Same crash-safety pattern as /// `app_state::save_key_file`. +/// +/// Portable exports selected through a native save panel must use +/// [`write_portable_backup_file`] instead: sandboxed macOS grants access to the +/// selected path, but not to the sibling temporary file this writer needs. +#[allow(dead_code)] // Retained for durable app-managed backups; portable exports must not use it. pub fn write_backup_file(path: &std::path::Path, ncryptsec: &str) -> Result<(), String> { use atomic_write_file::AtomicWriteFile; use std::io::Write; @@ -155,6 +160,56 @@ pub fn write_backup_file(path: &std::path::Path, ncryptsec: &str) -> Result<(), file.commit() .map_err(|e| format!("commit backup file: {e}"))?; + verify_backup_file(path, ncryptsec) +} + +/// Write a user-selected portable backup without creating a sibling file. +/// +/// Native macOS save panels authorize the exact selected path in protected +/// folders such as Downloads, not an atomic writer's hidden sibling. Opening +/// with `create_new` uses only that authorized path and also guarantees an +/// existing backup is never truncated: users must choose a new filename when +/// the destination already exists. After writing, the file is synced and its +/// persisted bytes are reread before success is reported. +pub fn write_portable_backup_file(path: &std::path::Path, ncryptsec: &str) -> Result<(), String> { + use std::io::Write; + + let mut options = std::fs::OpenOptions::new(); + options.write(true).create_new(true); + #[cfg(unix)] + { + use std::os::unix::fs::OpenOptionsExt; + options.mode(0o600); + } + + let mut file = options.open(path).map_err(|error| { + if error.kind() == std::io::ErrorKind::AlreadyExists { + "backup file already exists; choose a new filename so the existing backup stays safe" + .to_string() + } else { + format!("create portable backup file: {error}") + } + })?; + + let write_result = file + .write_all(ncryptsec.as_bytes()) + .map_err(|e| format!("write portable backup file: {e}")) + .and_then(|()| { + file.sync_all() + .map_err(|e| format!("sync portable backup file: {e}")) + }); + drop(file); + + let result = write_result.and_then(|()| verify_backup_file(path, ncryptsec)); + if result.is_err() { + // This function created the destination exclusively, so cleanup cannot + // clobber a backup that existed before the save attempt. + let _ = std::fs::remove_file(path); + } + result +} + +fn verify_backup_file(path: &std::path::Path, ncryptsec: &str) -> Result<(), String> { // Reread and byte-compare: only report success for bytes that are // actually on disk. let on_disk = std::fs::read_to_string(path).map_err(|e| format!("reread backup file: {e}"))?; diff --git a/desktop/src-tauri/src/key_backup_tests.rs b/desktop/src-tauri/src/key_backup_tests.rs index b9713201e1..35b486f78d 100644 --- a/desktop/src-tauri/src/key_backup_tests.rs +++ b/desktop/src-tauri/src/key_backup_tests.rs @@ -160,6 +160,45 @@ fn write_backup_file_overwrites_atomically() { assert_eq!(entries, vec![std::ffi::OsString::from(BACKUP_FILE_NAME)]); } +#[test] +fn write_portable_backup_file_persists_0600_without_a_sibling() { + let dir = tempfile::tempdir().unwrap(); + let path = dir.path().join("portable.ncryptsec"); + write_portable_backup_file(&path, SPEC_NCRYPTSEC).unwrap(); + + assert_eq!(std::fs::read_to_string(&path).unwrap(), SPEC_NCRYPTSEC); + let entries: Vec<_> = std::fs::read_dir(dir.path()) + .unwrap() + .map(|entry| entry.unwrap().file_name()) + .collect(); + assert_eq!( + entries, + vec![std::ffi::OsString::from("portable.ncryptsec")] + ); + + #[cfg(unix)] + { + use std::os::unix::fs::PermissionsExt; + let mode = std::fs::metadata(&path).unwrap().permissions().mode(); + assert_eq!(mode & 0o777, 0o600, "portable backup must be owner-only"); + } +} + +#[test] +fn write_portable_backup_file_preserves_an_existing_backup() { + let dir = tempfile::tempdir().unwrap(); + let path = dir.path().join("portable.ncryptsec"); + std::fs::write(&path, "ncryptsec1existing").unwrap(); + + let error = write_portable_backup_file(&path, SPEC_NCRYPTSEC).unwrap_err(); + + assert!(error.contains("already exists"), "{error}"); + assert_eq!( + std::fs::read_to_string(&path).unwrap(), + "ncryptsec1existing" + ); +} + #[test] fn delete_backup_file_is_idempotent() { let dir = tempfile::tempdir().unwrap(); diff --git a/desktop/src-tauri/src/lib.rs b/desktop/src-tauri/src/lib.rs index 6814008f0d..c4b733e3e0 100644 --- a/desktop/src-tauri/src/lib.rs +++ b/desktop/src-tauri/src/lib.rs @@ -65,10 +65,7 @@ use mesh_llm_stubs::*; #[cfg(all(feature = "mesh-llm", target_os = "macos"))] use shutdown::{hard_exit_after_mesh_shutdown, relaunch_after_mesh_shutdown}; use shutdown::{is_restart_request, shut_down_app}; -use std::sync::{ - atomic::{AtomicBool, Ordering}, - Arc, -}; +use std::sync::{atomic::AtomicBool, atomic::Ordering, Arc}; use tauri::{Emitter, Manager, RunEvent}; #[cfg(target_os = "macos")] use tauri::{Listener, WindowEvent}; @@ -851,6 +848,12 @@ pub fn run() { update_team, delete_team, export_agent_snapshot, + card_mint_key_status, + card_mint_save_openai_key, + mint_agent_card, + save_agent_card, + list_agent_cards, + load_agent_card, preview_agent_snapshot_import, confirm_agent_snapshot_import, encode_agent_snapshot_for_send, diff --git a/desktop/src-tauri/src/managed_agents/agent_snapshot.rs b/desktop/src-tauri/src/managed_agents/agent_snapshot.rs index 16a0d35b23..7c08e7095f 100644 --- a/desktop/src-tauri/src/managed_agents/agent_snapshot.rs +++ b/desktop/src-tauri/src/managed_agents/agent_snapshot.rs @@ -306,9 +306,22 @@ pub fn encode_snapshot_png( ); } - // Manifest → JSON → base64 for the tEXt chunk payload. + // Manifest → JSON for the tEXt chunk payload. The payload/PNG composition + // is shared with the locked-card encoder in `agent_snapshot_envelope`; + // plain cards remain byte-identical to the pre-envelope encoder. let json_bytes = encode_snapshot_json(snapshot)?; - let chunk_text = STANDARD.encode(&json_bytes); + encode_chunk_payload_png(&json_bytes, avatar_bytes) +} + +/// Encode arbitrary chunk-payload JSON (plain manifest or locked envelope) +/// into a PNG carrying it base64-encoded in the `buzz_agent_snapshot` tEXt +/// chunk. Shared by the plain encoder above and +/// `agent_snapshot_envelope::encode_locked_snapshot_png`. +pub(crate) fn encode_chunk_payload_png( + json_bytes: &[u8], + avatar_bytes: Option<&[u8]>, +) -> Result, String> { + let chunk_text = STANDARD.encode(json_bytes); // Use the avatar as the PNG image body, transcoding decodable non-PNG // avatars. Fall back to a minimal 1×1 transparent placeholder only when @@ -334,8 +347,11 @@ pub fn encode_snapshot_png( Ok(png_bytes) } -/// Decode a manifest from a `.agent.png` tEXt chunk. -pub fn decode_snapshot_png(png_bytes: &[u8]) -> Result { +/// Extract and base64-decode the raw `buzz_agent_snapshot` chunk payload +/// (JSON bytes) from a PNG, without interpreting it. The payload may be a +/// plain manifest or a locked envelope — callers dispatch on the parsed +/// `format` via `agent_snapshot_envelope::parse_chunk_payload`. +pub(crate) fn extract_chunk_payload_png(png_bytes: &[u8]) -> Result, String> { let decoder = Decoder::new(Cursor::new(png_bytes)); let reader = decoder .read_info() @@ -349,10 +365,18 @@ pub fn decode_snapshot_png(png_bytes: &[u8]) -> Result { .map(|c| c.text.as_str()) .ok_or_else(|| "PNG does not contain a buzz_agent_snapshot tEXt chunk".to_string())?; - let json_bytes = STANDARD + STANDARD .decode(chunk_text.trim()) - .map_err(|e| format!("Invalid base64 in PNG chunk: {e}"))?; + .map_err(|e| format!("Invalid base64 in PNG chunk: {e}")) +} +/// Decode a manifest from a `.agent.png` tEXt chunk. +/// +/// Plain snapshots only — a locked (encrypted) chunk payload fails here with +/// the manifest format error. Import paths that must handle locked cards go +/// through `agent_snapshot_envelope::parse_chunk_payload` instead. +pub fn decode_snapshot_png(png_bytes: &[u8]) -> Result { + let json_bytes = extract_chunk_payload_png(png_bytes)?; decode_snapshot_json(&json_bytes) } @@ -473,527 +497,5 @@ fn inject_text_chunk(png_bytes: &[u8], keyword: &str, text: &str) -> Result ManagedAgentRecord { - ManagedAgentRecord { - pubkey: "deadbeef".to_string(), - name: "Test Agent".to_string(), - display_name: Some("Test Agent Display".to_string()), - persona_id: Some("SENTINEL_PERSONA_ID".to_string()), // MUST NOT appear in snapshot - team_id: Some("SENTINEL_TEAM_ID".to_string()), // MUST NOT appear in snapshot - private_key_nsec: "nsec1secret".to_string(), // MUST NOT appear in snapshot - auth_tag: Some("auth-tag-secret".to_string()), // MUST NOT appear in snapshot - relay_url: "wss://relay.example.com".to_string(), // MUST NOT appear in snapshot - avatar_url: Some("https://example.com/avatar.png".to_string()), - acp_command: "/usr/local/bin/acp".to_string(), // MUST NOT appear in snapshot - agent_command: "goose".to_string(), // MUST NOT appear in snapshot - agent_command_override: Some("goose-override".to_string()), // MUST NOT appear - agent_args: vec!["--arg".to_string()], // MUST NOT appear in snapshot - mcp_command: "mcp-server".to_string(), // MUST NOT appear in snapshot - turn_timeout_seconds: 120, // deprecated, MUST NOT appear - idle_timeout_seconds: Some(30), - max_turn_duration_seconds: Some(600), - parallelism: 2, - system_prompt: Some("You are a test agent.".to_string()), - model: Some("claude-opus-4".to_string()), - provider: Some("anthropic".to_string()), - persona_source_version: Some("v1.0".to_string()), // MUST NOT appear - env_vars: { - let mut m = BTreeMap::new(); - m.insert("API_KEY".to_string(), "secret123".to_string()); // MUST NOT appear - m - }, - start_on_app_launch: true, - auto_restart_on_config_change: true, - runtime_pid: Some(12345), // MUST NOT appear - backend: BackendKind::Provider { - // MUST NOT appear — carries a provider secret - id: "SENTINEL_BACKEND_ID".to_string(), - config: serde_json::json!({"api_key": "SENTINEL_BACKEND_SECRET"}), - }, - backend_agent_id: Some("SENTINEL_BACKEND_AGENT_ID".to_string()), // MUST NOT appear - provider_binary_path: Some("/usr/bin/SENTINEL_PROVIDER_BINARY".to_string()), // MUST NOT appear - persona_team_dir: Some(std::path::PathBuf::from("SENTINEL_TEAM_DIR")), // MUST NOT appear - persona_name_in_team: Some("SENTINEL_NAME_IN_TEAM".to_string()), // MUST NOT appear - created_at: "2024-01-01T00:00:00Z".to_string(), - updated_at: "2024-01-02T00:00:00Z".to_string(), - last_started_at: Some("2024-01-03T00:00:00Z".to_string()), // MUST NOT appear - last_stopped_at: None, - last_exit_code: Some(0), // MUST NOT appear - last_error: Some("SENTINEL_LAST_ERROR".to_string()), // MUST NOT appear - last_error_code: Some(42), // MUST NOT appear - respond_to: RespondTo::default(), - respond_to_allowlist: vec!["pubkey1hex".to_string()], - slug: Some("test-agent".to_string()), - runtime: Some("goose".to_string()), - name_pool: vec!["Alice".to_string(), "Bob".to_string()], - is_builtin: false, - is_active: true, - shared: false, - source_team: Some("team-id-123".to_string()), // MUST NOT appear - source_team_persona_slug: Some("lep".to_string()), // MUST NOT appear - definition_respond_to: Some("allowlist".to_string()), - catalog_source: None, - definition_respond_to_allowlist: vec!["abc123def".to_string()], - definition_parallelism: Some(4), - relay_mesh: None, - } - } - - // ── Round-trip tests ────────────────────────────────────────────────────── - - #[test] - fn json_round_trip_config_only() { - let record = minimal_record(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); - let bytes = encode_snapshot_json(&snapshot).unwrap(); - let parsed = decode_snapshot_json(&bytes).unwrap(); - assert_eq!(parsed, snapshot); - } - - #[test] - fn json_round_trip_with_memory() { - let record = minimal_record(); - let entries = vec![ - AgentSnapshotMemoryEntry { - slug: "core".to_string(), - body: "I am a test agent.".to_string(), - }, - AgentSnapshotMemoryEntry { - slug: "mem/research".to_string(), - body: "Some research notes.".to_string(), - }, - ]; - let snapshot = build_snapshot(&record, MemoryLevel::Everything, entries, None); - let bytes = encode_snapshot_json(&snapshot).unwrap(); - let parsed = decode_snapshot_json(&bytes).unwrap(); - assert_eq!(parsed, snapshot); - } - - #[test] - fn png_round_trip_no_memory() { - let record = minimal_record(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); - let png_bytes = encode_snapshot_png(&snapshot, None).unwrap(); - let parsed = decode_snapshot_png(&png_bytes).unwrap(); - assert_eq!(parsed.definition.name, snapshot.definition.name); - assert_eq!(parsed.profile.display_name, snapshot.profile.display_name); - assert_eq!(parsed.memory.level, MemoryLevel::None); - } - - #[test] - fn png_round_trip_with_avatar_png() { - // Build a minimal PNG avatar. - let avatar = make_png_with_text("dummy", "value").unwrap(); - let record = minimal_record(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&avatar)); - // Avatar should be inlined as a data URL. - assert!(snapshot - .profile - .avatar_data_url - .as_deref() - .unwrap_or("") - .starts_with("data:image/png;base64,")); - - let png_bytes = encode_snapshot_png(&snapshot, Some(&avatar)).unwrap(); - let parsed = decode_snapshot_png(&png_bytes).unwrap(); - assert_eq!(parsed.definition.name, snapshot.definition.name); - } - - #[test] - fn png_snapshot_transcodes_jpeg_avatar_into_image_body() { - let avatar = image::DynamicImage::ImageRgb8(image::RgbImage::from_pixel( - 3, - 2, - image::Rgb([0x12, 0x34, 0x56]), - )); - let mut jpeg_bytes = Vec::new(); - avatar - .write_to(&mut Cursor::new(&mut jpeg_bytes), image::ImageFormat::Jpeg) - .unwrap(); - - let snapshot = build_snapshot( - &minimal_record(), - MemoryLevel::None, - vec![], - Some(&jpeg_bytes), - ); - let png_bytes = encode_snapshot_png(&snapshot, Some(&jpeg_bytes)).unwrap(); - let decoder = Decoder::new(Cursor::new(png_bytes)); - let reader = decoder.read_info().unwrap(); - - assert_eq!((reader.info().width, reader.info().height), (3, 2)); - } - - // ── PNG memory parity ───────────────────────────────────────────────────── - - #[test] - fn png_round_trip_with_core_memory() { - let record = minimal_record(); - let entries = vec![AgentSnapshotMemoryEntry { - slug: "core".to_string(), - body: "remember this".to_string(), - }]; - let snapshot = build_snapshot(&record, MemoryLevel::Core, entries, None); - - let png_bytes = encode_snapshot_png(&snapshot, None).unwrap(); - let parsed = decode_snapshot_png(&png_bytes).unwrap(); - - assert_eq!(parsed.memory, snapshot.memory); - } - - #[test] - fn png_round_trip_with_everything_memory() { - let record = minimal_record(); - let entries = vec![ - AgentSnapshotMemoryEntry { - slug: "core".to_string(), - body: "remember this".to_string(), - }, - AgentSnapshotMemoryEntry { - slug: "mem/notes".to_string(), - body: "private notes".to_string(), - }, - ]; - let snapshot = build_snapshot(&record, MemoryLevel::Everything, entries, None); - - let png_bytes = encode_snapshot_png(&snapshot, None).unwrap(); - let parsed = decode_snapshot_png(&png_bytes).unwrap(); - - assert_eq!(parsed.memory, snapshot.memory); - } - - #[test] - fn png_export_with_no_memory_succeeds() { - let record = minimal_record(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); - assert!(encode_snapshot_png(&snapshot, None).is_ok()); - } - - #[test] - fn png_export_rejects_none_level_with_nonempty_entries() { - // Inconsistent state: level == None but entries is non-empty. - // The encoder must reject this to prevent a memory-leak bypass. - let record = minimal_record(); - let entries = vec![AgentSnapshotMemoryEntry { - slug: "core".to_string(), - body: "leaked memory".to_string(), - }]; - // Build with entries, then override level to None in the struct. - let mut snapshot = build_snapshot(&record, MemoryLevel::Core, entries, None); - snapshot.memory.level = MemoryLevel::None; // force inconsistency - let result = encode_snapshot_png(&snapshot, None); - assert!( - result.is_err(), - "PNG encoder must reject level=None with non-empty entries" - ); - assert!( - result - .unwrap_err() - .contains("memory.level 'none' and non-empty memory entries"), - "Error must explain the malformed memory state" - ); - } - - // ── Secret exclusion tests ──────────────────────────────────────────────── - // - // These tests assert that every field in the exclusion list is absent from - // the serialized snapshot. We serialize to JSON and assert the key is NOT - // present. - - fn snapshot_json_string(record: &ManagedAgentRecord) -> String { - let snapshot = build_snapshot(record, MemoryLevel::None, vec![], None); - let bytes = encode_snapshot_json(&snapshot).unwrap(); - String::from_utf8(bytes).unwrap() - } - - #[test] - fn secret_exclusion_private_key_nsec_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("nsec1secret"), - "nsec must not appear in snapshot" - ); - assert!( - !json.contains("privateKeyNsec") && !json.contains("private_key_nsec"), - "privateKeyNsec field must not appear in snapshot" - ); - } - - #[test] - fn secret_exclusion_auth_tag_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("auth-tag-secret"), - "auth_tag value must not appear in snapshot" - ); - assert!( - !json.contains("authTag") && !json.contains("auth_tag"), - "authTag field must not appear in snapshot" - ); - } - - #[test] - fn secret_exclusion_env_vars_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("API_KEY") && !json.contains("secret123"), - "env_vars content must not appear in snapshot" - ); - assert!( - !json.contains("envVars") && !json.contains("env_vars"), - "envVars field must not appear in snapshot" - ); - } - - #[test] - fn secret_exclusion_relay_url_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("wss://relay.example.com"), - "relay_url value must not appear in snapshot" - ); - assert!( - !json.contains("relayUrl") && !json.contains("relay_url"), - "relayUrl field must not appear in snapshot" - ); - } - - #[test] - fn snapshot_omits_removed_mcp_toolsets_config() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("mcpToolsets") && !json.contains("mcp_toolsets"), - "removed MCP toolsets config must not re-enter snapshots" - ); - } - - #[test] - fn secret_exclusion_machine_commands_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - // acp_command / agent_command / agent_command_override / agent_args / mcp_command - assert!( - !json.contains("/usr/local/bin/acp"), - "acp_command path must not appear" - ); - assert!( - !json.contains("acpCommand") && !json.contains("acp_command"), - "acpCommand field must not appear" - ); - assert!( - !json.contains("agentCommand") && !json.contains("agent_command"), - "agentCommand field must not appear" - ); - assert!( - !json.contains("mcpCommand") && !json.contains("mcp_command"), - "mcpCommand field must not appear" - ); - } - - #[test] - fn secret_exclusion_runtime_state_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("runtimePid") && !json.contains("runtime_pid"), - "runtimePid must not appear" - ); - assert!( - !json.contains("backendAgentId") && !json.contains("backend_agent_id"), - "backendAgentId must not appear" - ); - assert!( - !json.contains("SENTINEL_BACKEND_AGENT_ID"), - "backendAgentId value must not appear" - ); - assert!( - !json.contains("providerBinaryPath") && !json.contains("provider_binary_path"), - "providerBinaryPath must not appear" - ); - assert!( - !json.contains("SENTINEL_PROVIDER_BINARY"), - "providerBinaryPath value must not appear" - ); - assert!( - !json.contains("lastStartedAt") && !json.contains("last_started_at"), - "lastStartedAt must not appear" - ); - assert!( - !json.contains("lastExitCode") && !json.contains("last_exit_code"), - "lastExitCode must not appear" - ); - // backend blob — neither the type tag nor provider secret must leak. - assert!( - !json.contains("\"backend\"") && !json.contains("backend"), - "backend field must not appear" - ); - assert!( - !json.contains("SENTINEL_BACKEND_ID") && !json.contains("SENTINEL_BACKEND_SECRET"), - "backend config values must not appear" - ); - // last_error / last_error_code - assert!( - !json.contains("lastError") && !json.contains("last_error"), - "lastError must not appear" - ); - assert!( - !json.contains("SENTINEL_LAST_ERROR"), - "lastError value must not appear" - ); - assert!( - !json.contains("lastErrorCode") && !json.contains("last_error_code"), - "lastErrorCode must not appear" - ); - } - - #[test] - fn secret_exclusion_lineage_ids_absent() { - let record = minimal_record(); - let json = snapshot_json_string(&record); - assert!( - !json.contains("team-id-123"), - "source_team value must not appear" - ); - assert!( - !json.contains("sourceTeam") && !json.contains("source_team"), - "sourceTeam field must not appear" - ); - assert!( - !json.contains("sourceTeamPersonaSlug"), - "sourceTeamPersonaSlug must not appear" - ); - assert!( - !json.contains("personaSourceVersion") && !json.contains("persona_source_version"), - "personaSourceVersion must not appear" - ); - // personaId - assert!( - !json.contains("personaId") && !json.contains("persona_id"), - "personaId field must not appear" - ); - assert!( - !json.contains("SENTINEL_PERSONA_ID"), - "personaId value must not appear" - ); - // teamId - assert!( - !json.contains("teamId") && !json.contains("team_id"), - "teamId field must not appear" - ); - assert!( - !json.contains("SENTINEL_TEAM_ID"), - "teamId value must not appear" - ); - // personaTeamDir - assert!( - !json.contains("personaTeamDir") && !json.contains("persona_team_dir"), - "personaTeamDir field must not appear" - ); - assert!( - !json.contains("SENTINEL_TEAM_DIR"), - "personaTeamDir value must not appear" - ); - // personaNameInTeam - assert!( - !json.contains("personaNameInTeam") && !json.contains("persona_name_in_team"), - "personaNameInTeam field must not appear" - ); - assert!( - !json.contains("SENTINEL_NAME_IN_TEAM"), - "personaNameInTeam value must not appear" - ); - } - - // ── Definition field presence tests ────────────────────────────────────── - - #[test] - fn definition_fields_present_in_snapshot() { - let record = minimal_record(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); - - assert_eq!(snapshot.definition.name, "Test Agent Display"); - assert!(!snapshot.definition.source_is_builtin); - assert_eq!( - snapshot.definition.system_prompt.as_deref(), - Some("You are a test agent.") - ); - assert_eq!(snapshot.definition.runtime.as_deref(), Some("goose")); - assert_eq!(snapshot.definition.model.as_deref(), Some("claude-opus-4")); - assert_eq!(snapshot.definition.provider.as_deref(), Some("anthropic")); - assert_eq!(snapshot.definition.name_pool, vec!["Alice", "Bob"]); - // definition_respond_to maps to respond_to in the snapshot definition - assert_eq!(snapshot.definition.respond_to.as_deref(), Some("allowlist")); - // definition_respond_to_allowlist should be included - assert!(!snapshot.definition.respond_to_allowlist.is_empty()); - } - - #[test] - fn profile_fields_present_in_snapshot() { - let record = minimal_record(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); - assert_eq!(snapshot.profile.display_name, "Test Agent Display"); - // No bytes → should fall back to avatar_url - assert_eq!( - snapshot.profile.avatar_url.as_deref(), - Some("https://example.com/avatar.png") - ); - assert!(snapshot.profile.avatar_data_url.is_none()); - } - - #[test] - fn avatar_inlined_when_under_size_limit() { - let record = minimal_record(); - let small_png = make_png_with_text("k", "v").unwrap(); - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&small_png)); - assert!(snapshot.profile.avatar_data_url.is_some()); - assert!(snapshot.profile.avatar_url.is_none()); - } - - #[test] - fn avatar_url_fallback_when_over_size_limit() { - let mut record = minimal_record(); - record.avatar_url = Some("https://example.com/big.png".to_string()); - // Synthesize oversized avatar bytes (> 2 MB) — just a large zeroed vec. - let big_bytes = vec![0u8; MAX_AVATAR_INLINE_BYTES + 1]; - let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&big_bytes)); - assert!(snapshot.profile.avatar_data_url.is_none()); - assert_eq!( - snapshot.profile.avatar_url.as_deref(), - Some("https://example.com/big.png") - ); - } - - // ── Format/version validation ───────────────────────────────────────────── - - #[test] - fn invalid_format_discriminator_is_rejected() { - let mut snapshot = build_snapshot(&minimal_record(), MemoryLevel::None, vec![], None); - snapshot.format = "not-a-buzz-snapshot".to_string(); - let bytes = serde_json::to_vec(&snapshot).unwrap(); - let result = decode_snapshot_json(&bytes); - assert!(result.is_err()); - assert!(result.unwrap_err().contains("Unsupported snapshot format")); - } - - #[test] - fn unsupported_version_is_rejected() { - let mut snapshot = build_snapshot(&minimal_record(), MemoryLevel::None, vec![], None); - snapshot.version = 99; - let bytes = serde_json::to_vec(&snapshot).unwrap(); - let result = decode_snapshot_json(&bytes); - assert!(result.is_err()); - assert!(result.unwrap_err().contains("Unsupported snapshot version")); - } -} +#[path = "agent_snapshot_tests.rs"] +mod tests; diff --git a/desktop/src-tauri/src/managed_agents/agent_snapshot_envelope.rs b/desktop/src-tauri/src/managed_agents/agent_snapshot_envelope.rs new file mode 100644 index 0000000000..8508c27073 --- /dev/null +++ b/desktop/src-tauri/src/managed_agents/agent_snapshot_envelope.rs @@ -0,0 +1,638 @@ +//! Locked (encrypted) agent-card envelope — NIP-44 v2 over the snapshot manifest. +//! +//! A locked card carries the same `buzz_agent_snapshot` tEXt chunk as a plain +//! card, but the chunk JSON is a typed outer envelope whose ciphertext +//! decrypts to the ordinary manifest. The NIP-44 v2 conversation key is +//! symmetric over the (owner, agent) pair, so BOTH the owner's and the +//! agent's nsec decrypt the card — nobody else's does (NIP-AE's scheme). +//! +//! Wire contract (agreed with Wren, buzz-agent-trading-cards thread): +//! - Plain cards keep today's exact bytes; detection dispatches once on the +//! exact `format` discriminator and rejects unknown versions/schemes +//! rather than falling through to manifest parsing. +//! - Key lookup is exact-endpoint only: the owner identity key when its +//! pubkey equals `ownerPubkey`, or a hydrated local managed-agent record +//! whose record pubkey AND derived-secret pubkey equal `agentPubkey`. +//! No trial decryption; anything else fails closed as locked. +//! - Caps beyond the outer 10 MiB PNG gate: 65,535-byte NIP-44 plaintext +//! limit on the serialized manifest BEFORE encryption; envelope JSON and +//! ciphertext are capped before serde/base64/decrypt work; decrypted bytes +//! are capped before snapshot parsing. +//! - Decrypt/auth failures return only the locked-card refusal — never +//! partial plaintext or crypto details. + +use buzz_core_pkg::engram::NIP44_PLAINTEXT_MAX; +use nostr::nips::nip44::{self, Version}; +use nostr::{Keys, PublicKey, SecretKey}; +use serde::{Deserialize, Serialize}; + +use super::agent_snapshot::{ + decode_snapshot_json, encode_chunk_payload_png, encode_snapshot_json, AgentSnapshot, + MemoryLevel, FORMAT_DISCRIMINATOR, +}; +use super::types::ManagedAgentRecord; + +/// Discriminator for the locked envelope. Distinct from the plain manifest's +/// `buzz-agent-snapshot` so detection never guesses. +pub const LOCKED_FORMAT: &str = "buzz-agent-snapshot-encrypted"; +/// Envelope schema version this module produces and accepts. +pub const LOCKED_VERSION: u32 = 1; +/// Encryption scheme identifier this module produces and accepts. +pub const LOCKED_SCHEME: &str = "nip44-v2"; + +/// A max-size NIP-44 v2 payload (1 version + 32 nonce + 2 len + 65,536 +/// padded + 32 MAC = 65,603 bytes) base64-encodes to 87,472 chars. +/// Anything larger is rejected before base64/decrypt work. +pub const MAX_LOCKED_CIPHERTEXT_BYTES: usize = 90_000; +/// Envelope JSON = ciphertext + two pubkeys + fixed keys. Rejected before +/// typed deserialization. +pub const MAX_LOCKED_ENVELOPE_JSON_BYTES: usize = MAX_LOCKED_CIPHERTEXT_BYTES + 1024; + +/// The only error a failed unlock may surface. Deliberately says nothing +/// about which key was tried or why decryption failed. +pub const LOCKED_CARD_REFUSAL: &str = + "This card is locked to its owner and agent. Only they can import it."; + +// ── Envelope types ──────────────────────────────────────────────────────────── + +/// Typed outer envelope stored (base64 JSON) in the `buzz_agent_snapshot` +/// chunk of a locked card. +#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)] +#[serde(rename_all = "camelCase")] +pub struct LockedSnapshotEnvelope { + /// Always [`LOCKED_FORMAT`]. + pub format: String, + /// Always [`LOCKED_VERSION`]. + pub version: u32, + pub encryption: LockedEncryption, +} + +#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)] +#[serde(rename_all = "camelCase")] +pub struct LockedEncryption { + /// Always [`LOCKED_SCHEME`]. + pub scheme: String, + /// Owner identity pubkey (64 lowercase hex). Plaintext so a decryptor + /// knows which counterparty to pair with. + pub owner_pubkey: String, + /// Agent instance pubkey (64 lowercase hex). + pub agent_pubkey: String, + /// NIP-44 v2 ciphertext (base64) of the plain manifest JSON. + pub ciphertext: String, +} + +/// Result of parsing a chunk payload: either today's plain manifest or a +/// validated locked envelope. The plain manifest is boxed because it may +/// inline a multi-KB avatar data URL, dwarfing the envelope variant. +#[derive(Debug)] +pub enum ChunkPayload { + Plain(Box), + Locked(LockedSnapshotEnvelope), +} + +/// Minimal probe used to read the `format` discriminator without building a +/// full JSON tree for large plain manifests. +#[derive(Deserialize)] +struct FormatProbe { + #[serde(default)] + format: Option, +} + +// ── Validation ──────────────────────────────────────────────────────────────── + +/// Canonical pubkey check: exactly 64 lowercase hex chars that parse as a +/// valid x-only pubkey. Lowercase is required so string comparisons against +/// record pubkeys (always `to_hex()` output) stay sound. Curve validation is +/// explicit: nostr's `PublicKey::from_hex` only decodes 32 bytes and defers +/// lift-x validation to `xonly()`, so a non-point like `"f" * 64` would +/// otherwise pass structurally and fail only at decrypt time. +pub(crate) fn parse_canonical_pubkey(field: &str, value: &str) -> Result { + if value.len() != 64 + || !value + .chars() + .all(|c| c.is_ascii_digit() || ('a'..='f').contains(&c)) + { + return Err(format!( + "Locked card envelope has a malformed {field} (expected 64 lowercase hex chars)." + )); + } + let pubkey = PublicKey::from_hex(value) + .map_err(|_| format!("Locked card envelope has an invalid {field}."))?; + pubkey + .xonly() + .map_err(|_| format!("Locked card envelope has an invalid {field} (not a curve point)."))?; + Ok(pubkey) +} + +/// Structural validation of a locked envelope: exact version + scheme, +/// canonical pubkeys, distinct endpoints, bounded ciphertext. Does no +/// key lookup or crypto. +pub fn validate_envelope( + envelope: &LockedSnapshotEnvelope, +) -> Result<(PublicKey, PublicKey), String> { + if envelope.format != LOCKED_FORMAT { + return Err(format!( + "Unsupported locked card format: {:?} (expected {LOCKED_FORMAT:?})", + envelope.format + )); + } + if envelope.version != LOCKED_VERSION { + return Err(format!( + "Unsupported locked card envelope version: {} (expected {LOCKED_VERSION})", + envelope.version + )); + } + if envelope.encryption.scheme != LOCKED_SCHEME { + return Err(format!( + "Unsupported locked card encryption scheme: {:?} (expected {LOCKED_SCHEME:?})", + envelope.encryption.scheme + )); + } + let owner = parse_canonical_pubkey("ownerPubkey", &envelope.encryption.owner_pubkey)?; + let agent = parse_canonical_pubkey("agentPubkey", &envelope.encryption.agent_pubkey)?; + if owner == agent { + return Err("Locked card envelope owner and agent pubkeys must differ.".to_string()); + } + if envelope.encryption.ciphertext.len() > MAX_LOCKED_CIPHERTEXT_BYTES { + return Err("Locked card ciphertext exceeds the maximum size.".to_string()); + } + if envelope.encryption.ciphertext.is_empty() { + return Err("Locked card ciphertext is empty.".to_string()); + } + Ok((owner, agent)) +} + +// ── Dispatch ────────────────────────────────────────────────────────────────── + +/// Parse a raw chunk payload (JSON bytes from `extract_chunk_payload_png` or +/// an `.agent.json` file) and dispatch on the exact `format` discriminator. +/// +/// - `buzz-agent-snapshot` → full plain-manifest decode + validation. +/// - `buzz-agent-snapshot-encrypted` → size caps, typed envelope parse, +/// structural validation. No decryption happens here. +/// - anything else (including missing `format`) → error, never a fall-through. +pub fn parse_chunk_payload(json_bytes: &[u8]) -> Result { + let probe: FormatProbe = + serde_json::from_slice(json_bytes).map_err(|e| format!("Invalid snapshot JSON: {e}"))?; + match probe.format.as_deref() { + Some(f) if f == FORMAT_DISCRIMINATOR => Ok(ChunkPayload::Plain(Box::new( + decode_snapshot_json(json_bytes)?, + ))), + Some(f) if f == LOCKED_FORMAT => { + // Cap the envelope JSON before typed deserialization; a locked + // envelope is small by construction (unlike plain manifests, + // which may inline a multi-MB avatar). + if json_bytes.len() > MAX_LOCKED_ENVELOPE_JSON_BYTES { + return Err("Locked card envelope exceeds the maximum size.".to_string()); + } + let envelope: LockedSnapshotEnvelope = serde_json::from_slice(json_bytes) + .map_err(|e| format!("Invalid locked card envelope: {e}"))?; + validate_envelope(&envelope)?; + Ok(ChunkPayload::Locked(envelope)) + } + Some(other) => Err(format!("Unsupported snapshot format: {other:?}")), + None => Err("Snapshot payload has no format discriminator.".to_string()), + } +} + +// ── Encrypt ─────────────────────────────────────────────────────────────────── + +/// Encrypt a snapshot manifest into a locked envelope under the NIP-44 v2 +/// conversation key for (owner secret, agent pubkey). +/// +/// Fails clearly (never silently truncates) when the serialized manifest +/// exceeds the NIP-44 plaintext limit. +pub fn encrypt_snapshot_envelope( + snapshot: &AgentSnapshot, + owner_keys: &Keys, + agent_pubkey: &PublicKey, +) -> Result { + let json_bytes = encode_snapshot_json(snapshot)?; + if json_bytes.len() > NIP44_PLAINTEXT_MAX { + return Err(format!( + "Agent manifest is too large to lock ({} bytes; the encrypted \ + format caps at {NIP44_PLAINTEXT_MAX}). Reduce the avatar size \ + or mint an unlocked card.", + json_bytes.len() + )); + } + let plaintext = std::str::from_utf8(&json_bytes) + .map_err(|e| format!("Manifest JSON was not UTF-8: {e}"))?; + let ciphertext = nip44::encrypt( + owner_keys.secret_key(), + agent_pubkey, + plaintext, + Version::V2, + ) + .map_err(|e| format!("Failed to encrypt card manifest: {e}"))?; + + Ok(LockedSnapshotEnvelope { + format: LOCKED_FORMAT.to_string(), + version: LOCKED_VERSION, + encryption: LockedEncryption { + scheme: LOCKED_SCHEME.to_string(), + owner_pubkey: owner_keys.public_key().to_hex(), + agent_pubkey: agent_pubkey.to_hex(), + ciphertext, + }, + }) +} + +/// Encode a snapshot into a LOCKED `.agent.png`: encrypt the manifest into +/// the envelope, then compose the PNG through the same chunk encoder plain +/// cards use. Mirrors `encode_snapshot_png`'s structural memory guard. +pub fn encode_locked_snapshot_png( + snapshot: &AgentSnapshot, + owner_keys: &Keys, + agent_pubkey: &PublicKey, + avatar_bytes: Option<&[u8]>, +) -> Result, String> { + if snapshot.memory.level == MemoryLevel::None && !snapshot.memory.entries.is_empty() { + return Err( + "Cannot write a snapshot with memory.level 'none' and non-empty memory entries." + .to_string(), + ); + } + let envelope = encrypt_snapshot_envelope(snapshot, owner_keys, agent_pubkey)?; + let envelope_json = serde_json::to_vec(&envelope) + .map_err(|e| format!("Failed to serialize locked card envelope: {e}"))?; + encode_chunk_payload_png(&envelope_json, avatar_bytes) +} + +// ── Decrypt ─────────────────────────────────────────────────────────────────── + +/// Exact-endpoint key resolution (no trial decryption): +/// - the owner identity secret, only when its pubkey equals `ownerPubkey`; +/// - a hydrated local managed-agent record whose record pubkey AND +/// derived-secret pubkey both equal `agentPubkey`. +/// +/// Returns `None` when neither exact endpoint exists — callers fail closed +/// with [`LOCKED_CARD_REFUSAL`]. +pub fn resolve_unlock_secret( + envelope: &LockedSnapshotEnvelope, + owner_keys: Option<&Keys>, + records: &[ManagedAgentRecord], +) -> Option { + if let Some(keys) = owner_keys { + if keys.public_key().to_hex() == envelope.encryption.owner_pubkey { + return Some(keys.secret_key().clone()); + } + } + let record = records + .iter() + .find(|r| r.pubkey == envelope.encryption.agent_pubkey)?; + let agent_keys = Keys::parse(record.private_key_nsec.trim()).ok()?; + if agent_keys.public_key().to_hex() != envelope.encryption.agent_pubkey { + return None; + } + Some(agent_keys.secret_key().clone()) +} + +/// Decrypt a validated envelope with `my_secret`, which must be one of the +/// envelope's two exact endpoints (its derived pubkey selects the +/// counterparty). Returns the decoded, validated snapshot manifest. +/// +/// Every auth/crypto failure maps to [`LOCKED_CARD_REFUSAL`] — nothing about +/// the failure mode leaks. Manifest decode errors after a successful decrypt +/// are surfaced normally (the caller proved key possession). +pub fn decrypt_envelope( + envelope: &LockedSnapshotEnvelope, + my_secret: &SecretKey, +) -> Result { + let (owner_pub, agent_pub) = validate_envelope(envelope)?; + let my_pub = Keys::new(my_secret.clone()).public_key(); + let counterparty = if my_pub == owner_pub { + agent_pub + } else if my_pub == agent_pub { + owner_pub + } else { + return Err(LOCKED_CARD_REFUSAL.to_string()); + }; + + let plaintext = nip44::decrypt(my_secret, &counterparty, &envelope.encryption.ciphertext) + .map_err(|_| LOCKED_CARD_REFUSAL.to_string())?; + if plaintext.len() > NIP44_PLAINTEXT_MAX { + return Err(LOCKED_CARD_REFUSAL.to_string()); + } + decode_snapshot_json(plaintext.as_bytes()) +} + +// ── Tests ───────────────────────────────────────────────────────────────────── + +#[cfg(test)] +mod tests { + use super::*; + use crate::managed_agents::agent_snapshot::{ + extract_chunk_payload_png, AgentSnapshotDefinition, AgentSnapshotMemory, + AgentSnapshotProfile, FORMAT_VERSION, + }; + + fn sample_snapshot() -> AgentSnapshot { + AgentSnapshot { + format: FORMAT_DISCRIMINATOR.to_string(), + version: FORMAT_VERSION, + definition: AgentSnapshotDefinition { + name: "Locked Test".to_string(), + system_prompt: Some("You are a locked test agent.".to_string()), + runtime: None, + model: None, + provider: None, + parallelism: Some(1), + respond_to: None, + respond_to_allowlist: Vec::new(), + name_pool: Vec::new(), + idle_timeout_seconds: None, + max_turn_duration_seconds: None, + source_is_builtin: false, + }, + profile: AgentSnapshotProfile { + display_name: "Locked Test".to_string(), + about: None, + avatar_data_url: None, + avatar_url: None, + }, + memory: AgentSnapshotMemory { + level: MemoryLevel::None, + entries: Vec::new(), + }, + } + } + + fn owner_agent_keys() -> (Keys, Keys) { + (Keys::generate(), Keys::generate()) + } + + /// Minimal hydrated record for endpoint-resolution tests. Only the + /// pubkey/nsec pair matters here. + fn record_with_keys(pubkey: String, private_key_nsec: String) -> ManagedAgentRecord { + ManagedAgentRecord { + pubkey, + name: "Locked Test".to_string(), + persona_id: None, + private_key_nsec, + auth_tag: None, + relay_url: "ws://localhost:3000".to_string(), + avatar_url: None, + acp_command: "buzz-acp".to_string(), + agent_command: "goose".to_string(), + agent_args: vec![], + mcp_command: String::new(), + turn_timeout_seconds: 300, + idle_timeout_seconds: None, + max_turn_duration_seconds: None, + parallelism: 1, + system_prompt: None, + model: None, + env_vars: std::collections::BTreeMap::new(), + start_on_app_launch: false, + auto_restart_on_config_change: true, + runtime_pid: None, + backend: crate::managed_agents::types::BackendKind::Local, + backend_agent_id: None, + provider_binary_path: None, + team_id: None, + persona_team_dir: None, + persona_name_in_team: None, + created_at: String::new(), + updated_at: String::new(), + last_started_at: None, + last_stopped_at: None, + last_exit_code: None, + last_error: None, + last_error_code: None, + respond_to: crate::managed_agents::types::RespondTo::OwnerOnly, + respond_to_allowlist: vec![], + display_name: None, + slug: None, + runtime: None, + name_pool: Vec::new(), + is_builtin: false, + is_active: true, + shared: false, + source_team: None, + source_team_persona_slug: None, + catalog_source: None, + definition_respond_to: None, + definition_respond_to_allowlist: Vec::new(), + definition_parallelism: None, + relay_mesh: None, + agent_command_override: None, + persona_source_version: None, + provider: None, + } + } + + fn locked_envelope() -> (LockedSnapshotEnvelope, Keys, Keys) { + let (owner, agent) = owner_agent_keys(); + let env = + encrypt_snapshot_envelope(&sample_snapshot(), &owner, &agent.public_key()).unwrap(); + (env, owner, agent) + } + + #[test] + fn owner_secret_decrypts() { + let (env, owner, _agent) = locked_envelope(); + let decoded = decrypt_envelope(&env, owner.secret_key()).unwrap(); + assert_eq!(decoded, sample_snapshot()); + } + + #[test] + fn agent_secret_decrypts() { + let (env, _owner, agent) = locked_envelope(); + let decoded = decrypt_envelope(&env, agent.secret_key()).unwrap(); + assert_eq!(decoded, sample_snapshot()); + } + + #[test] + fn unrelated_key_fails_closed_with_refusal_only() { + let (env, _owner, _agent) = locked_envelope(); + let stranger = Keys::generate(); + let err = decrypt_envelope(&env, stranger.secret_key()).unwrap_err(); + assert_eq!(err, LOCKED_CARD_REFUSAL); + } + + #[test] + fn tampered_ciphertext_fails_with_refusal_only() { + let (mut env, owner, _agent) = locked_envelope(); + // Flip a character mid-ciphertext (keep valid base64 alphabet). + let mid = env.encryption.ciphertext.len() / 2; + let mut bytes = env.encryption.ciphertext.into_bytes(); + bytes[mid] = if bytes[mid] == b'A' { b'B' } else { b'A' }; + env.encryption.ciphertext = String::from_utf8(bytes).unwrap(); + let err = decrypt_envelope(&env, owner.secret_key()).unwrap_err(); + assert_eq!(err, LOCKED_CARD_REFUSAL); + } + + #[test] + fn swapped_pubkeys_fail_closed_at_endpoint_resolution() { + let (mut env, owner, agent) = locked_envelope(); + std::mem::swap( + &mut env.encryption.owner_pubkey, + &mut env.encryption.agent_pubkey, + ); + // The NIP-44 conversation key is symmetric over the pair, so a swap + // cannot grant a stranger anything — but it desyncs the routing + // hints, and exact-endpoint resolution fails closed rather than + // guessing: the owner identity no longer matches `ownerPubkey`, and + // no local record holds the pubkey now in `agentPubkey`. + assert!(resolve_unlock_secret(&env, Some(&owner), &[]).is_none()); + let nsec = nostr::ToBech32::to_bech32(agent.secret_key()).unwrap(); + let record = record_with_keys(agent.public_key().to_hex(), nsec); + assert!(resolve_unlock_secret(&env, None, std::slice::from_ref(&record)).is_none()); + } + + #[test] + fn mislabeled_pubkey_fails_decryption_with_refusal_only() { + // Replacing `agentPubkey` with a third party's key makes the owner + // derive the wrong conversation key — the NIP-44 MAC fails and only + // the refusal surfaces. + let (mut env, owner, _agent) = locked_envelope(); + env.encryption.agent_pubkey = Keys::generate().public_key().to_hex(); + let err = decrypt_envelope(&env, owner.secret_key()).unwrap_err(); + assert_eq!(err, LOCKED_CARD_REFUSAL); + } + + #[test] + fn malformed_pubkeys_rejected_structurally() { + let (env, _owner, _agent) = locked_envelope(); + + let mut short = env.clone(); + short.encryption.owner_pubkey = "abc123".to_string(); + assert!(validate_envelope(&short).unwrap_err().contains("malformed")); + + let mut upper = env.clone(); + upper.encryption.agent_pubkey = upper.encryption.agent_pubkey.to_uppercase(); + assert!(validate_envelope(&upper).unwrap_err().contains("malformed")); + + // A 64-hex string that is not a curve point (lift-x fails for + // x = p-1... all-f) must be rejected STRUCTURALLY — before any key + // lookup or decrypt work — per the wire contract. + let mut not_a_point = env.clone(); + not_a_point.encryption.agent_pubkey = "f".repeat(64); + assert!(validate_envelope(¬_a_point) + .unwrap_err() + .contains("not a curve point")); + + let mut same = env; + same.encryption.agent_pubkey = same.encryption.owner_pubkey.clone(); + assert!(validate_envelope(&same).unwrap_err().contains("differ")); + } + + #[test] + fn unknown_format_version_scheme_rejected() { + let (env, ..) = locked_envelope(); + + let mut bad_version = env.clone(); + bad_version.version = 2; + assert!(validate_envelope(&bad_version) + .unwrap_err() + .contains("version")); + + let mut bad_scheme = env.clone(); + bad_scheme.encryption.scheme = "nip44-v3".to_string(); + assert!(validate_envelope(&bad_scheme) + .unwrap_err() + .contains("scheme")); + + // Unknown top-level format never falls through to manifest parsing. + let unknown = serde_json::json!({"format": "buzz-agent-snapshot-v9", "version": 1}); + let err = parse_chunk_payload(unknown.to_string().as_bytes()).unwrap_err(); + assert!(err.contains("Unsupported snapshot format"), "{err}"); + + let missing = serde_json::json!({"version": 1}); + let err = parse_chunk_payload(missing.to_string().as_bytes()).unwrap_err(); + assert!(err.contains("no format discriminator"), "{err}"); + } + + #[test] + fn plaintext_cap_enforced_before_encryption() { + let (owner, agent) = owner_agent_keys(); + let mut snapshot = sample_snapshot(); + // Inflate the manifest beyond the NIP-44 plaintext limit. + snapshot.definition.system_prompt = Some("x".repeat(NIP44_PLAINTEXT_MAX)); + let err = encrypt_snapshot_envelope(&snapshot, &owner, &agent.public_key()).unwrap_err(); + assert!(err.contains("too large to lock"), "{err}"); + } + + #[test] + fn ciphertext_and_envelope_caps_enforced_before_crypto() { + let (mut env, ..) = locked_envelope(); + env.encryption.ciphertext = "A".repeat(MAX_LOCKED_CIPHERTEXT_BYTES + 1); + assert!(validate_envelope(&env) + .unwrap_err() + .contains("maximum size")); + + // Oversized envelope JSON is rejected before typed deserialization. + let huge = format!( + r#"{{"format":"{LOCKED_FORMAT}","version":1,"pad":"{}","encryption":{{}}}}"#, + "p".repeat(MAX_LOCKED_ENVELOPE_JSON_BYTES) + ); + let err = parse_chunk_payload(huge.as_bytes()).unwrap_err(); + assert!(err.contains("maximum size"), "{err}"); + } + + #[test] + fn locked_png_round_trips_through_chunk_and_decrypt() { + let (owner, agent) = owner_agent_keys(); + let snapshot = sample_snapshot(); + let png = encode_locked_snapshot_png(&snapshot, &owner, &agent.public_key(), None).unwrap(); + + let payload = extract_chunk_payload_png(&png).unwrap(); + let ChunkPayload::Locked(env) = parse_chunk_payload(&payload).unwrap() else { + panic!("locked PNG must parse as a locked envelope"); + }; + // Both endpoints decrypt to the same logical manifest (compare + // manifests, never ciphertext — the NIP-44 nonce is random). + assert_eq!( + decrypt_envelope(&env, owner.secret_key()).unwrap(), + snapshot + ); + assert_eq!( + decrypt_envelope(&env, agent.secret_key()).unwrap(), + snapshot + ); + } + + #[test] + fn plain_manifest_dispatches_to_plain() { + let json = encode_snapshot_json(&sample_snapshot()).unwrap(); + let ChunkPayload::Plain(decoded) = parse_chunk_payload(&json).unwrap() else { + panic!("plain manifest must parse as Plain"); + }; + assert_eq!(*decoded, sample_snapshot()); + } + + #[test] + fn resolve_unlock_secret_owner_exact_endpoint() { + let (env, owner, _agent) = locked_envelope(); + let secret = resolve_unlock_secret(&env, Some(&owner), &[]).unwrap(); + assert_eq!(&secret, owner.secret_key()); + + // A different identity key is NOT tried. + let other = Keys::generate(); + assert!(resolve_unlock_secret(&env, Some(&other), &[]).is_none()); + assert!(resolve_unlock_secret(&env, None, &[]).is_none()); + } + + #[test] + fn resolve_unlock_secret_agent_requires_record_and_derived_match() { + let (env, _owner, agent) = locked_envelope(); + let nsec = nostr::ToBech32::to_bech32(agent.secret_key()).unwrap(); + + let record = record_with_keys(agent.public_key().to_hex(), nsec); + let secret = resolve_unlock_secret(&env, None, std::slice::from_ref(&record)).unwrap(); + assert_eq!(&secret, agent.secret_key()); + + // Record pubkey matches but the stored secret derives a DIFFERENT + // pubkey → refused (no trial decryption on mismatched material). + let mut forged = record.clone(); + forged.private_key_nsec = + nostr::ToBech32::to_bech32(Keys::generate().secret_key()).unwrap(); + assert!(resolve_unlock_secret(&env, None, std::slice::from_ref(&forged)).is_none()); + + // Record for some other agent → not an endpoint. + let mut unrelated = record; + unrelated.pubkey = Keys::generate().public_key().to_hex(); + assert!(resolve_unlock_secret(&env, None, std::slice::from_ref(&unrelated)).is_none()); + } +} diff --git a/desktop/src-tauri/src/managed_agents/agent_snapshot_tests.rs b/desktop/src-tauri/src/managed_agents/agent_snapshot_tests.rs new file mode 100644 index 0000000000..b4492418e5 --- /dev/null +++ b/desktop/src-tauri/src/managed_agents/agent_snapshot_tests.rs @@ -0,0 +1,599 @@ +//! Unit tests for `managed_agents/agent_snapshot.rs`. +//! +//! Kept in a sibling file so `agent_snapshot.rs` stays under the +//! 1000-line gate; `#[path]`-included from there. + +use super::*; +use crate::managed_agents::types::{BackendKind, ManagedAgentRecord, RespondTo}; +use std::collections::BTreeMap; + +/// Build a minimal `ManagedAgentRecord` for testing. Only the fields +/// relevant to snapshot export are filled; the rest use defaults. +fn minimal_record() -> ManagedAgentRecord { + ManagedAgentRecord { + pubkey: "deadbeef".to_string(), + name: "Test Agent".to_string(), + display_name: Some("Test Agent Display".to_string()), + persona_id: Some("SENTINEL_PERSONA_ID".to_string()), // MUST NOT appear in snapshot + team_id: Some("SENTINEL_TEAM_ID".to_string()), // MUST NOT appear in snapshot + private_key_nsec: "nsec1secret".to_string(), // MUST NOT appear in snapshot + auth_tag: Some("auth-tag-secret".to_string()), // MUST NOT appear in snapshot + relay_url: "wss://relay.example.com".to_string(), // MUST NOT appear in snapshot + avatar_url: Some("https://example.com/avatar.png".to_string()), + acp_command: "/usr/local/bin/acp".to_string(), // MUST NOT appear in snapshot + agent_command: "goose".to_string(), // MUST NOT appear in snapshot + agent_command_override: Some("goose-override".to_string()), // MUST NOT appear + agent_args: vec!["--arg".to_string()], // MUST NOT appear in snapshot + mcp_command: "mcp-server".to_string(), // MUST NOT appear in snapshot + turn_timeout_seconds: 120, // deprecated, MUST NOT appear + idle_timeout_seconds: Some(30), + max_turn_duration_seconds: Some(600), + parallelism: 2, + system_prompt: Some("You are a test agent.".to_string()), + model: Some("claude-opus-4".to_string()), + provider: Some("anthropic".to_string()), + persona_source_version: Some("v1.0".to_string()), // MUST NOT appear + env_vars: { + let mut m = BTreeMap::new(); + m.insert("API_KEY".to_string(), "secret123".to_string()); // MUST NOT appear + m + }, + start_on_app_launch: true, + auto_restart_on_config_change: true, + runtime_pid: Some(12345), // MUST NOT appear + backend: BackendKind::Provider { + // MUST NOT appear — carries a provider secret + id: "SENTINEL_BACKEND_ID".to_string(), + config: serde_json::json!({"api_key": "SENTINEL_BACKEND_SECRET"}), + }, + backend_agent_id: Some("SENTINEL_BACKEND_AGENT_ID".to_string()), // MUST NOT appear + provider_binary_path: Some("/usr/bin/SENTINEL_PROVIDER_BINARY".to_string()), // MUST NOT appear + persona_team_dir: Some(std::path::PathBuf::from("SENTINEL_TEAM_DIR")), // MUST NOT appear + persona_name_in_team: Some("SENTINEL_NAME_IN_TEAM".to_string()), // MUST NOT appear + created_at: "2024-01-01T00:00:00Z".to_string(), + updated_at: "2024-01-02T00:00:00Z".to_string(), + last_started_at: Some("2024-01-03T00:00:00Z".to_string()), // MUST NOT appear + last_stopped_at: None, + last_exit_code: Some(0), // MUST NOT appear + last_error: Some("SENTINEL_LAST_ERROR".to_string()), // MUST NOT appear + last_error_code: Some(42), // MUST NOT appear + respond_to: RespondTo::default(), + respond_to_allowlist: vec!["pubkey1hex".to_string()], + slug: Some("test-agent".to_string()), + runtime: Some("goose".to_string()), + name_pool: vec!["Alice".to_string(), "Bob".to_string()], + is_builtin: false, + is_active: true, + shared: false, + source_team: Some("team-id-123".to_string()), // MUST NOT appear + source_team_persona_slug: Some("lep".to_string()), // MUST NOT appear + definition_respond_to: Some("allowlist".to_string()), + catalog_source: None, + definition_respond_to_allowlist: vec!["abc123def".to_string()], + definition_parallelism: Some(4), + relay_mesh: None, + } +} + +// ── Round-trip tests ────────────────────────────────────────────────────── + +#[test] +fn json_round_trip_config_only() { + let record = minimal_record(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); + let bytes = encode_snapshot_json(&snapshot).unwrap(); + let parsed = decode_snapshot_json(&bytes).unwrap(); + assert_eq!(parsed, snapshot); +} + +#[test] +fn json_round_trip_with_memory() { + let record = minimal_record(); + let entries = vec![ + AgentSnapshotMemoryEntry { + slug: "core".to_string(), + body: "I am a test agent.".to_string(), + }, + AgentSnapshotMemoryEntry { + slug: "mem/research".to_string(), + body: "Some research notes.".to_string(), + }, + ]; + let snapshot = build_snapshot(&record, MemoryLevel::Everything, entries, None); + let bytes = encode_snapshot_json(&snapshot).unwrap(); + let parsed = decode_snapshot_json(&bytes).unwrap(); + assert_eq!(parsed, snapshot); +} + +#[test] +fn png_round_trip_no_memory() { + let record = minimal_record(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); + let png_bytes = encode_snapshot_png(&snapshot, None).unwrap(); + let parsed = decode_snapshot_png(&png_bytes).unwrap(); + assert_eq!(parsed.definition.name, snapshot.definition.name); + assert_eq!(parsed.profile.display_name, snapshot.profile.display_name); + assert_eq!(parsed.memory.level, MemoryLevel::None); +} + +#[test] +fn png_round_trip_with_avatar_png() { + // Build a minimal PNG avatar. + let avatar = make_png_with_text("dummy", "value").unwrap(); + let record = minimal_record(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&avatar)); + // Avatar should be inlined as a data URL. + assert!(snapshot + .profile + .avatar_data_url + .as_deref() + .unwrap_or("") + .starts_with("data:image/png;base64,")); + + let png_bytes = encode_snapshot_png(&snapshot, Some(&avatar)).unwrap(); + let parsed = decode_snapshot_png(&png_bytes).unwrap(); + assert_eq!(parsed.definition.name, snapshot.definition.name); +} + +/// Plain-card byte compatibility: `encode_snapshot_png` was refactored +/// through the shared `encode_chunk_payload_png` when locked cards were +/// added. Plain cards must emit byte-identical PNGs to the pre-envelope +/// encoder. This vector reimplements the legacy encoder body verbatim and +/// asserts equality on all three composition paths: placeholder (no avatar), +/// PNG-avatar (where tEXt chunk injection ordering matters), and +/// JPEG-avatar transcode. +#[test] +fn plain_encoder_bytes_identical_to_pre_envelope_encoder() { + // Verbatim pre-refactor `encode_snapshot_png` body (post memory guard). + fn legacy_encode( + snapshot: &AgentSnapshot, + avatar_bytes: Option<&[u8]>, + ) -> Result, String> { + let json_bytes = encode_snapshot_json(snapshot)?; + let chunk_text = STANDARD.encode(&json_bytes); + let png_bytes = match avatar_bytes.filter(|bytes| !bytes.is_empty()) { + Some(bytes) => { + let encoded_avatar = if bytes.starts_with(b"\x89PNG") { + inject_text_chunk(bytes, PNG_CHUNK_KEYWORD, &chunk_text).or_else(|_| { + transcode_avatar_to_png_with_text(bytes, PNG_CHUNK_KEYWORD, &chunk_text) + }) + } else { + transcode_avatar_to_png_with_text(bytes, PNG_CHUNK_KEYWORD, &chunk_text) + }; + match encoded_avatar { + Ok(png_bytes) => png_bytes, + Err(_) => make_png_with_text(PNG_CHUNK_KEYWORD, &chunk_text)?, + } + } + None => make_png_with_text(PNG_CHUNK_KEYWORD, &chunk_text)?, + }; + Ok(png_bytes) + } + + let record = minimal_record(); + + // Placeholder path (no avatar). + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); + assert_eq!( + encode_snapshot_png(&snapshot, None).unwrap(), + legacy_encode(&snapshot, None).unwrap(), + "placeholder-path plain PNG bytes must match the pre-envelope encoder" + ); + + // PNG-avatar path: chunk injected into the avatar image body. + let avatar = make_png_with_text("dummy", "value").unwrap(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&avatar)); + assert_eq!( + encode_snapshot_png(&snapshot, Some(&avatar)).unwrap(), + legacy_encode(&snapshot, Some(&avatar)).unwrap(), + "avatar-path plain PNG bytes must match the pre-envelope encoder" + ); + + // JPEG-avatar path: transcode-to-PNG composition. + let jpeg_avatar = image::DynamicImage::ImageRgb8(image::RgbImage::from_pixel( + 4, + 4, + image::Rgb([0x10, 0x20, 0x30]), + )); + let mut jpeg_bytes = Vec::new(); + jpeg_avatar + .write_to(&mut Cursor::new(&mut jpeg_bytes), image::ImageFormat::Jpeg) + .unwrap(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&jpeg_bytes)); + assert_eq!( + encode_snapshot_png(&snapshot, Some(&jpeg_bytes)).unwrap(), + legacy_encode(&snapshot, Some(&jpeg_bytes)).unwrap(), + "transcode-path plain PNG bytes must match the pre-envelope encoder" + ); +} + +#[test] +fn png_snapshot_transcodes_jpeg_avatar_into_image_body() { + let avatar = image::DynamicImage::ImageRgb8(image::RgbImage::from_pixel( + 3, + 2, + image::Rgb([0x12, 0x34, 0x56]), + )); + let mut jpeg_bytes = Vec::new(); + avatar + .write_to(&mut Cursor::new(&mut jpeg_bytes), image::ImageFormat::Jpeg) + .unwrap(); + + let snapshot = build_snapshot( + &minimal_record(), + MemoryLevel::None, + vec![], + Some(&jpeg_bytes), + ); + let png_bytes = encode_snapshot_png(&snapshot, Some(&jpeg_bytes)).unwrap(); + let decoder = Decoder::new(Cursor::new(png_bytes)); + let reader = decoder.read_info().unwrap(); + + assert_eq!((reader.info().width, reader.info().height), (3, 2)); +} + +// ── PNG memory parity ───────────────────────────────────────────────────── + +#[test] +fn png_round_trip_with_core_memory() { + let record = minimal_record(); + let entries = vec![AgentSnapshotMemoryEntry { + slug: "core".to_string(), + body: "remember this".to_string(), + }]; + let snapshot = build_snapshot(&record, MemoryLevel::Core, entries, None); + + let png_bytes = encode_snapshot_png(&snapshot, None).unwrap(); + let parsed = decode_snapshot_png(&png_bytes).unwrap(); + + assert_eq!(parsed.memory, snapshot.memory); +} + +#[test] +fn png_round_trip_with_everything_memory() { + let record = minimal_record(); + let entries = vec![ + AgentSnapshotMemoryEntry { + slug: "core".to_string(), + body: "remember this".to_string(), + }, + AgentSnapshotMemoryEntry { + slug: "mem/notes".to_string(), + body: "private notes".to_string(), + }, + ]; + let snapshot = build_snapshot(&record, MemoryLevel::Everything, entries, None); + + let png_bytes = encode_snapshot_png(&snapshot, None).unwrap(); + let parsed = decode_snapshot_png(&png_bytes).unwrap(); + + assert_eq!(parsed.memory, snapshot.memory); +} + +#[test] +fn png_export_with_no_memory_succeeds() { + let record = minimal_record(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); + assert!(encode_snapshot_png(&snapshot, None).is_ok()); +} + +#[test] +fn png_export_rejects_none_level_with_nonempty_entries() { + // Inconsistent state: level == None but entries is non-empty. + // The encoder must reject this to prevent a memory-leak bypass. + let record = minimal_record(); + let entries = vec![AgentSnapshotMemoryEntry { + slug: "core".to_string(), + body: "leaked memory".to_string(), + }]; + // Build with entries, then override level to None in the struct. + let mut snapshot = build_snapshot(&record, MemoryLevel::Core, entries, None); + snapshot.memory.level = MemoryLevel::None; // force inconsistency + let result = encode_snapshot_png(&snapshot, None); + assert!( + result.is_err(), + "PNG encoder must reject level=None with non-empty entries" + ); + assert!( + result + .unwrap_err() + .contains("memory.level 'none' and non-empty memory entries"), + "Error must explain the malformed memory state" + ); +} + +// ── Secret exclusion tests ──────────────────────────────────────────────── +// +// These tests assert that every field in the exclusion list is absent from +// the serialized snapshot. We serialize to JSON and assert the key is NOT +// present. + +fn snapshot_json_string(record: &ManagedAgentRecord) -> String { + let snapshot = build_snapshot(record, MemoryLevel::None, vec![], None); + let bytes = encode_snapshot_json(&snapshot).unwrap(); + String::from_utf8(bytes).unwrap() +} + +#[test] +fn secret_exclusion_private_key_nsec_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("nsec1secret"), + "nsec must not appear in snapshot" + ); + assert!( + !json.contains("privateKeyNsec") && !json.contains("private_key_nsec"), + "privateKeyNsec field must not appear in snapshot" + ); +} + +#[test] +fn secret_exclusion_auth_tag_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("auth-tag-secret"), + "auth_tag value must not appear in snapshot" + ); + assert!( + !json.contains("authTag") && !json.contains("auth_tag"), + "authTag field must not appear in snapshot" + ); +} + +#[test] +fn secret_exclusion_env_vars_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("API_KEY") && !json.contains("secret123"), + "env_vars content must not appear in snapshot" + ); + assert!( + !json.contains("envVars") && !json.contains("env_vars"), + "envVars field must not appear in snapshot" + ); +} + +#[test] +fn secret_exclusion_relay_url_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("wss://relay.example.com"), + "relay_url value must not appear in snapshot" + ); + assert!( + !json.contains("relayUrl") && !json.contains("relay_url"), + "relayUrl field must not appear in snapshot" + ); +} + +#[test] +fn snapshot_omits_removed_mcp_toolsets_config() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("mcpToolsets") && !json.contains("mcp_toolsets"), + "removed MCP toolsets config must not re-enter snapshots" + ); +} + +#[test] +fn secret_exclusion_machine_commands_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + // acp_command / agent_command / agent_command_override / agent_args / mcp_command + assert!( + !json.contains("/usr/local/bin/acp"), + "acp_command path must not appear" + ); + assert!( + !json.contains("acpCommand") && !json.contains("acp_command"), + "acpCommand field must not appear" + ); + assert!( + !json.contains("agentCommand") && !json.contains("agent_command"), + "agentCommand field must not appear" + ); + assert!( + !json.contains("mcpCommand") && !json.contains("mcp_command"), + "mcpCommand field must not appear" + ); +} + +#[test] +fn secret_exclusion_runtime_state_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("runtimePid") && !json.contains("runtime_pid"), + "runtimePid must not appear" + ); + assert!( + !json.contains("backendAgentId") && !json.contains("backend_agent_id"), + "backendAgentId must not appear" + ); + assert!( + !json.contains("SENTINEL_BACKEND_AGENT_ID"), + "backendAgentId value must not appear" + ); + assert!( + !json.contains("providerBinaryPath") && !json.contains("provider_binary_path"), + "providerBinaryPath must not appear" + ); + assert!( + !json.contains("SENTINEL_PROVIDER_BINARY"), + "providerBinaryPath value must not appear" + ); + assert!( + !json.contains("lastStartedAt") && !json.contains("last_started_at"), + "lastStartedAt must not appear" + ); + assert!( + !json.contains("lastExitCode") && !json.contains("last_exit_code"), + "lastExitCode must not appear" + ); + // backend blob — neither the type tag nor provider secret must leak. + assert!( + !json.contains("\"backend\"") && !json.contains("backend"), + "backend field must not appear" + ); + assert!( + !json.contains("SENTINEL_BACKEND_ID") && !json.contains("SENTINEL_BACKEND_SECRET"), + "backend config values must not appear" + ); + // last_error / last_error_code + assert!( + !json.contains("lastError") && !json.contains("last_error"), + "lastError must not appear" + ); + assert!( + !json.contains("SENTINEL_LAST_ERROR"), + "lastError value must not appear" + ); + assert!( + !json.contains("lastErrorCode") && !json.contains("last_error_code"), + "lastErrorCode must not appear" + ); +} + +#[test] +fn secret_exclusion_lineage_ids_absent() { + let record = minimal_record(); + let json = snapshot_json_string(&record); + assert!( + !json.contains("team-id-123"), + "source_team value must not appear" + ); + assert!( + !json.contains("sourceTeam") && !json.contains("source_team"), + "sourceTeam field must not appear" + ); + assert!( + !json.contains("sourceTeamPersonaSlug"), + "sourceTeamPersonaSlug must not appear" + ); + assert!( + !json.contains("personaSourceVersion") && !json.contains("persona_source_version"), + "personaSourceVersion must not appear" + ); + // personaId + assert!( + !json.contains("personaId") && !json.contains("persona_id"), + "personaId field must not appear" + ); + assert!( + !json.contains("SENTINEL_PERSONA_ID"), + "personaId value must not appear" + ); + // teamId + assert!( + !json.contains("teamId") && !json.contains("team_id"), + "teamId field must not appear" + ); + assert!( + !json.contains("SENTINEL_TEAM_ID"), + "teamId value must not appear" + ); + // personaTeamDir + assert!( + !json.contains("personaTeamDir") && !json.contains("persona_team_dir"), + "personaTeamDir field must not appear" + ); + assert!( + !json.contains("SENTINEL_TEAM_DIR"), + "personaTeamDir value must not appear" + ); + // personaNameInTeam + assert!( + !json.contains("personaNameInTeam") && !json.contains("persona_name_in_team"), + "personaNameInTeam field must not appear" + ); + assert!( + !json.contains("SENTINEL_NAME_IN_TEAM"), + "personaNameInTeam value must not appear" + ); +} + +// ── Definition field presence tests ────────────────────────────────────── + +#[test] +fn definition_fields_present_in_snapshot() { + let record = minimal_record(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); + + assert_eq!(snapshot.definition.name, "Test Agent Display"); + assert!(!snapshot.definition.source_is_builtin); + assert_eq!( + snapshot.definition.system_prompt.as_deref(), + Some("You are a test agent.") + ); + assert_eq!(snapshot.definition.runtime.as_deref(), Some("goose")); + assert_eq!(snapshot.definition.model.as_deref(), Some("claude-opus-4")); + assert_eq!(snapshot.definition.provider.as_deref(), Some("anthropic")); + assert_eq!(snapshot.definition.name_pool, vec!["Alice", "Bob"]); + // definition_respond_to maps to respond_to in the snapshot definition + assert_eq!(snapshot.definition.respond_to.as_deref(), Some("allowlist")); + // definition_respond_to_allowlist should be included + assert!(!snapshot.definition.respond_to_allowlist.is_empty()); +} + +#[test] +fn profile_fields_present_in_snapshot() { + let record = minimal_record(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], None); + assert_eq!(snapshot.profile.display_name, "Test Agent Display"); + // No bytes → should fall back to avatar_url + assert_eq!( + snapshot.profile.avatar_url.as_deref(), + Some("https://example.com/avatar.png") + ); + assert!(snapshot.profile.avatar_data_url.is_none()); +} + +#[test] +fn avatar_inlined_when_under_size_limit() { + let record = minimal_record(); + let small_png = make_png_with_text("k", "v").unwrap(); + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&small_png)); + assert!(snapshot.profile.avatar_data_url.is_some()); + assert!(snapshot.profile.avatar_url.is_none()); +} + +#[test] +fn avatar_url_fallback_when_over_size_limit() { + let mut record = minimal_record(); + record.avatar_url = Some("https://example.com/big.png".to_string()); + // Synthesize oversized avatar bytes (> 2 MB) — just a large zeroed vec. + let big_bytes = vec![0u8; MAX_AVATAR_INLINE_BYTES + 1]; + let snapshot = build_snapshot(&record, MemoryLevel::None, vec![], Some(&big_bytes)); + assert!(snapshot.profile.avatar_data_url.is_none()); + assert_eq!( + snapshot.profile.avatar_url.as_deref(), + Some("https://example.com/big.png") + ); +} + +// ── Format/version validation ───────────────────────────────────────────── + +#[test] +fn invalid_format_discriminator_is_rejected() { + let mut snapshot = build_snapshot(&minimal_record(), MemoryLevel::None, vec![], None); + snapshot.format = "not-a-buzz-snapshot".to_string(); + let bytes = serde_json::to_vec(&snapshot).unwrap(); + let result = decode_snapshot_json(&bytes); + assert!(result.is_err()); + assert!(result.unwrap_err().contains("Unsupported snapshot format")); +} + +#[test] +fn unsupported_version_is_rejected() { + let mut snapshot = build_snapshot(&minimal_record(), MemoryLevel::None, vec![], None); + snapshot.version = 99; + let bytes = serde_json::to_vec(&snapshot).unwrap(); + let result = decode_snapshot_json(&bytes); + assert!(result.is_err()); + assert!(result.unwrap_err().contains("Unsupported snapshot version")); +} diff --git a/desktop/src-tauri/src/managed_agents/backend.rs b/desktop/src-tauri/src/managed_agents/backend.rs index 5debae41cb..84dd7e99da 100644 --- a/desktop/src-tauri/src/managed_agents/backend.rs +++ b/desktop/src-tauri/src/managed_agents/backend.rs @@ -1,3 +1,4 @@ +use sha2::{Digest, Sha256}; use std::io::{BufReader, Read, Write}; use std::path::{Path, PathBuf}; use std::sync::mpsc; @@ -7,6 +8,61 @@ const STDERR_CAP: usize = 65536; /// Provider responses should be small JSON objects. Cap stdout to prevent a /// buggy or malicious provider from OOM-ing the desktop process. const STDOUT_CAP: usize = 1_048_576; // 1 MB +const PROVIDER_PROTOCOL_VERSION: u64 = 1; + +fn validate_provider_info(info: &serde_json::Value) -> Result<(), String> { + let object = info + .as_object() + .ok_or_else(|| "provider info response must be a JSON object".to_string())?; + let actual_version = object + .get("protocol_version") + .and_then(serde_json::Value::as_u64); + if actual_version != Some(PROVIDER_PROTOCOL_VERSION) { + return Err(match actual_version { + Some(version) => format!( + "unsupported provider protocol version {version}; desktop requires {PROVIDER_PROTOCOL_VERSION}" + ), + None => "provider info response missing integer protocol_version".to_string(), + }); + } + if object.get("ok") != Some(&serde_json::Value::Bool(true)) { + return Err("provider info response must contain ok: true".to_string()); + } + for field in ["name", "version", "description"] { + if object + .get(field) + .is_none_or(|value| value.as_str().is_none_or(str::is_empty)) + { + return Err(format!( + "provider info response missing non-empty string {field}" + )); + } + } + if !object + .get("config_schema") + .is_some_and(serde_json::Value::is_object) + { + return Err("provider info response missing object config_schema".to_string()); + } + + const FIELDS: &[&str] = &[ + "ok", + "name", + "version", + "protocol_version", + "description", + "config_schema", + ]; + if let Some(field) = object + .keys() + .find(|field| !FIELDS.contains(&field.as_str())) + { + return Err(format!( + "provider info response contains unknown field {field}" + )); + } + Ok(()) +} /// Invoke a provider binary: write JSON to stdin, read JSON from stdout. /// @@ -333,23 +389,29 @@ pub(crate) fn redact_secrets_with(s: &str, extras: &[&str]) -> String { result } -/// Collect string values from `request["agent"]["env_vars"]` (if present) -/// to feed into [`redact_secrets_with`]. Returns an empty Vec if the -/// request shape doesn't match, which is fine — falls back to the default -/// prefix-based scrubbing. +/// Collect string values from every environment map a deploy request can +/// carry. Providers may echo any of these values in diagnostics, including +/// definition/baked values that exist only in the resolved launch block. fn env_secrets_from_request(request: &serde_json::Value) -> Vec { - request - .get("agent") - .and_then(|a| a.get("env_vars")) - .and_then(|e| e.as_object()) - .map(|obj| { - obj.values() - .filter_map(|v| v.as_str()) - .filter(|s| !s.is_empty()) - .map(String::from) - .collect() - }) - .unwrap_or_default() + let agent = request.get("agent"); + let maps = [ + agent.and_then(|value| value.get("env_vars")), + agent + .and_then(|value| value.get("launch")) + .and_then(|value| value.get("env")), + agent + .and_then(|value| value.get("launch")) + .and_then(|value| value.get("policy_env")), + ]; + + maps.into_iter() + .flatten() + .filter_map(serde_json::Value::as_object) + .flat_map(|map| map.values()) + .filter_map(serde_json::Value::as_str) + .filter(|value| !value.is_empty()) + .map(String::from) + .collect() } /// Public-in-crate helper: redact every non-empty value from `env` (plus @@ -368,22 +430,102 @@ pub(crate) fn redact_env_values_in( redact_secrets_with(s, &values) } -/// Deploy an agent via provider binary. Returns the provider-assigned agent_id. -/// -/// `request_id` is included for provider-side logging/correlation but is not -/// validated in the response — the stdin→stdout exchange is 1:1 per process. +/// Copy a resolved provider into a private staging directory while hashing +/// exactly the bytes copied. The staged file becomes non-writable before either +/// invocation, closing the path replacement and in-place rewrite races. +fn stage_provider( + binary: &Path, +) -> Result<(tempfile::TempDir, PathBuf, String, std::fs::File), String> { + let directory = tempfile::Builder::new() + .prefix("buzz-provider-") + .tempdir() + .map_err(|error| format!("failed to create provider staging directory: {error}"))?; + let suffix = if cfg!(windows) { ".exe" } else { "" }; + let staged_path = directory.path().join(format!("provider{suffix}")); + let mut source = std::fs::File::open(binary) + .map_err(|error| format!("failed to open provider for staging: {error}"))?; + let mut staged = std::fs::OpenOptions::new() + .write(true) + .create_new(true) + .open(&staged_path) + .map_err(|error| format!("failed to create staged provider: {error}"))?; + let mut hasher = Sha256::new(); + let mut buffer = [0_u8; 64 * 1024]; + loop { + let count = source + .read(&mut buffer) + .map_err(|error| format!("failed to read provider for staging: {error}"))?; + if count == 0 { + break; + } + staged + .write_all(&buffer[..count]) + .map_err(|error| format!("failed to write staged provider: {error}"))?; + hasher.update(&buffer[..count]); + } + staged + .sync_all() + .map_err(|error| format!("failed to sync staged provider: {error}"))?; + + let mut permissions = staged + .metadata() + .map_err(|error| format!("failed to inspect staged provider: {error}"))? + .permissions(); + #[cfg(unix)] + { + use std::os::unix::fs::PermissionsExt; + permissions.set_mode(0o500); + } + #[cfg(not(unix))] + permissions.set_readonly(true); + std::fs::set_permissions(&staged_path, permissions) + .map_err(|error| format!("failed to protect staged provider: {error}"))?; + drop(staged); + + #[cfg(windows)] + let execution_guard = { + use std::os::windows::fs::OpenOptionsExt; + std::fs::OpenOptions::new() + .read(true) + // Permit CreateProcess to read the image while denying replacement, + // writes, and deletion until both invocations finish. + .share_mode(windows_sys::Win32::Storage::FileSystem::FILE_SHARE_READ) + .open(&staged_path) + }; + #[cfg(not(windows))] + let execution_guard = std::fs::File::open(&staged_path); + let execution_guard = execution_guard + .map_err(|error| format!("failed to lock staged provider for execution: {error}"))?; + Ok(( + directory, + staged_path, + hex::encode(hasher.finalize()), + execution_guard, + )) +} + +/// Deploy through one immutable staged copy: negotiate protocol v1 before the +/// secret-bearing request, then invoke deploy on those exact same bytes. pub fn provider_deploy( binary: &Path, agent: &serde_json::Value, provider_config: &serde_json::Value, ) -> Result { + let (_directory, staged, _digest, _execution_guard) = stage_provider(binary)?; + let info_request = serde_json::json!({ + "op": "info", + "request_id": uuid::Uuid::new_v4().to_string(), + }); + let info = invoke_provider(&staged, &info_request, Duration::from_secs(10))?; + validate_provider_info(&info)?; + let request = serde_json::json!({ "op": "deploy", "request_id": uuid::Uuid::new_v4().to_string(), "agent": agent, "provider_config": provider_config, }); - let resp = invoke_provider(binary, &request, Duration::from_secs(600))?; + let resp = invoke_provider(&staged, &request, Duration::from_secs(600))?; resp["agent_id"] .as_str() .map(String::from) @@ -423,6 +565,24 @@ pub fn validate_provider_config(config: &serde_json::Value) -> Result<(), String Ok(()) } +/// Derive a provider id from the filename Tauri stages at runtime. Tauri +/// removes its target-triple suffix while copying an external binary, but on +/// Windows leaves the executable/script extension, which is not part of the +/// provider id. +fn provider_id_from_filename(name: &str) -> Option<&str> { + let raw = name.strip_prefix("buzz-backend-")?; + let id = [".exe", ".bat", ".cmd"] + .into_iter() + .find_map(|extension| { + raw.get(raw.len().saturating_sub(extension.len())..) + .filter(|suffix| suffix.eq_ignore_ascii_case(extension)) + .map(|_| &raw[..raw.len() - extension.len()]) + }) + .unwrap_or(raw); + + (!id.is_empty()).then_some(id) +} + /// Enumerate PATH for buzz-backend-* executables. Returns (id, path) pairs. /// Only includes files that are executable. Does NOT execute any binaries. /// @@ -464,10 +624,12 @@ pub fn discover_provider_candidates() -> Vec<(String, PathBuf)> { }; for entry in entries.flatten() { let name = entry.file_name().to_string_lossy().to_string(); - if let Some(id) = name.strip_prefix(prefix) { - if !id.is_empty() && !seen.contains(&name) && is_executable(&entry.path()) { - seen.insert(name.clone()); - results.push((id.to_string(), entry.path())); + if name.starts_with(prefix) { + if let Some(id) = provider_id_from_filename(&name) { + if !seen.contains(&name) && is_executable(&entry.path()) { + seen.insert(name.clone()); + results.push((id.to_string(), entry.path())); + } } } } @@ -538,203 +700,5 @@ pub struct BackendProviderInfo { } #[cfg(test)] -mod tests { - use super::*; - - #[test] - fn redact_secrets_replaces_nsec() { - let s = "key=nsec1abc123def456 other"; - let r = redact_secrets(s); - assert!(r.contains("[REDACTED]")); - assert!(!r.contains("nsec1abc123def456")); - } - - #[test] - fn redact_secrets_replaces_token() { - let s = r#"{"token":"sprt_tok_xyz789"}"#; - let r = redact_secrets(s); - assert!(r.contains("[REDACTED]")); - assert!(!r.contains("sprt_tok_xyz789")); - } - - #[test] - fn redact_secrets_with_extras_scrubs_user_env_values() { - // If a provider echoes back a user-supplied API key in its error - // output, the desktop must not surface that secret unredacted via - // `last_error`. We scrub the literal values that came from the - // request's `agent.env_vars`. - let secret = "sk-ant-api03-abc123def456"; - let stderr = format!("auth failed with key {secret} on host api.anthropic.com"); - let r = redact_secrets_with(&stderr, &[secret]); - assert!(r.contains("[REDACTED]")); - assert!(!r.contains(secret)); - } - - #[test] - fn redact_secrets_with_extras_skips_short_values() { - // Don't scrub values shorter than 4 chars — too noisy. - let r = redact_secrets_with("error code: 42", &["42"]); - assert!(r.contains("42")); - } - - /// GitHub tokens are recognised by shape, so one that never passed through - /// our environment — embedded in a remote URL an installer echoes — is - /// still scrubbed. The scan runs to the next whitespace or quote, so the - /// rest of the URL goes with it; over-redaction is the safe direction. - #[test] - fn redact_secrets_with_scrubs_github_token_prefixes() { - for token in [ - "ghp_abcdefghij0123456789", - "gho_abcdefghij0123456789", - "ghu_abcdefghij0123456789", - "ghs_abcdefghij0123456789", - "ghr_abcdefghij0123456789", - "github_pat_abcdefghij0123456789", - ] { - let r = - redact_secrets_with(&format!("cloning https://{token}@github.com/o/r now"), &[]); - assert!(!r.contains(token), "leaked {token}: {r}"); - assert!(r.contains("[REDACTED]"), "got: {r}"); - assert!(r.contains("cloning"), "scan must stop at whitespace: {r}"); - assert!(r.ends_with(" now"), "scan must stop at whitespace: {r}"); - } - } - - #[test] - fn redact_secrets_with_extras_terminates_when_value_substring_of_marker() { - // Regression: an earlier impl used `while let Some(pos) = find(value)` - // which never terminates if the user's env value is a substring of - // the replacement marker `[REDACTED]` — each replacement - // reintroduces the same text. Now uses `str::replace` (single-pass). - for value in ["REDACTED", "EDACTE", "REDA", "ACTED"] { - let r = redact_secrets_with(&format!("leak={value}"), &[value]); - assert!(r.contains("[REDACTED]")); - } - } - - #[test] - fn redact_secrets_with_extras_handles_overlapping_secrets() { - // Longer entries get scrubbed first so the substring "abc12" isn't - // matched before "abc123" is consumed. - let s = "key1=abc123 key2=abc12"; - let r = redact_secrets_with(s, &["abc12", "abc123"]); - assert!(!r.contains("abc123")); - assert!(!r.contains("abc12 ")); - } - - #[test] - fn env_secrets_from_request_extracts_string_values() { - let req = serde_json::json!({ - "op": "deploy", - "agent": { - "env_vars": { - "ANTHROPIC_API_KEY": "sk-ant-test", - "EMPTY": "", - "NUMERIC": 42, - }, - }, - }); - let secrets = env_secrets_from_request(&req); - assert!(secrets.iter().any(|v| v == "sk-ant-test")); - // Empty and non-string values are filtered out. - assert_eq!(secrets.len(), 1); - } - - #[test] - fn env_secrets_from_request_handles_missing_shape() { - assert!(env_secrets_from_request(&serde_json::json!({})).is_empty()); - assert!(env_secrets_from_request(&serde_json::json!({"agent": {}})).is_empty()); - assert!( - env_secrets_from_request(&serde_json::json!({"agent": {"env_vars": null}})).is_empty() - ); - } - - #[test] - fn redact_env_values_in_scrubs_map_values() { - let mut env = std::collections::BTreeMap::new(); - env.insert("ANTHROPIC_API_KEY".to_string(), "sk-ant-real".to_string()); - env.insert("EMPTY".to_string(), String::new()); - let stderr = "auth=sk-ant-real failed; other context"; - let r = redact_env_values_in(stderr, &env); - assert!(!r.contains("sk-ant-real")); - assert!(r.contains("[REDACTED]")); - } - - #[test] - fn validate_provider_config_rejects_secret_key() { - let cfg = serde_json::json!({"api_key": "val"}); - assert!(validate_provider_config(&cfg).is_err()); - } - - #[test] - fn validate_provider_config_rejects_nested() { - let cfg = serde_json::json!({"region": {"us": "east"}}); - assert!(validate_provider_config(&cfg).is_err()); - } - - #[test] - fn validate_provider_config_accepts_scalars() { - let cfg = serde_json::json!({"region": "us-east-1", "tier": "standard"}); - assert!(validate_provider_config(&cfg).is_ok()); - } - - #[test] - fn validate_provider_config_allows_key_as_substring() { - // "keyboard", "monkey" contain "key" as substring but not as a word segment. - let cfg = serde_json::json!({"keyboard_layout": "us", "monkey_wrench": "tight"}); - assert!(validate_provider_config(&cfg).is_ok()); - } - - #[test] - fn validate_provider_config_rejects_camel_case_secrets() { - assert!(validate_provider_config(&serde_json::json!({"apiKey": "val"})).is_err()); - assert!(validate_provider_config(&serde_json::json!({"accessToken": "val"})).is_err()); - assert!(validate_provider_config(&serde_json::json!({"clientSecret": "val"})).is_err()); - // ALL-CAPS variants - assert!(validate_provider_config(&serde_json::json!({"apiKEY": "val"})).is_err()); - assert!(validate_provider_config(&serde_json::json!({"accessTOKEN": "val"})).is_err()); - } - - #[test] - fn split_config_key_handles_all_styles() { - assert_eq!(split_config_key("apiKey"), vec!["api", "key"]); - assert_eq!(split_config_key("access_token"), vec!["access", "token"]); - assert_eq!(split_config_key("keyboard"), vec!["keyboard"]); - assert_eq!(split_config_key("client-secret"), vec!["client", "secret"]); - // Acronym runs stay together - assert_eq!(split_config_key("APIKey"), vec!["api", "key"]); - assert_eq!(split_config_key("apiKEY"), vec!["api", "key"]); - assert_eq!(split_config_key("accessTOKEN"), vec!["access", "token"]); - assert_eq!(split_config_key("MyAPIKey"), vec!["my", "api", "key"]); - } - - #[test] - fn resolve_provider_binary_rejects_invalid_ids() { - // Path traversal - assert!(resolve_provider_binary("../evil").is_err()); - // Empty - assert!(resolve_provider_binary("").is_err()); - // Uppercase - assert!(resolve_provider_binary("MyProvider").is_err()); - // Spaces - assert!(resolve_provider_binary("my provider").is_err()); - // Shell metacharacters - assert!(resolve_provider_binary("foo;rm -rf /").is_err()); - // Valid format but not on PATH — should fail with "not found" - assert!(resolve_provider_binary("nonexistent-test-id-12345").is_err()); - } - - #[test] - fn resolve_provider_binary_accepts_valid_id_format() { - // Valid ID format should pass validation. If the binary happens to - // exist on PATH, Ok is returned; otherwise Err contains "not found" - // (not "invalid provider ID"). Either outcome proves validation passed. - match resolve_provider_binary("zzz-nonexistent-test-provider") { - Ok(_) => {} // unlikely but fine — binary exists - Err(e) => assert!( - e.contains("not found"), - "expected 'not found' error, got: {e}" - ), - } - } -} +#[path = "backend_tests.rs"] +mod tests; diff --git a/desktop/src-tauri/src/managed_agents/backend_tests.rs b/desktop/src-tauri/src/managed_agents/backend_tests.rs new file mode 100644 index 0000000000..ce1f81466f --- /dev/null +++ b/desktop/src-tauri/src/managed_agents/backend_tests.rs @@ -0,0 +1,452 @@ +use super::*; + +#[test] +fn redact_secrets_replaces_nsec() { + let s = "key=nsec1abc123def456 other"; + let r = redact_secrets(s); + assert!(r.contains("[REDACTED]")); + assert!(!r.contains("nsec1abc123def456")); +} + +#[test] +fn redact_secrets_replaces_token() { + let s = r#"{"token":"sprt_tok_xyz789"}"#; + let r = redact_secrets(s); + assert!(r.contains("[REDACTED]")); + assert!(!r.contains("sprt_tok_xyz789")); +} + +#[test] +fn redact_secrets_with_extras_scrubs_user_env_values() { + // If a provider echoes back a user-supplied API key in its error + // output, the desktop must not surface that secret unredacted via + // `last_error`. We scrub the literal values that came from the + // request's `agent.env_vars`. + let secret = "sk-ant-api03-abc123def456"; + let stderr = format!("auth failed with key {secret} on host api.anthropic.com"); + let r = redact_secrets_with(&stderr, &[secret]); + assert!(r.contains("[REDACTED]")); + assert!(!r.contains(secret)); +} + +#[test] +fn redact_secrets_with_extras_skips_short_values() { + // Don't scrub values shorter than 4 chars — too noisy. + let r = redact_secrets_with("error code: 42", &["42"]); + assert!(r.contains("42")); +} + +/// GitHub tokens are recognised by shape, so one that never passed through +/// our environment — embedded in a remote URL an installer echoes — is +/// still scrubbed. The scan runs to the next whitespace or quote, so the +/// rest of the URL goes with it; over-redaction is the safe direction. +#[test] +fn redact_secrets_with_scrubs_github_token_prefixes() { + for token in [ + "ghp_abcdefghij0123456789", + "gho_abcdefghij0123456789", + "ghu_abcdefghij0123456789", + "ghs_abcdefghij0123456789", + "ghr_abcdefghij0123456789", + "github_pat_abcdefghij0123456789", + ] { + let r = redact_secrets_with(&format!("cloning https://{token}@github.com/o/r now"), &[]); + assert!(!r.contains(token), "leaked {token}: {r}"); + assert!(r.contains("[REDACTED]"), "got: {r}"); + assert!(r.contains("cloning"), "scan must stop at whitespace: {r}"); + assert!(r.ends_with(" now"), "scan must stop at whitespace: {r}"); + } +} + +#[test] +fn redact_secrets_with_extras_terminates_when_value_substring_of_marker() { + // Regression: an earlier impl used `while let Some(pos) = find(value)` + // which never terminates if the user's env value is a substring of + // the replacement marker `[REDACTED]` — each replacement + // reintroduces the same text. Now uses `str::replace` (single-pass). + for value in ["REDACTED", "EDACTE", "REDA", "ACTED"] { + let r = redact_secrets_with(&format!("leak={value}"), &[value]); + assert!(r.contains("[REDACTED]")); + } +} + +#[test] +fn redact_secrets_with_extras_handles_overlapping_secrets() { + // Longer entries get scrubbed first so the substring "abc12" isn't + // matched before "abc123" is consumed. + let s = "key1=abc123 key2=abc12"; + let r = redact_secrets_with(s, &["abc12", "abc123"]); + assert!(!r.contains("abc123")); + assert!(!r.contains("abc12 ")); +} + +#[test] +fn env_secrets_from_request_extracts_string_values() { + let req = serde_json::json!({ + "op": "deploy", + "agent": { + "env_vars": { + "ANTHROPIC_API_KEY": "sk-ant-test", + "EMPTY": "", + "NUMERIC": 42, + }, + }, + }); + let secrets = env_secrets_from_request(&req); + assert!(secrets.iter().any(|v| v == "sk-ant-test")); + // Empty and non-string values are filtered out. + assert_eq!(secrets.len(), 1); +} + +#[test] +fn env_secrets_from_request_handles_missing_shape() { + assert!(env_secrets_from_request(&serde_json::json!({})).is_empty()); + assert!(env_secrets_from_request(&serde_json::json!({"agent": {}})).is_empty()); + assert!(env_secrets_from_request(&serde_json::json!({"agent": {"env_vars": null}})).is_empty()); +} + +#[test] +fn redact_env_values_in_scrubs_map_values() { + let mut env = std::collections::BTreeMap::new(); + env.insert("ANTHROPIC_API_KEY".to_string(), "sk-ant-real".to_string()); + env.insert("EMPTY".to_string(), String::new()); + let stderr = "auth=sk-ant-real failed; other context"; + let r = redact_env_values_in(stderr, &env); + assert!(!r.contains("sk-ant-real")); + assert!(r.contains("[REDACTED]")); +} + +#[test] +fn env_secrets_from_request_includes_resolved_launch_maps() { + let req = serde_json::json!({ + "agent": { + "env_vars": {"LEGACY": "legacy-secret"}, + "launch": { + "env": {"PERSONA": "persona-secret"}, + "policy_env": {"POLICY": "policy-secret"} + } + } + }); + let secrets = env_secrets_from_request(&req); + assert_eq!(secrets.len(), 3); + for secret in ["legacy-secret", "persona-secret", "policy-secret"] { + assert!(secrets.iter().any(|candidate| candidate == secret)); + } +} + +#[cfg(unix)] +fn write_test_provider(path: &Path, body: &str) { + use std::os::unix::fs::PermissionsExt; + std::fs::write(path, format!("#!/bin/sh\nset -eu\n{body}\n")).unwrap(); + std::fs::set_permissions(path, std::fs::Permissions::from_mode(0o700)).unwrap(); +} + +#[cfg(unix)] +#[test] +fn provider_deploy_negotiates_and_deploys_the_same_staged_bytes() { + let directory = tempfile::tempdir().unwrap(); + let provider = directory.path().join("provider"); + let log = directory.path().join("invocations"); + let body = format!( + r#"read request +printf '%s\n' "$0" >> '{}' +case "$request" in + *\"op\":\"info\"*) printf '%s\n' '{{"ok":true,"name":"test","version":"1.0.0","protocol_version":1,"description":"test provider","config_schema":{{}}}}' ;; + *\"op\":\"deploy\"*) printf '%s\n' '{{"ok":true,"agent_id":"remote-1"}}' ;; +esac"#, + log.display() + ); + write_test_provider(&provider, &body); + + let id = provider_deploy(&provider, &serde_json::json!({}), &serde_json::json!({})) + .expect("staged deploy"); + assert_eq!(id, "remote-1"); + let paths: Vec<_> = std::fs::read_to_string(log) + .unwrap() + .lines() + .map(str::to_owned) + .collect(); + assert_eq!(paths.len(), 2); + assert_eq!(paths[0], paths[1]); + assert_ne!(Path::new(&paths[0]), provider); + assert!( + !Path::new(&paths[0]).exists(), + "staging directory must be deleted" + ); +} + +#[cfg(unix)] +fn replacement_provider() -> &'static str { + r#"#!/bin/sh +set -eu +read request +case "$request" in + *\"op\":\"info\"*) printf '%s\n' '{"ok":true,"name":"replacement","version":"9.9.9","protocol_version":1,"description":"replacement provider","config_schema":{}}' ;; + *\"op\":\"deploy\"*) printf '%s\n' '{"ok":true,"agent_id":"replacement-bytes-ran"}' ;; +esac +"# +} + +#[cfg(unix)] +#[test] +fn provider_deploy_uses_staged_bytes_after_same_inode_source_rewrite() { + use std::os::unix::fs::MetadataExt; + + let directory = tempfile::tempdir().unwrap(); + let provider = directory.path().join("provider"); + let replacement = directory.path().join("replacement"); + std::fs::write(&replacement, replacement_provider()).unwrap(); + let body = format!( + r#"read request +case "$request" in + *\"op\":\"info\"*) + cat '{}' > '{}' + chmod 700 '{}' + printf '%s\n' '{{"ok":true,"name":"original","version":"1.0.0","protocol_version":1,"description":"original provider","config_schema":{{}}}}' + ;; + *\"op\":\"deploy\"*) printf '%s\n' '{{"ok":true,"agent_id":"original-staged-bytes"}}' ;; +esac"#, + replacement.display(), + provider.display(), + provider.display(), + ); + write_test_provider(&provider, &body); + let inode_before = std::fs::metadata(&provider).unwrap().ino(); + + let id = provider_deploy(&provider, &serde_json::json!({}), &serde_json::json!({})) + .expect("deploy from immutable staged copy"); + + assert_eq!(id, "original-staged-bytes"); + assert_eq!( + std::fs::metadata(&provider).unwrap().ino(), + inode_before, + "test must rewrite the source binary in place" + ); + assert_eq!( + std::fs::read_to_string(&provider).unwrap(), + replacement_provider(), + "source pathname must contain replacement bytes before deploy" + ); +} + +#[cfg(unix)] +#[test] +fn provider_deploy_uses_staged_bytes_after_source_pathname_replacement() { + use std::os::unix::fs::{MetadataExt, PermissionsExt}; + + let directory = tempfile::tempdir().unwrap(); + let provider = directory.path().join("provider"); + let replacement = directory.path().join("replacement"); + std::fs::write(&replacement, replacement_provider()).unwrap(); + std::fs::set_permissions(&replacement, std::fs::Permissions::from_mode(0o700)).unwrap(); + let body = format!( + r#"read request +case "$request" in + *\"op\":\"info\"*) + mv '{}' '{}' + printf '%s\n' '{{"ok":true,"name":"original","version":"1.0.0","protocol_version":1,"description":"original provider","config_schema":{{}}}}' + ;; + *\"op\":\"deploy\"*) printf '%s\n' '{{"ok":true,"agent_id":"original-staged-bytes"}}' ;; +esac"#, + replacement.display(), + provider.display(), + ); + write_test_provider(&provider, &body); + let inode_before = std::fs::metadata(&provider).unwrap().ino(); + + let id = provider_deploy(&provider, &serde_json::json!({}), &serde_json::json!({})) + .expect("deploy from immutable staged copy"); + + assert_eq!(id, "original-staged-bytes"); + assert_ne!( + std::fs::metadata(&provider).unwrap().ino(), + inode_before, + "test must replace the source pathname with a different inode" + ); + assert_eq!( + std::fs::read_to_string(&provider).unwrap(), + replacement_provider(), + "source pathname must contain replacement bytes before deploy" + ); +} + +#[cfg(unix)] +#[test] +fn provider_deploy_refuses_mismatch_before_sending_agent_secret() { + let directory = tempfile::tempdir().unwrap(); + let provider = directory.path().join("provider"); + let marker = directory.path().join("deploy-received"); + let body = format!( + r#"read request +case "$request" in + *\"op\":\"info\"*) printf '%s\n' '{{"ok":true,"name":"test","version":"2.0.0","protocol_version":2,"description":"test provider","config_schema":{{}}}}' ;; + *\"op\":\"deploy\"*) touch '{}'; printf '%s\n' '{{"ok":true,"agent_id":"bad"}}' ;; +esac"#, + marker.display() + ); + write_test_provider(&provider, &body); + + let error = provider_deploy( + &provider, + &serde_json::json!({"private_key_nsec": "nsec1must-not-cross"}), + &serde_json::json!({}), + ) + .unwrap_err(); + assert!(error.contains("protocol version 2"), "{error}"); + assert!(!marker.exists()); + assert!(!error.contains("nsec1must-not-cross")); +} + +#[cfg(unix)] +#[test] +fn provider_deploy_requires_an_explicit_integer_protocol_version() { + let directory = tempfile::tempdir().unwrap(); + let provider = directory.path().join("provider"); + write_test_provider( + &provider, + r#"read request +printf '%s\n' '{"ok":true,"version":"1.0.0"}'"#, + ); + + let error = + provider_deploy(&provider, &serde_json::json!({}), &serde_json::json!({})).unwrap_err(); + assert!( + error.contains("missing integer protocol_version"), + "{error}" + ); +} + +#[test] +fn provider_info_requires_the_complete_flat_wire_shape() { + let complete = serde_json::json!({ + "ok": true, + "name": "kubernetes", + "version": "1.0.0", + "protocol_version": 1, + "description": "Kubernetes provider", + "config_schema": {} + }); + assert!(validate_provider_info(&complete).is_ok()); + + let mut missing = complete.clone(); + missing.as_object_mut().unwrap().remove("config_schema"); + assert!(validate_provider_info(&missing) + .unwrap_err() + .contains("config_schema")); + + let mut nested = complete; + nested.as_object_mut().unwrap().insert( + "provider".into(), + serde_json::json!({"protocol_version": 1}), + ); + assert!(validate_provider_info(&nested) + .unwrap_err() + .contains("unknown field provider")); +} + +#[test] +fn validate_provider_config_rejects_secret_key() { + let cfg = serde_json::json!({"api_key": "val"}); + assert!(validate_provider_config(&cfg).is_err()); +} + +#[test] +fn validate_provider_config_rejects_nested() { + let cfg = serde_json::json!({"region": {"us": "east"}}); + assert!(validate_provider_config(&cfg).is_err()); +} + +#[test] +fn validate_provider_config_accepts_scalars() { + let cfg = serde_json::json!({"region": "us-east-1", "tier": "standard"}); + assert!(validate_provider_config(&cfg).is_ok()); +} + +#[test] +fn validate_provider_config_allows_key_as_substring() { + // "keyboard", "monkey" contain "key" as substring but not as a word segment. + let cfg = serde_json::json!({"keyboard_layout": "us", "monkey_wrench": "tight"}); + assert!(validate_provider_config(&cfg).is_ok()); +} + +#[test] +fn validate_provider_config_rejects_camel_case_secrets() { + assert!(validate_provider_config(&serde_json::json!({"apiKey": "val"})).is_err()); + assert!(validate_provider_config(&serde_json::json!({"accessToken": "val"})).is_err()); + assert!(validate_provider_config(&serde_json::json!({"clientSecret": "val"})).is_err()); + // ALL-CAPS variants + assert!(validate_provider_config(&serde_json::json!({"apiKEY": "val"})).is_err()); + assert!(validate_provider_config(&serde_json::json!({"accessTOKEN": "val"})).is_err()); +} + +#[test] +fn split_config_key_handles_all_styles() { + assert_eq!(split_config_key("apiKey"), vec!["api", "key"]); + assert_eq!(split_config_key("access_token"), vec!["access", "token"]); + assert_eq!(split_config_key("keyboard"), vec!["keyboard"]); + assert_eq!(split_config_key("client-secret"), vec!["client", "secret"]); + // Acronym runs stay together + assert_eq!(split_config_key("APIKey"), vec!["api", "key"]); + assert_eq!(split_config_key("apiKEY"), vec!["api", "key"]); + assert_eq!(split_config_key("accessTOKEN"), vec!["access", "token"]); + assert_eq!(split_config_key("MyAPIKey"), vec!["my", "api", "key"]); +} + +#[test] +fn provider_filename_strips_the_windows_extension() { + assert_eq!( + provider_id_from_filename("buzz-backend-kubernetes"), + Some("kubernetes") + ); + assert_eq!( + provider_id_from_filename("buzz-backend-kubernetes.exe"), + Some("kubernetes") + ); + assert_eq!( + provider_id_from_filename("buzz-backend-kubernetes.EXE"), + Some("kubernetes") + ); + assert_eq!( + provider_id_from_filename("buzz-backend-kubernetes.bat"), + Some("kubernetes") + ); + assert_eq!( + provider_id_from_filename("buzz-backend-kubernetes.CMD"), + Some("kubernetes") + ); + assert_eq!( + provider_id_from_filename("buzz-backend-my-provider"), + Some("my-provider") + ); + assert_eq!(provider_id_from_filename("other"), None); +} + +#[test] +fn resolve_provider_binary_rejects_invalid_ids() { + // Path traversal + assert!(resolve_provider_binary("../evil").is_err()); + // Empty + assert!(resolve_provider_binary("").is_err()); + // Uppercase + assert!(resolve_provider_binary("MyProvider").is_err()); + // Spaces + assert!(resolve_provider_binary("my provider").is_err()); + // Shell metacharacters + assert!(resolve_provider_binary("foo;rm -rf /").is_err()); + // Valid format but not on PATH — should fail with "not found" + assert!(resolve_provider_binary("nonexistent-test-id-12345").is_err()); +} + +#[test] +fn resolve_provider_binary_accepts_valid_id_format() { + // Valid ID format should pass validation. If the binary happens to + // exist on PATH, Ok is returned; otherwise Err contains "not found" + // (not "invalid provider ID"). Either outcome proves validation passed. + match resolve_provider_binary("zzz-nonexistent-test-provider") { + Ok(_) => {} // unlikely but fine — binary exists + Err(e) => assert!( + e.contains("not found"), + "expected 'not found' error, got: {e}" + ), + } +} diff --git a/desktop/src-tauri/src/managed_agents/config_bridge/reader.rs b/desktop/src-tauri/src/managed_agents/config_bridge/reader.rs index 372d2cfde1..c51f325cf3 100644 --- a/desktop/src-tauri/src/managed_agents/config_bridge/reader.rs +++ b/desktop/src-tauri/src/managed_agents/config_bridge/reader.rs @@ -3,15 +3,17 @@ use crate::managed_agents::types::ManagedAgentRecord; use super::types::*; -/// Build the full config surface for an agent, merging all four tiers. +/// Build the full config surface for an agent, merging all tiers. /// -/// Pre-spawn (no session cache): tiers 2a (env vars / record) and 2b (config files). -/// Post-spawn (session cache present): adds tiers 1a (ACP native) and 1b (ACP configOptions). +/// Inherited values flow through `tiers` — a sanitized snapshot of the +/// persona and global tiers assembled at the command boundary. Each field +/// builder constructs its own candidate list and resolves via +/// `resolve_with_override`. pub(crate) fn read_config_surface( record: &ManagedAgentRecord, runtime_meta: Option<&KnownAcpRuntime>, session_cache: Option<&SessionConfigCache>, - baseline: Option<(&str, ConfigOrigin)>, + tiers: &InheritedConfigTiers, ) -> RuntimeConfigSurface { let is_pre_spawn = session_cache.is_none(); @@ -27,14 +29,7 @@ pub(crate) fn read_config_surface( }) .unwrap_or_else(|| (RuntimeFileConfig::default(), false)); - // Tier 2a: record-level values (Buzz-explicit). - let record_model = record.model.clone(); - let record_provider = record - .env_vars - .get(runtime_meta.and_then(|m| m.provider_env_var).unwrap_or("")) - .cloned() - .or_else(|| record.provider.clone()); // structured provider field as fallback - + // Runtime-specific env var keys. let supports_acp_model = runtime_meta.is_some_and(|m| m.supports_acp_model_switching); let model_env_var = runtime_meta.and_then(|m| m.model_env_var); let provider_env_var = runtime_meta.and_then(|m| m.provider_env_var); @@ -48,10 +43,6 @@ pub(crate) fn read_config_surface( let context_limit_env_var = runtime_meta.and_then(|m| m.context_limit_env_var); // Tier 1b: ACP configOptions from session cache. - // For unstable/switchable agents, current_model comes from the `models` - // field. For stable agents that only report model via configOptions - // (category="model", current_value), fall back to find_config_option_value - // so their current model is surfaced in the panel. let acp_model = session_cache.and_then(|c| { c.current_model .clone() @@ -59,61 +50,53 @@ pub(crate) fn read_config_surface( }); let acp_mode = session_cache.and_then(|c| find_config_option_value(c, "mode")); let acp_effort = session_cache.and_then(|c| find_config_option_value(c, "effort")); - let record_effort = thinking_env_var - .and_then(|k| record.env_vars.get(k)) - .cloned(); let model_overridden = session_cache.is_some_and(|c| c.model_overridden); let normalized = NormalizedConfig { - model: Some(apply_runtime_override( - build_model_field( - &record_model, - &file_config.model, - &acp_model, - model_env_var, - supports_acp_model, - is_pre_spawn, - session_cache, - required_fields.contains(&"model"), - ), - acp_model.as_deref(), - baseline, + model: Some(build_model_field( + record, + &file_config.model, + &acp_model, + model_env_var, + supports_acp_model, + is_pre_spawn, + session_cache, + required_fields.contains(&"model"), model_overridden, + tiers, )), provider: build_provider_field( - &record_provider, + record, &file_config.provider, provider_env_var, provider_locked, required_fields.contains(&"provider"), + tiers, ), mode: build_mode_field(&file_config.mode, &acp_mode, is_pre_spawn, session_cache), thinking_effort: build_thinking_field( - &record_effort, + record, &file_config.thinking_effort, &acp_effort, thinking_env_var, is_pre_spawn, session_cache, + tiers, ), max_output_tokens: build_numeric_env_field( max_tokens_env_var, - &record.env_vars, + record, &file_config.max_output_tokens, + tiers, ), context_limit: build_numeric_env_field( context_limit_env_var, - &record.env_vars, + record, &file_config.context_limit, + tiers, ), - system_prompt: build_system_prompt_field( - &record - .system_prompt - .clone() - .or_else(|| record.env_vars.get("BUZZ_ACP_SYSTEM_PROMPT").cloned()), - &file_config.system_prompt, - ), + system_prompt: build_system_prompt_field(record, &file_config.system_prompt, tiers), }; // Advanced fields from config file extras. @@ -130,7 +113,7 @@ pub(crate) fn read_config_surface( }) .collect(); - // Collect the env var keys already covered by normalized fields so we don't double-surface them. + // Collect the env var keys already covered by normalized fields. let normalized_env_keys: Vec<&str> = [ model_env_var, provider_env_var, @@ -144,15 +127,13 @@ pub(crate) fn read_config_surface( .collect(); // Tier 2a: remaining env vars not covered by normalized fields. - // Env var wins over config file for the same key (tier 2a > 2b), so skip - // keys already present in file_config.extra. let mut advanced = advanced; for (k, v) in &record.env_vars { if normalized_env_keys.contains(&k.as_str()) { continue; } if file_config.extra.contains_key(k) { - continue; // config file already surfaced this key + continue; } advanced.push(ConfigField { key: k.clone(), @@ -178,8 +159,6 @@ pub(crate) fn read_config_surface( { ConfigTierStatus::Available } else { - // Post-spawn without native config data is also Pending — it arrives - // asynchronously after the session/new response. ConfigTierStatus::Pending } } else { @@ -226,9 +205,27 @@ fn mcp_config_file_path_for_runtime(runtime: &KnownAcpRuntime) -> Option } } +/// Extract an env-backed candidate value for `env_key` from each tier in +/// spawn precedence: record env > persona env > global env > definition env. +/// Returns `[record, persona, global, definition]` — `None` when key is absent. +fn env_candidates<'a>( + env_key: &str, + record_env: &'a std::collections::BTreeMap, + persona_env: &'a std::collections::BTreeMap, + global_env: &'a std::collections::BTreeMap, + definition_env: &'a std::collections::BTreeMap, +) -> [Option<&'a str>; 4] { + [ + record_env.get(env_key).map(String::as_str), + persona_env.get(env_key).map(String::as_str), + global_env.get(env_key).map(String::as_str), + definition_env.get(env_key).map(String::as_str), + ] +} + #[allow(clippy::too_many_arguments)] fn build_model_field( - record_model: &Option, + record: &ManagedAgentRecord, file_model: &Option, acp_model: &Option, model_env_var: Option<&str>, @@ -236,30 +233,109 @@ fn build_model_field( is_pre_spawn: bool, session_cache: Option<&SessionConfigCache>, is_required: bool, + model_overridden: bool, + tiers: &InheritedConfigTiers, ) -> NormalizedField { - // Precedence: Buzz-explicit > ACP current > config file - let (value, origin) = if let Some(ref m) = record_model { - (Some(m.clone()), ConfigOrigin::BuzzExplicit) - } else if let Some(ref m) = acp_model { - (Some(m.clone()), ConfigOrigin::AcpConfigOption) - } else if let Some(ref m) = file_model { - (Some(m.clone()), ConfigOrigin::ConfigFile) - } else { - // No value from any tier. EnvVar is the sentinel origin for "no value - // resolved" — there is no dedicated None-origin variant. The panel - // renders this as an empty/absent field. - (None, ConfigOrigin::EnvVar) - }; + let [rec_env, pers_env, glob_env, def_env] = model_env_var + .map(|k| { + env_candidates( + k, + &record.env_vars, + &tiers.persona_env, + &tiers.global_env, + &tiers.definition_env, + ) + }) + .unwrap_or([None, None, None, None]); - // The secondary expresses ONLY the static record-vs-file precedence: a - // Buzz-explicit model shadowing a config-file model. The live-session - // override (acp vs record/persona) is exclusively `apply_runtime_override`'s - // job, gated on `model_overridden`. Surfacing `acp_model` here would leak an - // override row even when no live switch has been applied. - let (overridden_value, overridden_origin) = if record_model.is_some() && file_model.is_some() { - (file_model.clone(), Some(ConfigOrigin::ConfigFile)) + // Structured record model (definition-less only; linked cleared upstream). + let struct_record = record.model.as_deref(); + let struct_persona = tiers.persona_model.as_deref(); + let struct_global = tiers.global_model.as_deref(); + + // Configured candidates in spawn order: record env > persona env > global env > + // definition env > struct record > struct persona > struct global > file. + // The file entry is always last; everything before it is a "configured" candidate + // that gates whether ACP participates as a fallback (see any_configured below). + let configured: &[(Option<&str>, ConfigOrigin)] = &[ + (rec_env, ConfigOrigin::BuzzExplicit), + (pers_env, ConfigOrigin::PersonaDefault), + (glob_env, ConfigOrigin::GlobalDefault), + (def_env, ConfigOrigin::HarnessDefault), + (struct_record, ConfigOrigin::BuzzExplicit), + (struct_persona, ConfigOrigin::PersonaDefault), + (struct_global, ConfigOrigin::GlobalDefault), + (file_model.as_deref(), ConfigOrigin::ConfigFile), + ]; + // "Configured" = any non-file candidate. The file entry is always last, so + // slicing to len()-1 is equivalent to the old magic `[..6]` and stays correct + // if the array ever grows again. + let any_configured = configured[..configured.len() - 1] + .iter() + .any(|(v, _)| v.is_some()); + + // When model_overridden is true and ACP is present, ACP is the live winner. + // The top configured candidate becomes the secondary (the overridden baseline). + // Equal-value case: ACP == baseline → fall through to normal resolution so + // the field carries the correct baseline origin rather than RuntimeOverride. + if model_overridden { + if let Some(acp) = acp_model.as_deref() { + let baseline = configured.iter().find(|(v, _)| v.is_some()); + match baseline { + Some((Some(baseline_value), _)) if acp == *baseline_value => { + // Equal-value switch: no real divergence. + // Fall through to the normal resolve path below — it will + // return the same value with its true baseline origin, with + // no secondary row. + } + Some((Some(baseline_value), baseline_origin)) => { + return NormalizedField { + value: Some(acp.to_string()), + origin: ConfigOrigin::RuntimeOverride, + write_via: model_write_mechanism( + is_pre_spawn, + supports_acp_model, + session_cache, + model_env_var, + ), + overridden_value: Some(baseline_value.to_string()), + overridden_origin: Some(baseline_origin.clone()), + is_required, + }; + } + _ => { + // No configured baseline — ACP is the only source. + return NormalizedField { + value: Some(acp.to_string()), + origin: ConfigOrigin::RuntimeOverride, + write_via: model_write_mechanism( + is_pre_spawn, + supports_acp_model, + session_cache, + model_env_var, + ), + overridden_value: None, + overridden_origin: None, + is_required, + }; + } + } + } + } + + let (value, origin, overridden_value, overridden_origin) = if !any_configured { + // No configured candidate: ACP participates as AcpConfigOption fallback. + let full: &[(Option<&str>, ConfigOrigin)] = &[ + (acp_model.as_deref(), ConfigOrigin::AcpConfigOption), + (file_model.as_deref(), ConfigOrigin::ConfigFile), + ]; + resolve_with_override(full).unwrap_or((None, ConfigOrigin::EnvVar, None, None)) } else { - (None, None) + // ACP excluded: a configured value is pending and wins over live ACP. + match resolve_with_override(configured) { + Some(r) => r, + None => (None, ConfigOrigin::EnvVar, None, None), + } }; let write_via = model_write_mechanism( @@ -280,7 +356,6 @@ fn build_model_field( } /// Resolve how the model field is written back to the runtime. -/// Prefer ACP `set_config_option`/`set_model` post-spawn, else env-var respawn. fn model_write_mechanism( is_pre_spawn: bool, supports_acp_model: bool, @@ -301,67 +376,13 @@ fn model_write_mechanism( } } -/// Re-key the model field as a live runtime override when the harness signals -/// that a `SwitchModel` control signal set the model (Phase 3c). -/// -/// The override-active signal is `model_overridden` from the -/// `session_config_captured` payload — NOT `acp_model != persona_model`, which -/// would false-positive when a persona model is edited mid-life while the -/// session is stale on the old model. -/// -/// `baseline` is the value the live model overrides, paired with its true -/// origin — `(persona_model, PersonaDefault)` for a persona-linked agent, or -/// `(record_model, BuzzExplicit)` for a genuine-explicit agent that live- -/// switched. It is `Some` only when there is such a baseline to override -/// against; otherwise the field passes through unchanged. Carrying the origin -/// in the pair (rather than hardcoding it) lets the secondary be tagged by its -/// real source instead of always reading `PersonaDefault`. -/// -/// The `acp == baseline_value` short-circuit keeps a live pick of the baseline -/// model itself from rendering a no-op "override of X with X". It yields a -/// CLEAN single-value field — `overridden_value`/`overridden_origin` cleared — -/// rather than passing `base` through, because `build_model_field` already -/// populates `base`'s secondary with an `AcpConfigOption` row for the -/// record-model-plus-live-session case; returning `base` would leak that -/// spurious row. The override preserves the base field's write mechanism — only -/// the displayed value, origin, and secondary change. -fn apply_runtime_override( - base: NormalizedField, - acp_model: Option<&str>, - baseline: Option<(&str, ConfigOrigin)>, - model_overridden: bool, -) -> NormalizedField { - if !model_overridden { - return base; - } - let (Some(acp), Some((baseline_value, baseline_origin))) = (acp_model, baseline) else { - return base; - }; - if acp == baseline_value { - // Live pick equals the baseline — no real divergence. Strip any - // secondary `build_model_field` may have produced so the panel shows a - // single clean value rather than "X overridden by X". - return NormalizedField { - overridden_value: None, - overridden_origin: None, - ..base - }; - } - NormalizedField { - value: Some(acp.to_string()), - origin: ConfigOrigin::RuntimeOverride, - overridden_value: Some(baseline_value.to_string()), - overridden_origin: Some(baseline_origin), - ..base - } -} - fn build_provider_field( - record_provider: &Option, + record: &ManagedAgentRecord, file_provider: &Option, provider_env_var: Option<&str>, provider_locked: bool, is_required: bool, + tiers: &InheritedConfigTiers, ) -> Option { if provider_locked { return Some(NormalizedField { @@ -374,15 +395,43 @@ fn build_provider_field( }); } - let tiers: &[(Option<&str>, ConfigOrigin)] = &[ - (record_provider.as_deref(), ConfigOrigin::BuzzExplicit), + let [rec_env, pers_env, glob_env, def_env] = provider_env_var + .map(|k| { + env_candidates( + k, + &record.env_vars, + &tiers.persona_env, + &tiers.global_env, + &tiers.definition_env, + ) + }) + .unwrap_or([None, None, None, None]); + + let struct_record = record.provider.as_deref(); + + let tiers_list: &[(Option<&str>, ConfigOrigin)] = &[ + (rec_env, ConfigOrigin::BuzzExplicit), + (pers_env, ConfigOrigin::PersonaDefault), + (glob_env, ConfigOrigin::GlobalDefault), + (def_env, ConfigOrigin::HarnessDefault), + (struct_record, ConfigOrigin::BuzzExplicit), + ( + tiers.persona_provider.as_deref(), + ConfigOrigin::PersonaDefault, + ), + ( + tiers.global_provider.as_deref(), + ConfigOrigin::GlobalDefault, + ), (file_provider.as_deref(), ConfigOrigin::ConfigFile), ]; - let (value, origin, overridden_value, overridden_origin) = match resolve_with_override(tiers) { - Some(resolved) => resolved, - None if is_required => (None, ConfigOrigin::EnvVar, None, None), - None => return None, - }; + + let (value, origin, overridden_value, overridden_origin) = + match resolve_with_override(tiers_list) { + Some(resolved) => resolved, + None if is_required => (None, ConfigOrigin::EnvVar, None, None), + None => return None, + }; let write_via = if let Some(env_key) = provider_env_var { ConfigWriteMechanism::RespawnWithEnvVar { @@ -432,20 +481,38 @@ fn build_mode_field( }) } +#[allow(clippy::too_many_arguments)] fn build_thinking_field( - record_effort: &Option, + record: &ManagedAgentRecord, file_effort: &Option, acp_effort: &Option, thinking_env_var: Option<&str>, is_pre_spawn: bool, session_cache: Option<&SessionConfigCache>, + tiers: &InheritedConfigTiers, ) -> Option { - let tiers: &[(Option<&str>, ConfigOrigin)] = &[ - (record_effort.as_deref(), ConfigOrigin::BuzzExplicit), + // Tier ordering: record env > ACP > persona env > global env > definition env > config file. + let [rec_env, pers_env, glob_env, def_env] = thinking_env_var + .map(|k| { + env_candidates( + k, + &record.env_vars, + &tiers.persona_env, + &tiers.global_env, + &tiers.definition_env, + ) + }) + .unwrap_or([None, None, None, None]); + + let tiers_list: &[(Option<&str>, ConfigOrigin)] = &[ + (rec_env, ConfigOrigin::BuzzExplicit), (acp_effort.as_deref(), ConfigOrigin::AcpConfigOption), + (pers_env, ConfigOrigin::PersonaDefault), + (glob_env, ConfigOrigin::GlobalDefault), + (def_env, ConfigOrigin::HarnessDefault), (file_effort.as_deref(), ConfigOrigin::ConfigFile), ]; - let (value, origin, overridden_value, overridden_origin) = resolve_with_override(tiers)?; + let (value, origin, overridden_value, overridden_origin) = resolve_with_override(tiers_list)?; let write_via = if !is_pre_spawn && has_config_option(session_cache, "effort") { ConfigWriteMechanism::AcpSetConfigOption { @@ -469,67 +536,105 @@ fn build_thinking_field( }) } -/// Numeric fields (max_output_tokens, context_limit) — env-var tier wins over -/// config-file tier. When an env var key is given and present in the record's -/// env_vars map the field is BuzzExplicit + RespawnWithEnvVar; otherwise if the -/// config file supplied a value it is ConfigFile + ReadOnly; otherwise None. +/// Numeric fields (max_output_tokens, context_limit). +/// Tier ordering: record env > persona env > global env > config file. fn build_numeric_env_field( env_var: Option<&'static str>, - record_env: &std::collections::BTreeMap, + record: &ManagedAgentRecord, file_value: &Option, + tiers: &InheritedConfigTiers, ) -> Option { - if let Some(key) = env_var { - if let Some(v) = record_env.get(key) { - return Some(NormalizedField { - value: Some(v.clone()), - origin: ConfigOrigin::BuzzExplicit, - write_via: ConfigWriteMechanism::RespawnWithEnvVar { - env_key: key.to_string(), - }, - overridden_value: file_value.clone(), - overridden_origin: file_value.as_ref().map(|_| ConfigOrigin::ConfigFile), - is_required: false, - }); + let [rec_env, pers_env, glob_env, def_env] = env_var + .map(|k| { + env_candidates( + k, + &record.env_vars, + &tiers.persona_env, + &tiers.global_env, + &tiers.definition_env, + ) + }) + .unwrap_or([None, None, None, None]); + + let tiers_list: &[(Option<&str>, ConfigOrigin)] = &[ + (rec_env, ConfigOrigin::BuzzExplicit), + (pers_env, ConfigOrigin::PersonaDefault), + (glob_env, ConfigOrigin::GlobalDefault), + (def_env, ConfigOrigin::HarnessDefault), + (file_value.as_deref(), ConfigOrigin::ConfigFile), + ]; + + let (value, origin, overridden_value, overridden_origin) = resolve_with_override(tiers_list)?; + + let write_via = if let Some(key) = env_var { + ConfigWriteMechanism::RespawnWithEnvVar { + env_key: key.to_string(), } - } - file_value.as_ref().map(|v| NormalizedField { - value: Some(v.clone()), - origin: ConfigOrigin::ConfigFile, - write_via: ConfigWriteMechanism::ReadOnly, - overridden_value: None, - overridden_origin: None, + } else { + ConfigWriteMechanism::ReadOnly + }; + + Some(NormalizedField { + value, + origin, + write_via, + overridden_value, + overridden_origin, is_required: false, }) } -/// Record/env prompt wins (BuzzExplicit, respawnable); a config-file prompt it -/// shadows is reported as the overridden secondary. A config-file-only prompt -/// — no record/env value to shadow it — is surfaced directly (read-only) -/// instead of being dropped: a prompt that drives the agent should always be -/// visible somewhere in the panel. +/// System prompt field. +/// +/// Tier ordering per v3 plan: record env > persona env > global env > +/// struct record > struct persona > config file. +/// +/// Env tiers sit above structured per spawn contract: `descriptor.env` is +/// written last (after the structured prompt), so env wins on collision. +/// `GlobalAgentConfig` has no structured system_prompt, so the global tier +/// is env-only. `BUZZ_ACP_SYSTEM_PROMPT` is not reserved and is therefore +/// a real global env tier. fn build_system_prompt_field( - record_prompt: &Option, + record: &ManagedAgentRecord, file_prompt: &Option, + tiers: &InheritedConfigTiers, ) -> Option { - if let Some(v) = record_prompt { - return Some(NormalizedField { - value: Some(v.clone()), - origin: ConfigOrigin::BuzzExplicit, - write_via: ConfigWriteMechanism::RespawnWithEnvVar { - env_key: "BUZZ_ACP_SYSTEM_PROMPT".to_string(), - }, - overridden_value: file_prompt.clone(), - overridden_origin: file_prompt.as_ref().map(|_| ConfigOrigin::ConfigFile), - is_required: false, - }); - } + const PROMPT_ENV_KEY: &str = "BUZZ_ACP_SYSTEM_PROMPT"; - file_prompt.as_ref().map(|v| NormalizedField { - value: Some(v.clone()), - origin: ConfigOrigin::ConfigFile, - write_via: ConfigWriteMechanism::ReadOnly, - overridden_value: None, - overridden_origin: None, + let [rec_env, pers_env, glob_env, def_env] = env_candidates( + PROMPT_ENV_KEY, + &record.env_vars, + &tiers.persona_env, + &tiers.global_env, + &tiers.definition_env, + ); + + // Structured record prompt (definition-less only; linked cleared upstream). + let struct_record = record.system_prompt.as_deref(); + + let tiers_list: &[(Option<&str>, ConfigOrigin)] = &[ + (rec_env, ConfigOrigin::BuzzExplicit), // record env + (pers_env, ConfigOrigin::PersonaDefault), // persona env + (glob_env, ConfigOrigin::GlobalDefault), // global env + (def_env, ConfigOrigin::HarnessDefault), // definition env + (struct_record, ConfigOrigin::BuzzExplicit), // struct record + ( + tiers.persona_prompt.as_deref(), + ConfigOrigin::PersonaDefault, + ), // struct persona + (file_prompt.as_deref(), ConfigOrigin::ConfigFile), + ]; + + let (value, origin, overridden_value, overridden_origin) = resolve_with_override(tiers_list)?; + + Some(NormalizedField { + value, + origin, + write_via: ConfigWriteMechanism::RespawnWithEnvVar { + env_key: PROMPT_ENV_KEY.to_string(), + }, + overridden_value, + overridden_origin, is_required: false, }) } diff --git a/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests.rs b/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests.rs index 4ee4ec79c3..153db1bbd8 100644 --- a/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests.rs +++ b/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests.rs @@ -120,11 +120,53 @@ fn test_record() -> ManagedAgentRecord { } } +/// Default empty tiers: no persona or global inheritance. +fn no_tiers() -> InheritedConfigTiers { + InheritedConfigTiers::default() +} + +/// Tiers with only global env set (for AC-1 style tests). +fn global_env_tiers(key: &str, val: &str) -> InheritedConfigTiers { + let mut global_env = BTreeMap::new(); + global_env.insert(key.to_string(), val.to_string()); + InheritedConfigTiers { + global_env, + ..Default::default() + } +} + +/// Tiers with only persona env set. +fn persona_env_tiers(key: &str, val: &str) -> InheritedConfigTiers { + let mut persona_env = BTreeMap::new(); + persona_env.insert(key.to_string(), val.to_string()); + InheritedConfigTiers { + persona_env, + ..Default::default() + } +} + +/// Tiers with both persona and global env set for the same key. +fn persona_and_global_env_tiers( + key: &str, + persona_val: &str, + global_val: &str, +) -> InheritedConfigTiers { + let mut persona_env = BTreeMap::new(); + persona_env.insert(key.to_string(), persona_val.to_string()); + let mut global_env = BTreeMap::new(); + global_env.insert(key.to_string(), global_val.to_string()); + InheritedConfigTiers { + persona_env, + global_env, + ..Default::default() + } +} + #[test] fn pre_spawn_surface_reports_pending_acp_tiers() { let record = test_record(); let runtime = test_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); assert!(surface.is_pre_spawn); assert_eq!(surface.sources.acp_native, ConfigTierStatus::Pending); @@ -140,7 +182,7 @@ fn surface_reports_mcp_specific_config_path() { let record = test_record(); let runtime = test_runtime(); let surface = with_goose_path_root(None, || { - read_config_surface(&record, Some(runtime), None, None) + read_config_surface(&record, Some(runtime), None, &no_tiers()) }); let path = surface @@ -159,7 +201,7 @@ fn goose_mcp_config_path_follows_path_root_override() { let record = test_record(); let runtime = test_runtime(); let surface = with_goose_path_root(Some("/tmp/buzz-goose-root"), || { - read_config_surface(&record, Some(runtime), None, None) + read_config_surface(&record, Some(runtime), None, &no_tiers()) }); let expected_path = Path::new("/tmp/buzz-goose-root") @@ -183,7 +225,7 @@ fn claude_surface_uses_mcp_config_path_not_settings_path() { config_file_path: Some("~/.claude/settings.json"), ..*test_runtime() }; - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); assert!(surface .sources @@ -203,7 +245,7 @@ fn record_model_overrides_file_model() { record.model = Some("explicit-model".to_string()); let runtime = test_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let model = surface.normalized.model.unwrap(); assert_eq!(model.value.as_deref(), Some("explicit-model")); assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); @@ -216,7 +258,7 @@ fn provider_locked_shows_locked() { provider_locked: true, ..*test_runtime() }; - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let provider = surface.normalized.provider.unwrap(); assert_eq!(provider.value.as_deref(), Some("Anthropic (locked)")); assert_eq!(provider.origin, ConfigOrigin::HarnessConstraint); @@ -242,7 +284,7 @@ fn post_spawn_with_model_config_option_uses_acp() { captured_at: "".to_string(), }; - let surface = read_config_surface(&record, Some(runtime), Some(&cache), None); + let surface = read_config_surface(&record, Some(runtime), Some(&cache), &no_tiers()); assert!(!surface.is_pre_spawn); let model = surface.normalized.model.unwrap(); assert_eq!(model.value.as_deref(), Some("claude-opus-4")); @@ -266,53 +308,86 @@ fn acp_model_overrides_file_model_with_override_tracking() { captured_at: "".to_string(), }; - let surface = read_config_surface(&record, Some(runtime), Some(&cache), None); + let surface = read_config_surface(&record, Some(runtime), Some(&cache), &no_tiers()); let model = surface.normalized.model.unwrap(); assert_eq!(model.value.as_deref(), Some("acp-model")); assert_eq!(model.origin, ConfigOrigin::AcpConfigOption); - // The goose config file might have a model too — since we can't control - // the actual file in a unit test, just verify the override fields are populated - // when we manually construct the scenario via build_model_field. } -// ── Persona resolution integration tests ──────────────────────────── +// ── Persona / global tier integration tests ────────────────────────────────── // -// These simulate the call-site pattern in agent_config.rs: -// 1. Inject persona-resolved values into the record (as if absent) -// 2. Call read_config_surface (reader tags them BuzzExplicit) -// 3. Re-tag injected fields to PersonaDefault -// -// This exercises the same logic path as get_agent_config_surface without -// requiring Tauri AppHandle/State infrastructure. +// These exercise the tiers-based candidate resolution for model, provider, and +// system_prompt via `InheritedConfigTiers` — replacing the old inject+retag +// simulation tests. #[test] -fn persona_model_injection_produces_persona_default_origin() { - let mut record = test_record(); - // Simulate: record has no model, persona provides one. - // The call-site injects it before calling the reader. - record.model = Some("persona-model".to_string()); +fn persona_model_tier_produces_persona_default_origin() { + let record = test_record(); // no record.model let runtime = test_runtime(); + let tiers = InheritedConfigTiers { + persona_model: Some("persona-model".to_string()), + ..Default::default() + }; - let mut surface = read_config_surface(&record, Some(runtime), None, None); - - // Reader sees injected model as BuzzExplicit. - let model = surface.normalized.model.as_ref().unwrap(); - assert_eq!(model.value.as_deref(), Some("persona-model")); - assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); - - // Call-site re-tags (simulating had_model == false). - if let Some(ref mut field) = surface.normalized.model { - if field.origin == ConfigOrigin::BuzzExplicit { - field.origin = ConfigOrigin::PersonaDefault; - } - } + let surface = read_config_surface(&record, Some(runtime), None, &tiers); let model = surface.normalized.model.unwrap(); assert_eq!(model.value.as_deref(), Some("persona-model")); assert_eq!(model.origin, ConfigOrigin::PersonaDefault); } -// ── Runtime override (Phase 3c) ────────────────────────────────────── +#[test] +fn global_model_tier_produces_global_default_origin() { + let record = test_record(); + let runtime = test_runtime(); + let tiers = InheritedConfigTiers { + global_model: Some("global-model".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let model = surface.normalized.model.unwrap(); + assert_eq!(model.value.as_deref(), Some("global-model")); + assert_eq!(model.origin, ConfigOrigin::GlobalDefault); +} + +#[test] +fn persona_provider_tier_produces_persona_default_origin() { + let record = test_record(); + let runtime = test_runtime(); + let tiers = InheritedConfigTiers { + persona_provider: Some("anthropic".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let provider = surface.normalized.provider.unwrap(); + assert_eq!(provider.value.as_deref(), Some("anthropic")); + assert_eq!(provider.origin, ConfigOrigin::PersonaDefault); +} + +#[test] +fn persona_prompt_tier_produces_persona_default_origin() { + let record = test_record(); + let runtime = test_runtime(); + let tiers = InheritedConfigTiers { + persona_prompt: Some("You are a helpful assistant.".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let prompt = surface.normalized.system_prompt.unwrap(); + assert_eq!( + prompt.value.as_deref(), + Some("You are a helpful assistant.") + ); + assert_eq!(prompt.origin, ConfigOrigin::PersonaDefault); +} + +// ── Runtime override (model_overridden gate) ────────────────────────────────── // // A live ModelPicker switch is signalled by `model_overridden: true` in the // `session_config_captured` payload. The reader keys the override-active @@ -321,7 +396,7 @@ fn persona_model_injection_produces_persona_default_origin() { #[test] fn runtime_override_wins_display_when_model_overridden_is_true() { - // Persona-linked agent (record.model == None); persona == "persona-model". + // Persona-linked agent (record.model == None); persona model via tiers. // A live switch pushed "live-model" to the session and set model_overridden. let record = test_record(); let runtime = test_runtime(); @@ -334,29 +409,27 @@ fn runtime_override_wins_display_when_model_overridden_is_true() { goose_native_config: None, captured_at: "".to_string(), }; + let tiers = InheritedConfigTiers { + persona_model: Some("persona-model".to_string()), + ..Default::default() + }; - let surface = read_config_surface( - &record, - Some(runtime), - Some(&cache), - Some(("persona-model", ConfigOrigin::PersonaDefault)), - ); + let surface = read_config_surface(&record, Some(runtime), Some(&cache), &tiers); let model = surface.normalized.model.unwrap(); // Override wins the display value with a runtime-override origin. assert_eq!(model.value.as_deref(), Some("live-model")); assert_eq!(model.origin, ConfigOrigin::RuntimeOverride); - // Persona is the secondary value (not struck through — the UI keys off - // the RuntimeOverride origin to suppress strikethrough). + // Persona is the secondary value. assert_eq!(model.overridden_value.as_deref(), Some("persona-model")); assert_eq!(model.overridden_origin, Some(ConfigOrigin::PersonaDefault)); } #[test] fn no_runtime_override_when_model_overridden_is_false() { - // At spawn the session's current_model == persona model (BUZZ_ACP_MODEL - // is set to the persona model) and model_overridden is false. No override; - // the field falls through to normal precedence. + // At spawn the session's current_model == persona model and + // model_overridden is false. No override; field falls through to normal + // precedence. let record = test_record(); let runtime = test_runtime(); let cache = SessionConfigCache { @@ -368,17 +441,15 @@ fn no_runtime_override_when_model_overridden_is_false() { goose_native_config: None, captured_at: "".to_string(), }; + let tiers = InheritedConfigTiers { + persona_model: Some("persona-model".to_string()), + ..Default::default() + }; - let surface = read_config_surface( - &record, - Some(runtime), - Some(&cache), - Some(("persona-model", ConfigOrigin::PersonaDefault)), - ); + let surface = read_config_surface(&record, Some(runtime), Some(&cache), &tiers); let model = surface.normalized.model.unwrap(); - // model_overridden is false => the override branch is not taken: origin - // is the normal precedence result, never RuntimeOverride. + // model_overridden is false => the override branch is not taken. assert_ne!(model.origin, ConfigOrigin::RuntimeOverride); assert_eq!(model.value.as_deref(), Some("persona-model")); assert_ne!(model.overridden_origin, Some(ConfigOrigin::PersonaDefault)); @@ -402,106 +473,51 @@ fn no_false_positive_override_when_persona_edited_mid_life() { goose_native_config: None, captured_at: "".to_string(), }; + let tiers = InheritedConfigTiers { + persona_model: Some("new-persona-model".to_string()), + ..Default::default() + }; - let surface = read_config_surface( - &record, - Some(runtime), - Some(&cache), - Some(("new-persona-model", ConfigOrigin::PersonaDefault)), - ); + let surface = read_config_surface(&record, Some(runtime), Some(&cache), &tiers); let model = surface.normalized.model.unwrap(); // model_overridden is false => no RuntimeOverride, even though // acp_model != persona_model. The old divergence-based signal would - // have false-positived here. The persona is never surfaced as the - // overridden secondary (that marker is exclusive to a real override). + // have false-positived here. assert_ne!(model.origin, ConfigOrigin::RuntimeOverride); assert_ne!(model.overridden_origin, Some(ConfigOrigin::PersonaDefault)); } -#[test] -fn persona_provider_injection_produces_persona_default_origin() { - let mut record = test_record(); - // Simulate: record has no provider env var, persona provides one. - // The call-site injects it as GOOSE_PROVIDER before calling the reader. - record - .env_vars - .insert("GOOSE_PROVIDER".to_string(), "anthropic".to_string()); - let runtime = test_runtime(); - - let mut surface = read_config_surface(&record, Some(runtime), None, None); - - // Reader sees injected provider as BuzzExplicit. - let provider = surface.normalized.provider.as_ref().unwrap(); - assert_eq!(provider.value.as_deref(), Some("anthropic")); - assert_eq!(provider.origin, ConfigOrigin::BuzzExplicit); - - // Call-site re-tags (simulating had_provider == false). - if let Some(ref mut field) = surface.normalized.provider { - if field.origin == ConfigOrigin::BuzzExplicit { - field.origin = ConfigOrigin::PersonaDefault; - } - } - - let provider = surface.normalized.provider.unwrap(); - assert_eq!(provider.value.as_deref(), Some("anthropic")); - assert_eq!(provider.origin, ConfigOrigin::PersonaDefault); -} +// ── system_prompt builder unit tests ───────────────────────────────────────── #[test] -fn persona_system_prompt_injection_produces_persona_default_origin() { - let mut record = test_record(); - // Simulate: record has no system_prompt, persona provides one via env var. - // The call-site injects it as BUZZ_ACP_SYSTEM_PROMPT before calling the reader. - record.env_vars.insert( - "BUZZ_ACP_SYSTEM_PROMPT".to_string(), - "You are a helpful assistant.".to_string(), - ); - let runtime = test_runtime(); - - let mut surface = read_config_surface(&record, Some(runtime), None, None); - - // Reader sees injected prompt as BuzzExplicit. - let prompt = surface.normalized.system_prompt.as_ref().unwrap(); - assert_eq!( - prompt.value.as_deref(), - Some("You are a helpful assistant.") - ); - assert_eq!(prompt.origin, ConfigOrigin::BuzzExplicit); - - // Call-site re-tags (simulating had_prompt == false). - if let Some(ref mut field) = surface.normalized.system_prompt { - if field.origin == ConfigOrigin::BuzzExplicit { - field.origin = ConfigOrigin::PersonaDefault; - } - } - - let prompt = surface.normalized.system_prompt.unwrap(); - assert_eq!( - prompt.value.as_deref(), - Some("You are a helpful assistant.") - ); - assert_eq!(prompt.origin, ConfigOrigin::PersonaDefault); -} - -#[test] -fn config_file_only_system_prompt_surfaces_as_read_only_config_file_field() { - // Record/env has no prompt; the config file does. It must NOT be - // dropped — it should surface with ConfigFile origin, read-only. - let field = build_system_prompt_field(&None, &Some("File-driven prompt.".to_string())).unwrap(); +fn config_file_only_system_prompt_surfaces_as_config_file_origin() { + // Record/env has no prompt; the config file does. Must surface with + // ConfigFile origin. Write mechanism is always RespawnWithEnvVar for + // system_prompt — the UI writes back via BUZZ_ACP_SYSTEM_PROMPT. + let record = test_record(); + let field = build_system_prompt_field( + &record, + &Some("File-driven prompt.".to_string()), + &no_tiers(), + ) + .unwrap(); assert_eq!(field.value.as_deref(), Some("File-driven prompt.")); assert_eq!(field.origin, ConfigOrigin::ConfigFile); - assert!(matches!(field.write_via, ConfigWriteMechanism::ReadOnly)); + assert!(matches!( + field.write_via, + ConfigWriteMechanism::RespawnWithEnvVar { ref env_key } + if env_key == "BUZZ_ACP_SYSTEM_PROMPT" + )); assert!(field.overridden_value.is_none()); } #[test] fn record_system_prompt_shadows_config_file_prompt_as_secondary() { - let field = build_system_prompt_field( - &Some("Record prompt.".to_string()), - &Some("File prompt.".to_string()), - ) - .unwrap(); + let mut record = test_record(); + record.system_prompt = Some("Record prompt.".to_string()); + let field = + build_system_prompt_field(&record, &Some("File prompt.".to_string()), &no_tiers()).unwrap(); assert_eq!(field.value.as_deref(), Some("Record prompt.")); assert_eq!(field.origin, ConfigOrigin::BuzzExplicit); assert_eq!(field.overridden_value.as_deref(), Some("File prompt.")); @@ -510,19 +526,19 @@ fn record_system_prompt_shadows_config_file_prompt_as_secondary() { #[test] fn no_system_prompt_from_any_tier_yields_none() { - assert!(build_system_prompt_field(&None, &None).is_none()); + let record = test_record(); + assert!(build_system_prompt_field(&record, &None, &no_tiers()).is_none()); } #[test] fn explicit_record_model_not_retagged_when_already_present() { let mut record = test_record(); - // Record already has its own model — persona resolution should NOT re-tag. + // Record already has its own model — origin stays BuzzExplicit. record.model = Some("explicit-model".to_string()); let runtime = test_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); - // had_model == true, so no re-tagging occurs. Origin stays BuzzExplicit. let model = surface.normalized.model.unwrap(); assert_eq!(model.value.as_deref(), Some("explicit-model")); assert_eq!(model.origin, ConfigOrigin::BuzzExplicit); @@ -544,7 +560,7 @@ fn extra_env_vars_appear_in_advanced_as_buzz_explicit() { .insert("SPROUT_ACP_MEMORY".to_string(), "mem-value".to_string()); let runtime = test_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let advanced_keys: Vec<&str> = surface.advanced.iter().map(|f| f.key.as_str()).collect(); assert!( @@ -575,21 +591,15 @@ fn extra_env_vars_appear_in_advanced_as_buzz_explicit() { #[test] fn extra_env_var_skipped_when_already_in_file_config_extra() { - // If a key is in both record.env_vars and file_config.extra, the config - // file entry wins (it was already added to advanced). The env var must - // not produce a second entry. - // - // We can't inject into file_config.extra directly in a unit test (it - // comes from disk), so we verify the dedup logic via the normalized-key - // path: GOOSE_THINKING_EFFORT is a normalized key and must not appear - // in advanced even if set in env_vars. + // If a key is normalized, it must not appear in advanced even if set + // in env_vars. let mut record = test_record(); record .env_vars .insert("GOOSE_THINKING_EFFORT".to_string(), "high".to_string()); let runtime = test_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let advanced_keys: Vec<&str> = surface.advanced.iter().map(|f| f.key.as_str()).collect(); assert!( @@ -598,7 +608,7 @@ fn extra_env_var_skipped_when_already_in_file_config_extra() { ); } -// ── buzz-agent normalized env-var field tests ─────────────────────────────── +// ── buzz-agent normalized env-var field tests ───────────────────────────────── // // buzz-agent uses env vars (not a config file) for max_output_tokens and // context_limit. build_numeric_env_field must surface these as BuzzExplicit @@ -649,7 +659,7 @@ fn buzz_agent_max_output_tokens_from_env_is_buzz_explicit() { ); let runtime = buzz_agent_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let field = surface.normalized.max_output_tokens.unwrap(); assert_eq!(field.value.as_deref(), Some("8192")); @@ -670,7 +680,7 @@ fn buzz_agent_context_limit_from_env_is_buzz_explicit() { ); let runtime = buzz_agent_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let field = surface.normalized.context_limit.unwrap(); assert_eq!(field.value.as_deref(), Some("100000")); @@ -688,7 +698,7 @@ fn buzz_agent_max_tokens_absent_when_no_env_var_or_file() { let record = test_record(); let runtime = buzz_agent_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); assert!( surface.normalized.max_output_tokens.is_none(), @@ -713,7 +723,7 @@ fn buzz_agent_max_tokens_env_var_not_double_surfaced_in_advanced() { ); let runtime = buzz_agent_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let advanced_keys: Vec<&str> = surface.advanced.iter().map(|f| f.key.as_str()).collect(); assert!( @@ -734,7 +744,7 @@ fn buzz_agent_thinking_effort_from_env_is_buzz_explicit() { .insert("BUZZ_AGENT_THINKING_EFFORT".to_string(), "high".to_string()); let runtime = buzz_agent_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let field = surface.normalized.thinking_effort.unwrap(); assert_eq!(field.value.as_deref(), Some("high")); @@ -755,7 +765,7 @@ fn buzz_agent_thinking_effort_env_var_not_double_surfaced_in_advanced() { ); let runtime = buzz_agent_runtime(); - let surface = read_config_surface(&record, Some(runtime), None, None); + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); let advanced_keys: Vec<&str> = surface.advanced.iter().map(|f| f.key.as_str()).collect(); assert!( @@ -764,10 +774,20 @@ fn buzz_agent_thinking_effort_env_var_not_double_surfaced_in_advanced() { ); } +// ── provider builder unit tests ─────────────────────────────────────────────── + #[test] fn missing_required_provider_still_returns_dropdown_field() { - let provider = build_provider_field(&None, &None, Some("GOOSE_PROVIDER"), false, true) - .expect("required provider field should be surfaced even when empty"); + let record = test_record(); + let provider = build_provider_field( + &record, + &None, + Some("GOOSE_PROVIDER"), + false, + true, + &no_tiers(), + ) + .expect("required provider field should be surfaced even when empty"); assert_eq!(provider.value, None); assert_eq!(provider.origin, ConfigOrigin::EnvVar); @@ -776,5 +796,156 @@ fn missing_required_provider_still_returns_dropdown_field() { #[test] fn missing_optional_provider_stays_hidden() { - assert!(build_provider_field(&None, &None, Some("GOOSE_PROVIDER"), false, false).is_none()); + let record = test_record(); + assert!(build_provider_field( + &record, + &None, + Some("GOOSE_PROVIDER"), + false, + false, + &no_tiers() + ) + .is_none()); } + +// ── thinking_effort persona/global tier tests (AC-1..5) ────────────────────── +// +// The plan's acceptance criteria for effort tier resolution. +// Tier ordering: record env > ACP > persona env > global env > config file. + +fn buzz_agent_rt() -> &'static KnownAcpRuntime { + crate::managed_agents::discovery::known_acp_runtime_exact("buzz-agent") + .expect("buzz-agent must be in catalog") +} + +/// AC-1: no record effort, global env has effort → GlobalDefault. +/// Real-world case: global-agent-config has BUZZ_AGENT_THINKING_EFFORT=high, +/// per-agent record has no env_vars → effort must surface with GlobalDefault origin. +#[test] +fn global_effort_surfaces_as_global_default_when_record_has_none() { + let record = test_record(); + let runtime = buzz_agent_rt(); + let tiers = global_env_tiers("BUZZ_AGENT_THINKING_EFFORT", "high"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let effort = surface + .normalized + .thinking_effort + .expect("effort must surface from global tier"); + assert_eq!(effort.value.as_deref(), Some("high")); + assert_eq!(effort.origin, ConfigOrigin::GlobalDefault); +} + +/// AC-2: persona env has effort, global also has effort → PersonaDefault wins, shadows global. +#[test] +fn persona_effort_shadows_global_and_tags_persona_default() { + let record = test_record(); + let runtime = buzz_agent_rt(); + let tiers = persona_and_global_env_tiers("BUZZ_AGENT_THINKING_EFFORT", "medium", "high"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let effort = surface + .normalized + .thinking_effort + .expect("effort must surface from persona tier"); + assert_eq!(effort.value.as_deref(), Some("medium")); + assert_eq!(effort.origin, ConfigOrigin::PersonaDefault); + // global is the overridden baseline + assert_eq!(effort.overridden_value.as_deref(), Some("high")); + assert_eq!(effort.overridden_origin, Some(ConfigOrigin::GlobalDefault)); +} + +/// AC-3: record-level effort wins over persona and global, stays BuzzExplicit. +#[test] +fn record_effort_outranks_persona_and_global_keeps_buzz_explicit() { + let mut record = test_record(); + record.env_vars.insert( + "BUZZ_AGENT_THINKING_EFFORT".to_string(), + "xhigh".to_string(), + ); + let runtime = buzz_agent_rt(); + let tiers = persona_and_global_env_tiers("BUZZ_AGENT_THINKING_EFFORT", "medium", "high"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let effort = surface + .normalized + .thinking_effort + .expect("effort must surface from record tier"); + assert_eq!(effort.value.as_deref(), Some("xhigh")); + assert_eq!(effort.origin, ConfigOrigin::BuzzExplicit); +} + +/// AC-4: no effort from any tier → thinking_effort field is absent. +#[test] +fn no_effort_anywhere_yields_no_thinking_effort_field() { + let record = test_record(); + let runtime = buzz_agent_rt(); + + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); + + assert!( + surface.normalized.thinking_effort.is_none(), + "thinking_effort must be None when no tier has a value" + ); +} + +/// AC-5 (conflicting-ACP): inherited effort set (global=high) + live ACP effort=low +/// → ACP wins as primary (AcpConfigOption), global is the overridden secondary. +#[test] +fn acp_effort_wins_over_inherited_global_effort_as_secondary() { + let record = test_record(); + let runtime = buzz_agent_rt(); + let cache = SessionConfigCache { + config_options: vec![AcpConfigOptionEntry { + config_id: "effort".to_string(), + category: Some("effort".to_string()), + display_name: Some("Effort".to_string()), + current_value: Some("low".to_string()), + options: vec![], + }], + available_modes: vec![], + available_models: vec![], + current_model: None, + model_overridden: false, + goose_native_config: None, + captured_at: "".to_string(), + }; + let tiers = global_env_tiers("BUZZ_AGENT_THINKING_EFFORT", "high"); + + let surface = read_config_surface(&record, Some(runtime), Some(&cache), &tiers); + + let effort = surface + .normalized + .thinking_effort + .expect("effort must surface from ACP tier"); + // Live ACP value wins. + assert_eq!(effort.value.as_deref(), Some("low")); + assert_eq!(effort.origin, ConfigOrigin::AcpConfigOption); + // Global is surfaced as the overridden baseline. + assert_eq!(effort.overridden_value.as_deref(), Some("high")); + assert_eq!(effort.overridden_origin, Some(ConfigOrigin::GlobalDefault)); +} + +// ── Numerics inheritance tests ──────────────────────────────────────────────── +// +// max_output_tokens and context_limit gain persona/global tiers. + +#[test] +fn numeric_max_tokens_inherits_from_global_env() { + let record = test_record(); + let runtime = buzz_agent_runtime(); + let tiers = global_env_tiers("BUZZ_AGENT_MAX_OUTPUT_TOKENS", "16384"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let field = surface.normalized.max_output_tokens.unwrap(); + assert_eq!(field.value.as_deref(), Some("16384")); + assert_eq!(field.origin, ConfigOrigin::GlobalDefault); +} + +// ── Extended tests (split file to respect line-count ratchet) ──────────────── +#[path = "reader_tests_ext.rs"] +mod ext; diff --git a/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests_ext.rs b/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests_ext.rs new file mode 100644 index 0000000000..8613124f25 --- /dev/null +++ b/desktop/src-tauri/src/managed_agents/config_bridge/reader_tests_ext.rs @@ -0,0 +1,258 @@ +//! Additional tests for `config_bridge/reader.rs` — split out to keep +//! `reader_tests.rs` under the 1000-line file-size ratchet. +//! +//! Included as `mod ext` inside `reader_tests.rs`, so `use super::*` gives +//! access to all helpers and types from that module. + +use super::*; + +// ── Numerics inheritance tests ──────────────────────────────────────────────── +// +// max_output_tokens and context_limit gain persona/global tiers. + +#[test] +fn numeric_context_limit_inherits_from_persona_env() { + let record = test_record(); + let runtime = buzz_agent_runtime(); + let tiers = persona_env_tiers("BUZZ_AGENT_MAX_CONTEXT_TOKENS", "200000"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let field = surface.normalized.context_limit.unwrap(); + assert_eq!(field.value.as_deref(), Some("200000")); + assert_eq!(field.origin, ConfigOrigin::PersonaDefault); +} + +#[test] +fn record_max_tokens_overrides_global_env_with_secondary() { + let mut record = test_record(); + record.env_vars.insert( + "BUZZ_AGENT_MAX_OUTPUT_TOKENS".to_string(), + "8192".to_string(), + ); + let runtime = buzz_agent_runtime(); + let tiers = global_env_tiers("BUZZ_AGENT_MAX_OUTPUT_TOKENS", "16384"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let field = surface.normalized.max_output_tokens.unwrap(); + assert_eq!(field.value.as_deref(), Some("8192")); + assert_eq!(field.origin, ConfigOrigin::BuzzExplicit); + // Global value is the overridden secondary. + assert_eq!(field.overridden_value.as_deref(), Some("16384")); + assert_eq!(field.overridden_origin, Some(ConfigOrigin::GlobalDefault)); +} + +// ── Env-vs-structured collision tests (plan v3, Phase 2) ───────────────────── + +/// Collision test 1: persona structured prompt + global env BUZZ_ACP_SYSTEM_PROMPT +/// → global env wins (env block sits entirely above structured). +#[test] +fn global_env_prompt_wins_over_persona_structured_prompt() { + let record = test_record(); + let runtime = test_runtime(); + let tiers = InheritedConfigTiers { + global_env: { + let mut m = BTreeMap::new(); + m.insert( + "BUZZ_ACP_SYSTEM_PROMPT".to_string(), + "global-env-prompt".to_string(), + ); + m + }, + persona_prompt: Some("persona-structured-prompt".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let prompt = surface.normalized.system_prompt.unwrap(); + assert_eq!(prompt.value.as_deref(), Some("global-env-prompt")); + assert_eq!(prompt.origin, ConfigOrigin::GlobalDefault); +} + +/// Collision test 2: structured persona/record model + higher user-env value at +/// the runtime's model key → env value wins. +#[test] +fn persona_env_model_wins_over_persona_structured_model() { + let record = test_record(); // no record.model + let runtime = test_runtime(); // GOOSE_MODEL + let tiers = InheritedConfigTiers { + persona_env: { + let mut m = BTreeMap::new(); + m.insert("GOOSE_MODEL".to_string(), "env-model".to_string()); + m + }, + persona_model: Some("struct-persona-model".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let model = surface.normalized.model.unwrap(); + // persona env outranks persona struct because env candidates precede struct + assert_eq!(model.value.as_deref(), Some("env-model")); + assert_eq!(model.origin, ConfigOrigin::PersonaDefault); +} + +/// Collision test 3: no env representation → structured persona/record/global +/// fallback and provenance remain intact. +#[test] +fn structured_fallback_intact_when_no_env_representation() { + let record = test_record(); // no record.model, no env vars + let runtime = test_runtime(); + let tiers = InheritedConfigTiers { + persona_model: Some("struct-persona-model".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let model = surface.normalized.model.unwrap(); + assert_eq!(model.value.as_deref(), Some("struct-persona-model")); + assert_eq!(model.origin, ConfigOrigin::PersonaDefault); +} + +// ── Post-sanitization fallthrough test ─────────────────────────────────────── +// +// Sanitization itself happens at the command boundary in `build_inherited_tiers` +// (a value with a NUL byte or an oversize value is dropped from the tier) and is +// pinned by the tests in `commands/agent_config_tests.rs`. The reader only ever +// sees the sanitized result, so what it must guarantee is the downstream half: +// a key stripped from one tier falls through to the next. + +/// A key absent from the global env tier — the shape the reader sees after the +/// command boundary strips an invalid value — falls through to the persona tier. +#[test] +fn post_sanitization_empty_global_env_falls_through_to_persona_tier() { + let record = test_record(); + let runtime = buzz_agent_rt(); + // No global env (stripped); persona provides the valid fallback. + let tiers = persona_env_tiers("BUZZ_AGENT_THINKING_EFFORT", "medium"); + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + // Persona value surfaces instead of the stripped global value. + let effort = surface.normalized.thinking_effort.unwrap(); + assert_eq!(effort.value.as_deref(), Some("medium")); + assert_eq!(effort.origin, ConfigOrigin::PersonaDefault); +} + +// ── Pass-3 prompt collision test ───────────────────────────────────────────── +// +// From Thufir's pass-3 verdict MINOR clarification (promoted to required): +// definition-less record with both structured and env prompt — env wins. + +/// Pass-3 clarification: record.system_prompt = A + record env +/// BUZZ_ACP_SYSTEM_PROMPT = B → B wins as BuzzExplicit. +/// The env block sits above the struct block per v3 candidate-preparation +/// contract; current reader semantics (struct before env) would be wrong. +#[test] +fn record_env_prompt_wins_over_record_struct_prompt_as_buzz_explicit() { + let mut record = test_record(); + record.system_prompt = Some("struct-prompt-A".to_string()); + record.env_vars.insert( + "BUZZ_ACP_SYSTEM_PROMPT".to_string(), + "env-prompt-B".to_string(), + ); + let runtime = test_runtime(); + + let surface = read_config_surface(&record, Some(runtime), None, &no_tiers()); + + let prompt = surface.normalized.system_prompt.unwrap(); + assert_eq!(prompt.value.as_deref(), Some("env-prompt-B")); + assert_eq!(prompt.origin, ConfigOrigin::BuzzExplicit); + // Struct prompt is the secondary. + assert_eq!(prompt.overridden_value.as_deref(), Some("struct-prompt-A")); + assert_eq!(prompt.overridden_origin, Some(ConfigOrigin::BuzzExplicit)); +} + +// ── Definition env tier tests (Layer 2b) ───────────────────────────────────── +// +// The harness definition's `env` block sits below global env and above +// structured values in spawn's precedence (Layer 2b). These tests exercise +// the reader's mapping of that tier to `HarnessDefault` origin. + +/// Definition env wins over structured persona model when no user-env or +/// global-env candidate is present. +#[test] +fn definition_env_beats_structured_persona_model() { + let record = test_record(); // no record.model, no record.env_vars + let runtime = test_runtime(); // model_env_var = "GOOSE_MODEL" + let tiers = InheritedConfigTiers { + definition_env: { + let mut m = BTreeMap::new(); + m.insert("GOOSE_MODEL".to_string(), "harness-model".to_string()); + m + }, + persona_model: Some("persona-struct-model".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let model = surface.normalized.model.unwrap(); + assert_eq!(model.value.as_deref(), Some("harness-model")); + assert_eq!(model.origin, ConfigOrigin::HarnessDefault); + // Structured persona model is the overridden secondary. + assert_eq!( + model.overridden_value.as_deref(), + Some("persona-struct-model") + ); + assert_eq!(model.overridden_origin, Some(ConfigOrigin::PersonaDefault)); +} + +/// Global env beats definition env — user-settable tiers always win over the +/// harness author's defaults. +#[test] +fn global_env_beats_definition_env() { + let record = test_record(); + let runtime = test_runtime(); // model_env_var = "GOOSE_MODEL" + let tiers = InheritedConfigTiers { + global_env: { + let mut m = BTreeMap::new(); + m.insert("GOOSE_MODEL".to_string(), "global-model".to_string()); + m + }, + definition_env: { + let mut m = BTreeMap::new(); + m.insert("GOOSE_MODEL".to_string(), "harness-model".to_string()); + m + }, + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let model = surface.normalized.model.unwrap(); + assert_eq!(model.value.as_deref(), Some("global-model")); + assert_eq!(model.origin, ConfigOrigin::GlobalDefault); + // Harness default is the overridden secondary. + assert_eq!(model.overridden_value.as_deref(), Some("harness-model")); + assert_eq!(model.overridden_origin, Some(ConfigOrigin::HarnessDefault)); +} + +/// A reserved key in the definition env is stripped by sanitization and must +/// not reach the reader. This test exercises the reader's contract (a key +/// absent from the tier falls through) — sanitization itself is pinned in +/// the `agent_config_tests.rs` constructor tests. +#[test] +fn reserved_key_absent_from_definition_env_falls_through() { + let record = test_record(); + let runtime = test_runtime(); // model_env_var = "GOOSE_MODEL" + // definition_env contains only an unrelated key — the env map here is what + // the command boundary would produce after stripping a reserved key; the + // reader must fall through to the next tier (persona structured model). + let tiers = InheritedConfigTiers { + definition_env: BTreeMap::new(), // stripped — nothing survives + persona_model: Some("persona-struct-model".to_string()), + ..Default::default() + }; + + let surface = read_config_surface(&record, Some(runtime), None, &tiers); + + let model = surface.normalized.model.unwrap(); + // Falls through to persona structured model. + assert_eq!(model.value.as_deref(), Some("persona-struct-model")); + assert_eq!(model.origin, ConfigOrigin::PersonaDefault); +} diff --git a/desktop/src-tauri/src/managed_agents/config_bridge/types.rs b/desktop/src-tauri/src/managed_agents/config_bridge/types.rs index 15ccb718e7..6ca2592538 100644 --- a/desktop/src-tauri/src/managed_agents/config_bridge/types.rs +++ b/desktop/src-tauri/src/managed_agents/config_bridge/types.rs @@ -2,6 +2,41 @@ use std::collections::BTreeMap; use serde::{Deserialize, Serialize}; +/// Sanitized inherited config tiers passed to the reader. +/// +/// Built at the `agent_config` command boundary with spawn-equivalent +/// sanitization: reserved, malformed, NUL-value, and oversize-value env keys +/// are stripped (matching `merged_user_env`). Structured fields are +/// normalized: blank/whitespace-only values collapse to `None`. +/// +/// Orphaned persona links (persona_id references a missing persona) produce +/// an empty persona env tier and `None` for all structured persona fields — +/// the panel still renders from record/global. This diverges deliberately from +/// spawn's `OrphanedInstance` refusal, which is a spawn-safety property the +/// display surface does not need to enforce. +#[derive(Debug, Clone, Default)] +pub struct InheritedConfigTiers { + /// Sanitized env vars from the linked persona definition. + pub persona_env: BTreeMap, + /// Sanitized env vars from the global agent config. + pub global_env: BTreeMap, + /// Sanitized env vars from the resolved harness definition (`HarnessDefinition::env`). + /// Sits below global env and above structured values, matching spawn Layer 2b. + /// Empty for preset harnesses (all shipped presets have `env: {}`); only + /// user-authored custom harness JSONs with a non-empty `env` block contribute here. + pub definition_env: BTreeMap, + /// Structured model from the linked persona (non-blank only). + pub persona_model: Option, + /// Structured provider from the linked persona (non-blank only). + pub persona_provider: Option, + /// Structured system_prompt from the linked persona (non-blank only). + pub persona_prompt: Option, + /// Structured model from global config (non-blank only). + pub global_model: Option, + /// Structured provider from global config (non-blank only). + pub global_provider: Option, +} + /// Where a config value came from — determines precedence and UI annotations. #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] #[serde(rename_all = "camelCase")] @@ -17,14 +52,12 @@ pub enum ConfigOrigin { /// Read from harness config file on disk (tier 2b, lowest precedence). ConfigFile, /// Value inherited from persona defaults. - /// Populated by the `get_agent_config_surface` call site: persona values are - /// resolved before calling the reader, then the surface is post-processed to - /// re-tag injected fields from `BuzzExplicit` to `PersonaDefault`. + /// Populated when a persona's env var or structured field wins for this + /// field in the reader's candidate resolution. PersonaDefault, /// Value inherited from global agent configuration defaults. /// The lowest user-settable layer — active when neither the agent record nor - /// the linked persona specifies a value. Re-tagged from `BuzzExplicit` by the - /// `resolve_config_surface` call site, analogously to `PersonaDefault`. + /// the linked persona specifies a value. GlobalDefault, /// Live runtime model override applied via the ModelPicker (Phase 3). /// The ACP session's current model diverges from the persona model because @@ -35,6 +68,11 @@ pub enum ConfigOrigin { /// env var. E.g. Claude Code only supports Anthropic as a provider; the /// "locked" display is synthesized by the config bridge, not read from disk. HarnessConstraint, + /// Value comes from a custom harness definition's `env` block. + /// Sits below global env and above structured persona/global values, + /// matching spawn Layer 2b. Only reachable for user-authored custom harness + /// JSONs with a non-empty `env` block; preset harnesses always have empty env. + HarnessDefault, } /// How a config field can be written back to the runtime. diff --git a/desktop/src-tauri/src/managed_agents/discovery.rs b/desktop/src-tauri/src/managed_agents/discovery.rs index 8d1b8a5013..248625ce93 100644 --- a/desktop/src-tauri/src/managed_agents/discovery.rs +++ b/desktop/src-tauri/src/managed_agents/discovery.rs @@ -9,10 +9,10 @@ use crate::managed_agents::{ AcpAvailabilityStatus, AcpRuntimeCatalogEntry, AuthStatus, CommandAvailabilityInfo, HarnessSource, }; - mod presets; mod runtime_metadata; - +#[macro_use] +mod windows_install; use presets::{preset_catalog_entry, PRESET_HARNESSES}; pub(crate) use presets::{preset_harness_definitions, preset_harness_ids}; pub(crate) use runtime_metadata::KnownAcpRuntime; @@ -85,7 +85,7 @@ const KNOWN_ACP_RUNTIMES: &[KnownAcpRuntime] = &[ cli_install_commands: &["curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | CONFIGURE=false bash"], // Goose's stable release currently publishes only the Unix installer; // its official Windows instructions intentionally point at this main-branch script. - cli_install_commands_windows: &["powershell.exe -NoProfile -ExecutionPolicy Bypass -Command \"$env:CONFIGURE='false'; irm https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1 | iex\""], + cli_install_commands_windows: &[windows_install_command!("goose", "https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1", "$env:CONFIGURE='false'; ")], adapter_install_commands: &[], cli_install_instructions_url: "https://goose-docs.ai/docs/getting-started/installation/", adapter_install_instructions_url: "", @@ -117,7 +117,7 @@ const KNOWN_ACP_RUNTIMES: &[KnownAcpRuntime] = &[ mcp_hooks: false, underlying_cli: Some("claude"), cli_install_commands: &["curl -fsSL https://claude.ai/install.sh | bash"], - cli_install_commands_windows: &["powershell.exe -NoProfile -ExecutionPolicy Bypass -Command \"irm https://claude.ai/install.ps1 | iex\""], + cli_install_commands_windows: &[windows_install_command!("claude", "https://claude.ai/install.ps1")], adapter_install_commands: &["npm install -g @agentclientprotocol/claude-agent-acp"], cli_install_instructions_url: "https://code.claude.com/docs/en/getting-started", adapter_install_instructions_url: "https://github.com/agentclientprotocol/claude-agent-acp", @@ -149,7 +149,7 @@ const KNOWN_ACP_RUNTIMES: &[KnownAcpRuntime] = &[ mcp_hooks: false, underlying_cli: Some("codex"), cli_install_commands: &["curl -fsSL https://chatgpt.com/codex/install.sh | sh"], - cli_install_commands_windows: &["powershell.exe -NoProfile -ExecutionPolicy Bypass -Command \"irm https://chatgpt.com/codex/install.ps1 | iex\""], + cli_install_commands_windows: &[windows_install_command!("codex", "https://chatgpt.com/codex/install.ps1")], adapter_install_commands: &["npm install -g @agentclientprotocol/codex-acp"], cli_install_instructions_url: "https://developers.openai.com/codex/cli/", adapter_install_instructions_url: "https://github.com/agentclientprotocol/codex-acp", diff --git a/desktop/src-tauri/src/managed_agents/discovery/presets.rs b/desktop/src-tauri/src/managed_agents/discovery/presets.rs index 72c4657dc7..3622b21c4a 100644 --- a/desktop/src-tauri/src/managed_agents/discovery/presets.rs +++ b/desktop/src-tauri/src/managed_agents/discovery/presets.rs @@ -106,7 +106,7 @@ pub(super) const PRESET_HARNESSES: &[PresetHarness] = &[ label: "Oh My Pi", command: "omp", args: &["acp"], - install_instructions_url: "https://github.com/can1357/oh-my-pi", + install_instructions_url: "https://omp.sh/", install_hint: "Buzz talks to Oh My Pi through its CLI's ACP mode (omp acp).", underlying_cli: None, }, diff --git a/desktop/src-tauri/src/managed_agents/discovery/windows_install.rs b/desktop/src-tauri/src/managed_agents/discovery/windows_install.rs new file mode 100644 index 0000000000..09e27a62be --- /dev/null +++ b/desktop/src-tauri/src/managed_agents/discovery/windows_install.rs @@ -0,0 +1,225 @@ +//! Defender-safe construction of the Windows PowerShell CLI install commands. +//! +//! # Why the shape matters +//! +//! Windows Defender's ML classifier flags the bare `irm | iex` command +//! line as `Trojan:Win32/Commando.A!ml` — piping a downloaded string straight +//! into `Invoke-Expression` is a textbook dropper signature, so the *command +//! line itself* is scored, independent of what the URL actually serves. The +//! spawn is denied before PowerShell runs, surfacing as +//! `failed to spawn shell: Access is denied. (os error 5)`, and the block is +//! sticky: Defender's "Allow" button does not clear it. +//! +//! [`windows_install_command!`] emits the two-step form instead — download the +//! vendor script to a file, then execute the file — which does not match that +//! signature. All three runtimes use it, not only the one observed failing: +//! Goose and Claude escaped by scoring under the classifier threshold, which is +//! luck rather than design, and the threshold is not ours to depend on. +//! +//! # Why one macro instead of three literals +//! +//! The catalog needs `&'static str`, so the commands must be built at compile +//! time from literals. Emitting them from a single macro means the security +//! shape is defined once and cannot drift between runtimes as URLs change — +//! a per-runtime literal would let one entry silently regress to `iex`. +//! +//! # Exit-code fidelity +//! +//! [#2892](https://github.com/block/buzz/pull/2892) established that an install +//! step must not report success when the download failed. Two pieces preserve +//! that here, and both are load-bearing: +//! +//! - `$ErrorActionPreference='Stop'` makes a failed `Invoke-RestMethod` +//! terminate the whole command. Without it a failed download falls through to +//! `& $installer` on a path that does not exist, and PowerShell exits **0** — +//! the exact masking #2892 removed, in a new dress. `Stop` also prevents +//! executing a *stale* installer left in `$env:TEMP` by an earlier run. +//! - `exit $LASTEXITCODE` propagates the vendor script's own exit code. Without +//! it PowerShell reports its own status and a vendor failure of `3` flattens +//! to `1`, losing the distinction the retry logic reads. +//! +//! Verified against `pwsh` over a local HTTP server: vendor exit 3 surfaces as +//! 3, vendor exit 0 as 0, a 404 and an unresolvable host as non-zero, and a +//! planted stale installer is never executed. The old `irm | iex` shape +//! produces identical codes for all four, so this is not a behavior change. +//! +//! # Quoting contract +//! +//! The emitted body is wrapped in one double-quote pair, which +//! `install_powershell_command` strips before handing the body to PowerShell. +//! The body therefore uses **only single quotes** internally; a double quote +//! would terminate that pair early and truncate the command. + +/// Build the Windows CLI install command for one runtime. +/// +/// `slug` names the downloaded script (`buzz-install-.ps1`) so concurrent +/// installs of different runtimes cannot overwrite each other's file. The +/// optional third argument carries a runtime's env prefix (Goose's +/// `$env:CONFIGURE='false'; `) and must end with `; `. +/// +/// See the module docs for why each fragment is present. +macro_rules! windows_install_command { + ($slug:literal, $url:literal) => { + windows_install_command!($slug, $url, "") + }; + ($slug:literal, $url:literal, $env_prefix:literal) => { + concat!( + "powershell.exe -NoProfile -ExecutionPolicy Bypass -Command \"", + $env_prefix, + "$ErrorActionPreference='Stop'; ", + "$installer=Join-Path $env:TEMP 'buzz-install-", + $slug, + ".ps1'; ", + "Invoke-RestMethod ", + $url, + " -OutFile $installer; ", + "& $installer; ", + "exit $LASTEXITCODE\"", + ) + }; +} + +#[cfg(test)] +mod tests { + use crate::managed_agents::known_acp_runtime_exact; + + /// Every runtime that ships a Windows install command. `cli_install_commands_windows` + /// is read directly rather than through `cli_install_commands_for_os()` so these + /// assertions cover the Windows strings while running on the Linux CI host. + fn windows_install_commands() -> Vec<(&'static str, &'static str)> { + ["goose", "claude", "codex"] + .into_iter() + .flat_map(|id| { + known_acp_runtime_exact(id) + .expect("runtime must exist in the catalog") + .cli_install_commands_windows + .iter() + .map(move |command| (id, *command)) + }) + .collect() + } + + /// The whole point of the change: no runtime may carry the flagged + /// download-and-execute-in-one-line signature. + #[test] + fn test_no_windows_install_command_pipes_a_download_into_iex() { + for (id, command) in windows_install_commands() { + assert!( + !command.contains("| iex"), + "{id}: `irm | iex` is the shape Defender flags as Trojan:Win32/Commando.A!ml; \ + download to a file and execute the file instead. Got: {command}" + ); + assert!( + !command.contains("Invoke-Expression"), + "{id}: Invoke-Expression on downloaded content carries the same signature. \ + Got: {command}" + ); + } + } + + /// All three runtimes must be hardened, not just the one observed failing. + /// Goose and Claude escaped only by scoring under the classifier threshold. + #[test] + fn test_every_windows_install_command_downloads_to_a_file_then_executes_it() { + let commands = windows_install_commands(); + assert_eq!( + commands.len(), + 3, + "expected exactly one Windows install command for each of goose, claude, codex" + ); + for (id, command) in commands { + assert!( + command.contains("-OutFile $installer"), + "{id}: must download the vendor script to a file. Got: {command}" + ); + assert!( + command.contains("& $installer"), + "{id}: must execute the downloaded file. Got: {command}" + ); + assert!( + command.contains(&format!("buzz-install-{id}.ps1")), + "{id}: script name must be runtime-specific so concurrent installs of \ + different runtimes cannot overwrite each other. Got: {command}" + ); + } + } + + /// Guards the #2892 regression: without `Stop`, a failed download falls + /// through to a missing file and PowerShell exits 0, reporting a failed + /// install as a success. Without `exit $LASTEXITCODE`, the vendor's own + /// exit code is replaced by PowerShell's. + #[test] + fn test_every_windows_install_command_preserves_failure_exit_codes() { + for (id, command) in windows_install_commands() { + assert!( + command.contains("$ErrorActionPreference='Stop'"), + "{id}: a failed download must abort instead of running a missing or stale \ + installer and exiting 0 (see #2892). Got: {command}" + ); + assert!( + command.contains("exit $LASTEXITCODE"), + "{id}: the vendor script's exit code must propagate. Got: {command}" + ); + } + } + + /// `install_powershell_command` strips exactly one outer double-quote pair. + /// An inner double quote would close that pair early and truncate the body. + #[test] + fn test_every_windows_install_command_quotes_the_body_exactly_once() { + for (id, command) in windows_install_commands() { + let body = command + .split_once(" -Command ") + .map(|(_, body)| body) + .unwrap_or_else(|| panic!("{id}: command must pass a -Command body: {command}")); + assert!( + body.starts_with('"') && body.ends_with('"'), + "{id}: body must be wrapped in one double-quote pair. Got: {body}" + ); + assert_eq!( + body.matches('"').count(), + 2, + "{id}: body must contain no inner double quotes — one would terminate the \ + outer pair early and truncate the command. Got: {body}" + ); + } + } + + /// Goose's installer reads `CONFIGURE` to stay non-interactive; losing the + /// prefix hangs the install waiting on input that never comes. + #[test] + fn test_goose_windows_install_command_keeps_its_env_prefix() { + let goose = known_acp_runtime_exact("goose").unwrap(); + let command = goose.cli_install_commands_windows[0]; + assert!( + command.contains("$env:CONFIGURE='false'"), + "goose must stay non-interactive. Got: {command}" + ); + assert!( + command.find("$env:CONFIGURE='false'").unwrap() + < command.find("Invoke-RestMethod").unwrap(), + "the env prefix must be set before the installer runs. Got: {command}" + ); + } + + /// The vendor URLs are the payload; pin them so a refactor of the shared + /// shape cannot silently retarget a download. + #[test] + fn test_windows_install_commands_target_the_official_vendor_urls() { + for (id, expected) in [ + ( + "goose", + "https://raw.githubusercontent.com/aaif-goose/goose/main/download_cli.ps1", + ), + ("claude", "https://claude.ai/install.ps1"), + ("codex", "https://chatgpt.com/codex/install.ps1"), + ] { + let runtime = known_acp_runtime_exact(id).unwrap(); + let command = runtime.cli_install_commands_windows[0]; + assert!( + command.contains(&format!("Invoke-RestMethod {expected} -OutFile")), + "{id}: must download from {expected}. Got: {command}" + ); + } + } +} diff --git a/desktop/src-tauri/src/managed_agents/env_vars.rs b/desktop/src-tauri/src/managed_agents/env_vars.rs index 592a5cbbd9..1653371e7f 100644 --- a/desktop/src-tauri/src/managed_agents/env_vars.rs +++ b/desktop/src-tauri/src/managed_agents/env_vars.rs @@ -77,6 +77,10 @@ pub(crate) const RESERVED_ENV_KEYS: &[&str] = &[ "BUZZ_ACP_RESPOND_TO", "BUZZ_ACP_RESPOND_TO_ALLOWLIST", "BUZZ_ACP_AGENT_OWNER", + // Remote lifetime/presence policy: user env must not disable the + // desktop/provider-owned bounds while the saved record still promises them. + "BUZZ_ACP_EXIT_AFTER_INACTIVITY", + "BUZZ_ACP_NO_PRESENCE", // Readiness handoff: desktop is the ONLY readiness source. A saved or // ambient env var must not be able to forge setup mode (NotReady) on a // Ready agent or suppress it (empty/stale payload) on a NotReady one. @@ -307,28 +311,5 @@ pub(crate) fn live_persona_env( .unwrap_or_default() } -/// Resolve live env_vars for a linked persona, loading personas from disk. -/// -/// Returns the persona's `env_vars` map if a persona_id is provided and found; -/// returns an empty map if no persona is linked. Errors if the linked persona -/// is missing. Used by the provider deploy path, which has no pre-loaded -/// persona slice. -pub(crate) fn resolve_persona_env( - app: &tauri::AppHandle, - persona_id: Option<&str>, -) -> Result, String> { - let Some(pid) = persona_id else { - return Ok(std::collections::BTreeMap::new()); - }; - let personas = super::load_personas(app).map_err(|e| { - format!("failed to load personas while resolving env for persona `{pid}`: {e}") - })?; - let persona = personas - .into_iter() - .find(|p| p.id == pid) - .ok_or_else(|| format!("persona `{pid}` not found while resolving env"))?; - Ok(persona.env_vars) -} - #[cfg(test)] mod tests; diff --git a/desktop/src-tauri/src/managed_agents/env_vars/tests.rs b/desktop/src-tauri/src/managed_agents/env_vars/tests.rs index cf57b12546..534c2e0835 100644 --- a/desktop/src-tauri/src/managed_agents/env_vars/tests.rs +++ b/desktop/src-tauri/src/managed_agents/env_vars/tests.rs @@ -158,6 +158,15 @@ fn reserved_keys_include_respond_to_gate() { } } +#[test] +fn reserved_keys_include_remote_lifetime_policy() { + for key in ["BUZZ_ACP_EXIT_AFTER_INACTIVITY", "BUZZ_ACP_NO_PRESENCE"] { + assert!(is_reserved_env_key(key), "{key} should be reserved"); + let agent = map(&[(key, "0")]); + assert!(merged_user_env(&BTreeMap::new(), &agent).is_empty()); + } +} + #[test] fn reserved_keys_include_code_execution_surface() { // The agent/MCP command + args are what Buzz actually exec's. diff --git a/desktop/src-tauri/src/managed_agents/mod.rs b/desktop/src-tauri/src/managed_agents/mod.rs index be9b07cf11..772d707f27 100644 --- a/desktop/src-tauri/src/managed_agents/mod.rs +++ b/desktop/src-tauri/src/managed_agents/mod.rs @@ -1,6 +1,7 @@ mod agent_env; pub(crate) mod agent_events; pub(crate) mod agent_snapshot; +pub(crate) mod agent_snapshot_envelope; pub(crate) mod team_snapshot; pub(crate) use agent_env::{ baked_build_env, build_buzz_agent_provider_defaults, discovery_env_with_baked_floor, diff --git a/desktop/src-tauri/src/managed_agents/runtime.rs b/desktop/src-tauri/src/managed_agents/runtime.rs index 37927961ed..3173126b90 100644 --- a/desktop/src-tauri/src/managed_agents/runtime.rs +++ b/desktop/src-tauri/src/managed_agents/runtime.rs @@ -22,8 +22,7 @@ pub(crate) use path::should_use_inherited; mod metadata; pub(crate) use metadata::{ - resolve_effective_prompt_model_provider, resolve_session_title, runtime_metadata_env_vars, - SESSION_TITLE_ENV_VAR, + resolve_session_title, runtime_metadata_env_vars, SESSION_TITLE_ENV_VAR, }; mod stop; diff --git a/desktop/src-tauri/src/managed_agents/runtime/metadata.rs b/desktop/src-tauri/src/managed_agents/runtime/metadata.rs index 288ce06b0a..96ac73e347 100644 --- a/desktop/src-tauri/src/managed_agents/runtime/metadata.rs +++ b/desktop/src-tauri/src/managed_agents/runtime/metadata.rs @@ -57,32 +57,6 @@ pub(crate) fn resolve_session_title(display_name: Option<&str>, name: &str) -> O .find(|value| !value.is_empty()) } -/// Resolve effective prompt/model/provider using definition-authoritative -/// semantics for linked instances. -/// -/// Used by `agent_config.rs` to inject persona defaults into the config surface -/// before running the reader. -pub(crate) fn resolve_effective_prompt_model_provider( - persona_id: Option<&str>, - personas: &[crate::managed_agents::types::AgentDefinition], - record_prompt: Option, - record_model: Option, - record_provider: Option, -) -> (Option, Option, Option) { - match persona_id.and_then(|pid| personas.iter().find(|p| p.id == pid)) { - Some(p) => { - fn non_blank(v: Option<&str>) -> Option { - v.filter(|s| !s.trim().is_empty()).map(str::to_owned) - } - let prompt = non_blank(Some(&p.system_prompt)); - let model = non_blank(p.model.as_deref()); - let provider = non_blank(p.provider.as_deref()); - (prompt, model, provider) - } - None => (record_prompt, record_model, record_provider), - } -} - #[cfg(test)] mod tests { use super::resolve_session_title; diff --git a/desktop/src-tauri/src/mesh_llm/recovery.rs b/desktop/src-tauri/src/mesh_llm/recovery.rs index 89ca6396e9..7933fd291e 100644 --- a/desktop/src-tauri/src/mesh_llm/recovery.rs +++ b/desktop/src-tauri/src/mesh_llm/recovery.rs @@ -149,8 +149,25 @@ fn should_evict_after_probe( probe: MeshIngressProbe, consecutive: u32, ) -> bool { - urgency == MeshRecoveryUrgency::Foreground && probe == MeshIngressProbe::PortClosed - || consecutive >= DEAD_PROBE_EVICT_THRESHOLD + // Only a CLOSED port is evidence of death. A bound-but-HTTP-unresponsive + // port ("Unhealthy") is a BUSY node, not a dead one: mesh serializes all + // HTTP on the ingress — including the `/v1/models` liveness probe — behind + // in-flight inference, so a large-prompt turn on a big model leaves the + // control plane unresponsive for the whole turn (measured ~27s on a + // gemma-4-26B node) while TCP-connect keeps answering in ~0ms. Model load + // and package-layer download are unresponsive in exactly the same way. + // Evicting on any of those turns ordinary backpressure into a destructive + // whole-app restart loop, which is the regression this fixes. A genuinely + // wedged bound port cannot be distinguished from a busy one without a + // lock-free health endpoint on the ingress (tracked upstream in mesh-llm); + // until that exists we never evict a bound port and rely solely on the + // unambiguous closed-port signal. + match probe { + MeshIngressProbe::Live | MeshIngressProbe::Unhealthy => false, + MeshIngressProbe::PortClosed => { + urgency == MeshRecoveryUrgency::Foreground || consecutive >= DEAD_PROBE_EVICT_THRESHOLD + } + } } fn requires_process_restart( @@ -160,10 +177,11 @@ fn requires_process_restart( startup_in_progress || mode == crate::mesh_llm::MeshNodeMode::Serve } -/// Probe and, when justified, remove one stale runtime. A closed port is -/// decisive for a foreground agent start; watchdog and ambiguous/unhealthy -/// ports require consecutive failures to avoid restarting on a transient load -/// spike. +/// Probe and, when justified, remove one stale runtime. Only a CLOSED port is +/// treated as death: a foreground agent start evicts immediately, the watchdog +/// after a short consecutive-failure streak. A bound-but-unresponsive +/// ("Unhealthy") port is never evicted — it is a busy or still-loading node, +/// not a dead one (see `should_evict_after_probe`). pub(crate) async fn recover_stale_mesh_runtime( state: &AppState, urgency: MeshRecoveryUrgency, @@ -490,6 +508,89 @@ mod tests { )); } + #[test] + fn watchdog_closed_port_still_evicts_after_consecutive_streak() { + // A genuinely dead listener (crashed / released its port) must still be + // reclaimed — the closed-port signal is unchanged by this fix. + assert!(!should_evict_after_probe( + MeshRecoveryUrgency::Watchdog, + MeshIngressProbe::PortClosed, + 1 + )); + assert!(should_evict_after_probe( + MeshRecoveryUrgency::Watchdog, + MeshIngressProbe::PortClosed, + DEAD_PROBE_EVICT_THRESHOLD + )); + } + + #[test] + fn busy_or_loading_bound_port_is_never_evicted() { + // The regression this fixes: mesh serializes all ingress HTTP (incl. + // the `/v1/models` liveness probe) behind in-flight inference, so a + // large-prompt turn, a model load, or a layer download leaves the port + // bound-but-unresponsive ("Unhealthy"). No probe streak, and no + // urgency, may evict such a node — doing so restarts a node that is + // alive and working. + for consecutive in [1, 2, 5, 100] { + for urgency in [ + MeshRecoveryUrgency::Watchdog, + MeshRecoveryUrgency::Foreground, + ] { + assert!( + !should_evict_after_probe(urgency, MeshIngressProbe::Unhealthy, consecutive), + "a bound-but-busy port must never evict (urgency={urgency:?}, \ + consecutive={consecutive})" + ); + } + } + } + + #[test] + fn long_model_load_never_reaches_the_restart_path() { + // Pins the exact false positive this fix removes. A big model stays + // bound-but-unresponsive for MINUTES while it loads weights and + // downloads package layers, so the watchdog sees an unbroken run of + // `Unhealthy` probes. Walk ~5 minutes of watchdog passes at its 15s + // base interval and assert the eviction gate stays shut the whole way + // — for a serve node, one `true` here is a whole-app restart. + let state = MeshRecoveryState::default(); + let runtime_id = 42; + let passes = (5 * 60) / 15; + + for pass in 1..=passes { + let consecutive = state.record_dead_probe(runtime_id); + // The streak really does climb — the non-eviction below is the + // rule refusing to act, not the counter quietly resetting. + assert_eq!( + consecutive, pass, + "probe streak should keep climbing across a long load" + ); + for urgency in [ + MeshRecoveryUrgency::Watchdog, + MeshRecoveryUrgency::Foreground, + ] { + assert!( + !should_evict_after_probe(urgency, MeshIngressProbe::Unhealthy, consecutive), + "a still-loading node must never be evicted \ + (urgency={urgency:?}, minute={}, streak={consecutive})", + pass * 15 / 60 + ); + } + } + + // Sanity: the streak blew far past the threshold that used to evict, + // so the old logic WOULD have restarted this healthy loading node. + assert!( + passes >= DEAD_PROBE_EVICT_THRESHOLD, + "test must exceed the old eviction threshold to be meaningful" + ); + + // Once the load finishes and the ingress answers, the streak clears. + state.reset_probe_streak(); + assert_eq!(state.record_dead_probe(runtime_id), 1); + } + #[test] fn probe_streak_is_scoped_to_runtime_identity() { let state = MeshRecoveryState::default(); @@ -570,4 +671,39 @@ mod tests { ); assert!(!"user note: shared compute config".starts_with(MESH_REARM_ERROR_SENTINEL)); } + + // Black-box proof of the classification the eviction rule stands on. A + // mesh node busy in inference (or loading, or downloading) keeps its + // ingress TCP port accepting connections in ~0ms while HTTP does not answer + // within the probe timeout — measured directly against a gemma-4-26B node: + // a concurrent `/v1/models` took ~27s, queued behind one in-flight turn. + // This stands up exactly that shape — a listener that accepts then never + // replies — and asserts the probe reads `Unhealthy` (busy), the verdict + // `should_evict_after_probe` now refuses to evict on. If this regressed to + // `PortClosed`, a busy node would again be misread as dead and restarted. + #[tokio::test] + async fn bound_but_stalled_http_classifies_as_unhealthy_not_closed() { + let listener = tokio::net::TcpListener::bind("127.0.0.1:0") + .await + .expect("bind ephemeral port"); + let port = listener.local_addr().unwrap().port(); + // Accept and hold connections open without ever writing a response — + // the wire-level equivalent of a node serializing HTTP behind a turn. + let accept_task = tokio::spawn(async move { + let mut held = Vec::new(); + while let Ok((stream, _)) = listener.accept().await { + held.push(stream); // keep the socket open, never respond + } + }); + + let probe = probe_mesh_ingress_at(&format!("http://127.0.0.1:{port}/v1")).await; + accept_task.abort(); + + assert_eq!( + probe, + MeshIngressProbe::Unhealthy, + "a TCP-bound port that stalls HTTP (a busy/loading node) must read \ + Unhealthy, never PortClosed — the eviction fix depends on this" + ); + } } diff --git a/desktop/src-tauri/src/mouse_nav.rs b/desktop/src-tauri/src/mouse_nav.rs new file mode 100644 index 0000000000..cd729f7304 --- /dev/null +++ b/desktop/src-tauri/src/mouse_nav.rs @@ -0,0 +1,142 @@ +//! Native macOS handler for back/forward navigation inputs (mouse X1/X2 +//! buttons and horizontal swipe gestures). +//! +//! WKWebView never delivers these inputs to the web content layer, so a DOM +//! listener can't see them (Safari itself handles them natively in the app +//! layer, not in the page). This module installs an NSEvent local monitor +//! and emits a `mouse-nav` Tauri event that `useBackForwardControls` acts on +//! in the frontend. Two event shapes map to navigation: +//! +//! - `otherMouseUp` with button 3/4 — mice whose X1/X2 buttons reach the app +//! as plain mouse buttons. +//! - `swipe` with a horizontal delta — AppKit's page-swipe gesture +//! (`swipeWithEvent:`): `deltaX > 0` is back, `deltaX < 0` is forward. +//! Sent by mouse drivers that synthesize a page-swipe gesture for the +//! back/forward buttons instead of button-3/4 events (the hardware this +//! was verified on). Stock Apple trackpad and Magic Mouse swipes arrive +//! as phased scroll-wheel events instead, which this module does not +//! handle — that path (`ScrollWheel` + `trackSwipeEventWithOptions:`, +//! which also needs scroll-edge detection) is a follow-up. +//! +//! Compiled macOS-only (via `tray_menu`). Non-macOS X1/X2 behavior is left +//! to the underlying webview. + +/// Maps an `otherMouseUp` button number to a navigation direction. +/// Buttons 3 and 4 are X1 (back) and X2 (forward). +fn direction_for_button(button: isize) -> Option<&'static str> { + match button { + 3 => Some("back"), + 4 => Some("forward"), + _ => None, + } +} + +/// Maps a swipe gesture's horizontal delta to a navigation direction, +/// following the AppKit `swipeWithEvent:` convention: positive is back, +/// negative is forward. A swipe arrives as a begin/end pair and only the +/// end event carries the direction, so `deltaX == 0` maps to `None`. +fn direction_for_swipe(delta_x: f64) -> Option<&'static str> { + if delta_x > 0.0 { + Some("back") + } else if delta_x < 0.0 { + Some("forward") + } else { + None + } +} + +pub fn init(app_handle: &tauri::AppHandle) { + use block2::RcBlock; + use objc2_app_kit::{NSEvent, NSEventMask, NSEventType}; + use tauri::Emitter; + + let app = app_handle.clone(); + let block = RcBlock::new(move |event: std::ptr::NonNull| -> *mut NSEvent { + // SAFETY: the monitor hands us a valid NSEvent for the matched mask. + let ev = unsafe { event.as_ref() }; + + match ev.r#type() { + NSEventType::OtherMouseUp => { + if let Some(direction) = direction_for_button(ev.buttonNumber()) { + // Emit to the main window explicitly instead of + // broadcasting (`emit`) so navigation stays scoped if + // multi-window ever lands. "main" is the default label + // for the single configured window (see deep_link.rs). + let _ = app.emit_to("main", "mouse-nav", direction); + // Swallow the release: nothing downstream should also act + // on it. The matching press deliberately passes through: + // WKWebView never delivers X1/X2 to the page, so the + // unmatched down is inert, and swallowing presses risks + // interfering with AppKit behaviors keyed off mouse-down. + return std::ptr::null_mut(); + } + } + NSEventType::Swipe => { + if let Some(direction) = direction_for_swipe(ev.deltaX()) { + let _ = app.emit_to("main", "mouse-nav", direction); + } + // Pass swipes through: nothing else navigates on them, and + // swallowing mid-gesture events could confuse AppKit's + // gesture tracking. + } + _ => {} + } + + event.as_ptr() + }); + + // SAFETY: the block returns either null or the pointer it was given, both + // valid per the monitor contract. The returned monitor token is + // deliberately leaked: the monitor must live for the whole app lifetime. + let monitor = unsafe { + NSEvent::addLocalMonitorForEventsMatchingMask_handler( + NSEventMask::OtherMouseUp | NSEventMask::Swipe, + &block, + ) + }; + + if let Some(monitor) = monitor { + std::mem::forget(monitor); + } else { + eprintln!("buzz-desktop: mouse-nav: failed to install NSEvent monitor"); + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn button_3_is_back() { + assert_eq!(direction_for_button(3), Some("back")); + } + + #[test] + fn button_4_is_forward() { + assert_eq!(direction_for_button(4), Some("forward")); + } + + #[test] + fn other_buttons_do_not_navigate() { + for button in [0, 1, 2, 5, -1] { + assert_eq!(direction_for_button(button), None); + } + } + + #[test] + fn positive_swipe_delta_is_back() { + assert_eq!(direction_for_swipe(1.0), Some("back")); + assert_eq!(direction_for_swipe(0.5), Some("back")); + } + + #[test] + fn negative_swipe_delta_is_forward() { + assert_eq!(direction_for_swipe(-1.0), Some("forward")); + assert_eq!(direction_for_swipe(-0.5), Some("forward")); + } + + #[test] + fn zero_delta_swipe_begin_event_is_ignored() { + assert_eq!(direction_for_swipe(0.0), None); + } +} diff --git a/desktop/src-tauri/src/tray_menu.rs b/desktop/src-tauri/src/tray_menu.rs index d733cd1f13..6dcef0ecf7 100644 --- a/desktop/src-tauri/src/tray_menu.rs +++ b/desktop/src-tauri/src/tray_menu.rs @@ -3,6 +3,11 @@ //! The webview owns the live agent-turn state. It sends the small display //! projection here so the native menu can remain useful while Buzz is hidden. +// Mouse back/forward (X1/X2 buttons and swipe) is also macOS-only native I/O; +// group it here so both platform-layer init paths share one call site in lib.rs. +#[path = "mouse_nav.rs"] +pub(crate) mod mouse_nav; + use std::{ sync::{Mutex, OnceLock}, time::{Duration, Instant}, @@ -488,6 +493,7 @@ pub fn init(app: &AppHandle) -> tauri::Result<()> { if let Err(error) = apply_activity_presentation(&tray, activities, recent_activities) { eprintln!("buzz-desktop: failed to apply tray menu presentation: {error}"); } + mouse_nav::init(app); Ok(()) } diff --git a/desktop/src-tauri/tauri.conf.json b/desktop/src-tauri/tauri.conf.json index 1ff8bd20ef..4bda55fd09 100644 --- a/desktop/src-tauri/tauri.conf.json +++ b/desktop/src-tauri/tauri.conf.json @@ -1,7 +1,7 @@ { "$schema": "https://schema.tauri.app/config/2", "productName": "Buzz", - "version": "0.5.3", + "version": "0.5.4", "identifier": "xyz.block.buzz.app", "build": { "beforeDevCommand": { @@ -55,6 +55,7 @@ "externalBin": [ "binaries/buzz-acp", "binaries/buzz-agent", + "binaries/buzz-backend-kubernetes", "binaries/buzz-dev-mcp", "binaries/git-credential-nostr", "binaries/buzz" diff --git a/desktop/src-tauri/tauri.windows.conf.json b/desktop/src-tauri/tauri.windows.conf.json new file mode 100644 index 0000000000..abf09b5bfd --- /dev/null +++ b/desktop/src-tauri/tauri.windows.conf.json @@ -0,0 +1,11 @@ +{ + "bundle": { + "externalBin": [ + "binaries/buzz-acp", + "binaries/buzz-agent", + "binaries/buzz-dev-mcp", + "binaries/git-credential-nostr", + "binaries/buzz" + ] + } +} diff --git a/desktop/src/app/AppShell.helpers.test.mjs b/desktop/src/app/AppShell.helpers.test.mjs index 73505d1798..fa06232923 100644 --- a/desktop/src/app/AppShell.helpers.test.mjs +++ b/desktop/src/app/AppShell.helpers.test.mjs @@ -1,7 +1,10 @@ import assert from "node:assert/strict"; import test from "node:test"; -import { shouldBounceForChannelNotification } from "./AppShell.helpers.ts"; +import { + markAllReadSources, + shouldBounceForChannelNotification, +} from "./AppShell.helpers.ts"; test("shouldBounceForChannelNotification_allowsTopLevelChannelMessages", () => { assert.equal(shouldBounceForChannelNotification([["h", "channel"]]), true); @@ -27,3 +30,43 @@ test("shouldBounceForChannelNotification_allowsBroadcastReplies", () => { true, ); }); + +test("markAllReadSources clears Inbox overrides and active thread activity", () => { + const calls = []; + + markAllReadSources({ + activeChannelId: "active-channel", + channelActivityItems: [ + { channelId: "another-channel", createdAt: 100 }, + { channelId: "active-channel", createdAt: 200 }, + { channelId: "active-channel", createdAt: 300 }, + ], + unreadFeedItemIds: new Set(["first-inbox-item", "second-inbox-item"]), + undoUnreadFeedItem: (itemId) => calls.push(`inbox:${itemId}`), + markAllChannelReadMarkers: () => calls.push("channels"), + markActiveChannelRead: (channelId, createdAt) => + calls.push(`active:${channelId}:${createdAt}`), + }); + + assert.deepEqual(calls, [ + "inbox:first-inbox-item", + "inbox:second-inbox-item", + "channels", + "active:active-channel:300", + ]); +}); + +test("markAllReadSources skips the active marker without projected activity", () => { + const calls = []; + + markAllReadSources({ + activeChannelId: "active-channel", + channelActivityItems: [], + unreadFeedItemIds: new Set(), + undoUnreadFeedItem: () => calls.push("inbox"), + markAllChannelReadMarkers: () => calls.push("channels"), + markActiveChannelRead: () => calls.push("active"), + }); + + assert.deepEqual(calls, ["channels"]); +}); diff --git a/desktop/src/app/AppShell.helpers.ts b/desktop/src/app/AppShell.helpers.ts index b0ce894931..dd6b9195e8 100644 --- a/desktop/src/app/AppShell.helpers.ts +++ b/desktop/src/app/AppShell.helpers.ts @@ -86,6 +86,41 @@ export function shouldBounceForChannelNotification(tags: string[][]): boolean { return !isThreadReply(tags); } +export function markAllReadSources({ + activeChannelId, + channelActivityItems, + markAllChannelReadMarkers, + markActiveChannelRead, + undoUnreadFeedItem, + unreadFeedItemIds, +}: { + activeChannelId: string | null; + channelActivityItems: ReadonlyArray<{ + channelId: string | null; + createdAt: number; + }>; + markAllChannelReadMarkers: () => void; + markActiveChannelRead: (channelId: string, createdAt: number) => void; + undoUnreadFeedItem: (itemId: string) => void; + unreadFeedItemIds: ReadonlySet; +}) { + for (const itemId of unreadFeedItemIds) { + undoUnreadFeedItem(itemId); + } + markAllChannelReadMarkers(); + + if (!activeChannelId) return; + + let latestActivityAt: number | null = null; + for (const item of channelActivityItems) { + if (item.channelId !== activeChannelId) continue; + latestActivityAt = Math.max(latestActivityAt ?? 0, item.createdAt); + } + if (latestActivityAt !== null) { + markActiveChannelRead(activeChannelId, latestActivityAt); + } +} + export function toSearchHit( target: DesktopNotificationTarget, ): SearchHit | null { diff --git a/desktop/src/app/AppShell.tsx b/desktop/src/app/AppShell.tsx index 4eb0a42bbe..46e06b2f9f 100644 --- a/desktop/src/app/AppShell.tsx +++ b/desktop/src/app/AppShell.tsx @@ -1,7 +1,7 @@ import * as React from "react"; import { useQueryClient } from "@tanstack/react-query"; import { Outlet, useLocation } from "@tanstack/react-router"; -import { deriveShellRoute } from "@/app/AppShell.helpers"; +import { deriveShellRoute, markAllReadSources } from "@/app/AppShell.helpers"; import { AppShellProvider } from "@/app/AppShellContext"; import * as BuzzTheme from "@/app/BuzzThemeSurfaces"; import { AppShellOverlays } from "@/app/AppShellOverlays"; @@ -15,7 +15,7 @@ import { useMarkAsReadShortcuts } from "@/app/useMarkAsReadShortcuts"; import { useSettingsShortcuts } from "@/app/useSettingsShortcuts"; import { useAppShellDesktopNotifications } from "@/app/useAppShellDesktopNotifications"; import { useAppShellLifecycleEffects } from "@/app/useAppShellLifecycleEffects"; -import { useThreadActivityFeedItems } from "@/app/useThreadActivityFeedItems"; +import { useChannelActivityProjection } from "@/app/useChannelActivityProjection"; import { useTauriWindowDrag } from "@/app/useTauriWindowDrag"; import { useWebviewZoomShortcuts } from "@/app/useWebviewZoomShortcuts"; import { @@ -26,7 +26,6 @@ import { useOpenDmMutation, } from "@/features/channels/hooks"; import { useUnreadChannels } from "@/features/channels/useUnreadChannels"; -import { msgContextKey } from "@/features/channels/readState/readStateFormat"; import { useMembershipNotifications } from "@/features/channels/useMembershipNotifications"; import { useFeedItemState } from "@/features/home/useFeedItemState"; import { useThreadFollows } from "@/features/messages/lib/useThreadFollows"; @@ -324,10 +323,12 @@ export function AppShell() { } = useThreadFollows(identityQuery.data?.pubkey); const { - markAllChannelsRead, + markAllChannelsRead: markAllChannelReadMarkers, markChannelRead, markChannelUnread, + clearChannelUnreadSource, unreadChannelIds, + topLevelUnreadChannelIds, unreadChannelCounts, highPriorityUnreadChannelIds, unreadChannelNotificationCount, @@ -338,6 +339,7 @@ export function AppShell() { participatedRootIds, authoredRootIds, mentionedRootIds, + recordThreadInteraction, threadActivityItems, mutedRootIds, muteThread, @@ -356,55 +358,45 @@ export function AppShell() { followedRootIds, }); - const getThreadReadAt = React.useCallback( - (rootId: string, channelId?: string | null) => { - const threadReadAt = getOwnReadAt(`thread:${rootId}`); - if (!channelId) { - return threadReadAt; - } - - const channelReadAt = getChannelReadAt(channelId); - if (threadReadAt === null) { - return channelReadAt; - } - if (channelReadAt === null) { - return threadReadAt; - } - return Math.max(threadReadAt, channelReadAt); - }, - [getChannelReadAt, getOwnReadAt], - ); - - const markThreadRead = React.useCallback( - (rootId: string, timestamp: number) => { - markChannelRead( - `thread:${rootId}`, - new Date(timestamp * 1_000).toISOString(), - ); - }, - [markChannelRead], - ); - - // Per-message read frontier (LP4 v3): effective(msg:) folds through the - // channel, so a channel-read clears messages older than the top-level frontier. - const getMessageReadAt = React.useCallback( - (messageId: string) => getChannelReadAt(msgContextKey(messageId)), - [getChannelReadAt], - ); - const markMessageRead = React.useCallback( - (messageId: string, timestamp: number) => - markChannelRead( - msgContextKey(messageId), - new Date(timestamp * 1_000).toISOString(), - ), - [markChannelRead], - ); - const threadActivityFeedItems = useThreadActivityFeedItems( + const { + getThreadReadAt, + markThreadRead, + getMessageReadAt, + getChannelActivityItemReadAt, + markMessageRead, + threadActivityFeedItems, + locallyUnreadFeedItems, + unreadThreadFeedItems, + unreadThreadChannelIds, + } = useChannelActivityProjection({ + channels, + feed: homeFeedQuery.data?.feed, + unreadFeedItemIds: feedItemState.unreadSet, + getChannelReadAt, + getOwnReadAt, + markChannelRead, + readStateVersion, threadActivityItems, mutedRootIds, - channels, - ); - + }); + const markAllChannelsRead = React.useCallback(() => { + markAllReadSources({ + activeChannelId: activeChannel?.id ?? null, + channelActivityItems: unreadThreadFeedItems, + markAllChannelReadMarkers, + markActiveChannelRead: (channelId, createdAt) => + markChannelRead(channelId, new Date(createdAt * 1_000).toISOString()), + undoUnreadFeedItem: feedItemState.undoUnread, + unreadFeedItemIds: feedItemState.unreadSet, + }); + }, [ + activeChannel?.id, + feedItemState.undoUnread, + feedItemState.unreadSet, + markAllChannelReadMarkers, + markChannelRead, + unreadThreadFeedItems, + ]); // Badge count consumes the shared NIP-RS read-state from useUnreadChannels. const { homeBadgeCount, homeBadgeCountExcludingHighPriority } = useHomeFeedNotificationState( @@ -709,6 +701,7 @@ export function AppShell() { markAllChannelsRead, markChannelRead, markChannelUnread, + clearChannelUnreadSource, openBrowseChannels: handleOpenBrowseChannels, openCreateChannel: handleOpenCreateChannel, openChannelManagement: (channelId?: string) => { @@ -721,6 +714,7 @@ export function AppShell() { getThreadReadAt, markThreadRead, getMessageReadAt, + getChannelActivityItemReadAt, markMessageRead, readStateVersion, setContextParentResolver, @@ -728,9 +722,15 @@ export function AppShell() { unfollowThread: handleUnfollowThread, isFollowingThread, isNotifiedForThread, + recordThreadInteraction, isThreadMuted: (rootId) => mutedRootIds.has(rootId), threadActivityItems, threadActivityFeedItems, + locallyUnreadFeedItems, + unreadThreadFeedItems, + unreadThreadChannelIds, + topLevelUnreadChannelIds, + hasSidebarUnreadProjections: true, feedItemState, onOpenSettings: handleOpenSettings, }} diff --git a/desktop/src/app/AppShellContext.tsx b/desktop/src/app/AppShellContext.tsx index 4a64de0cb2..b909436178 100644 --- a/desktop/src/app/AppShellContext.tsx +++ b/desktop/src/app/AppShellContext.tsx @@ -1,4 +1,5 @@ import * as React from "react"; +import type { ForcedUnreadSource } from "@/features/channels/forcedUnreadStore"; import type { ContextParentResolver } from "@/features/channels/readState/readStateManager"; import type { ThreadActivityItem } from "@/features/channels/useUnreadChannels"; import type { FeedItemState } from "@/features/home/useFeedItemState"; @@ -12,9 +13,16 @@ type AppShellContextValue = { markChannelRead: ( channelId: string, readAt: string | null | undefined, - options?: { topLevelOnly?: boolean }, + options?: { + preserveForcedUnread?: boolean; + topLevelOnly?: boolean; + }, + ) => void; + markChannelUnread: (channelId: string, source?: ForcedUnreadSource) => void; + clearChannelUnreadSource: ( + channelId: string, + source: ForcedUnreadSource, ) => void; - markChannelUnread: (channelId: string) => void; openBrowseChannels: () => void; openCreateChannel: () => void; openChannelManagement: (channelId?: string) => void; @@ -31,6 +39,11 @@ type AppShellContextValue = { // read. Uses `msg:` context keys folded through the active channel by the // parent resolver (LP4 v3 per-message badge model). getMessageReadAt: (messageId: string) => number | null; + // Read frontier for a channel-activity item, scoped to that item's own + // message and channel rather than the currently mounted channel resolver. + getChannelActivityItemReadAt: ( + item: Pick, + ) => number | null; // Advance a single message's read marker to the given unix-seconds timestamp. markMessageRead: (messageId: string, timestamp: number) => void; // Bump-counter that invalidates whenever the read marker changes. Include @@ -43,9 +56,25 @@ type AppShellContextValue = { unfollowThread: (rootId: string) => void; isFollowingThread: (rootId: string) => boolean; isNotifiedForThread: (rootId: string) => boolean; + recordThreadInteraction: (rootId: string) => void; isThreadMuted: (rootId: string) => boolean; threadActivityItems: ThreadActivityItem[]; threadActivityFeedItems: FeedItem[]; + // Home-feed items explicitly reopened from Inbox. Kept separate from live + // thread activity so older rows can be projected into channel hover cards + // without duplicating the Home feed itself. + locallyUnreadFeedItems: FeedItem[]; + // Thread rows that remain unread until their own message/thread marker is + // advanced. Unlike the broad channel unread set, this includes the active + // channel so simply landing in it does not hide the wayfinding signal. + unreadThreadFeedItems: FeedItem[]; + unreadThreadChannelIds: ReadonlySet; + // Ordinary unread channel-level activity. Sidebar rows use this for text + // emphasis only; thread activity owns the dot. + topLevelUnreadChannelIds: ReadonlySet; + // Lets isolated component tests retain the legacy hasUnread fallback while + // the mounted shell uses the split projections above. + hasSidebarUnreadProjections: boolean; feedItemState: FeedItemState; // Open the Settings panel at the given section. Available on all surfaces // that render under AppShell (channel, home, projects, pulse, agents). @@ -57,6 +86,7 @@ const AppShellContext = React.createContext({ markAllChannelsRead: () => {}, markChannelRead: () => {}, markChannelUnread: () => {}, + clearChannelUnreadSource: () => {}, openBrowseChannels: () => {}, openCreateChannel: () => {}, openChannelManagement: () => {}, @@ -64,6 +94,7 @@ const AppShellContext = React.createContext({ getThreadReadAt: () => null, markThreadRead: () => {}, getMessageReadAt: () => null, + getChannelActivityItemReadAt: () => null, markMessageRead: () => {}, readStateVersion: 0, setContextParentResolver: () => {}, @@ -71,9 +102,15 @@ const AppShellContext = React.createContext({ unfollowThread: () => {}, isFollowingThread: () => false, isNotifiedForThread: () => false, + recordThreadInteraction: () => {}, isThreadMuted: () => false, threadActivityItems: [], threadActivityFeedItems: [], + locallyUnreadFeedItems: [], + unreadThreadFeedItems: [], + unreadThreadChannelIds: EMPTY_SET, + topLevelUnreadChannelIds: EMPTY_SET, + hasSidebarUnreadProjections: false, feedItemState: { doneSet: EMPTY_SET, markDone: () => {}, diff --git a/desktop/src/app/navigation/backForwardChords.test.mjs b/desktop/src/app/navigation/backForwardChords.test.mjs new file mode 100644 index 0000000000..cb60203d04 --- /dev/null +++ b/desktop/src/app/navigation/backForwardChords.test.mjs @@ -0,0 +1,139 @@ +import assert from "node:assert/strict"; +import test from "node:test"; + +import { matchBackForwardChord } from "./backForwardChords.ts"; + +function chord(overrides = {}) { + return { + altKey: false, + code: "", + ctrlKey: false, + key: "", + metaKey: false, + shiftKey: false, + ...overrides, + }; +} + +// ── macOS: ⌘[ / ⌘] ─────────────────────────────────────────────────────────── + +test("mac: ⌘[ matches back", () => { + assert.equal( + matchBackForwardChord(chord({ key: "[", metaKey: true }), true), + "back", + ); +}); + +test("mac: ⌘] matches forward", () => { + assert.equal( + matchBackForwardChord(chord({ key: "]", metaKey: true }), true), + "forward", + ); +}); + +test("mac: matches by code for non-US layouts", () => { + assert.equal( + matchBackForwardChord( + chord({ code: "BracketLeft", key: "Dead", metaKey: true }), + true, + ), + "back", + ); + assert.equal( + matchBackForwardChord( + chord({ code: "BracketRight", key: "Dead", metaKey: true }), + true, + ), + "forward", + ); +}); + +test("mac: requires meta", () => { + assert.equal(matchBackForwardChord(chord({ key: "[" }), true), null); +}); + +test("mac: rejects extra modifiers", () => { + for (const extra of [ + { altKey: true }, + { ctrlKey: true }, + { shiftKey: true }, + ]) { + assert.equal( + matchBackForwardChord(chord({ key: "[", metaKey: true, ...extra }), true), + null, + ); + } +}); + +test("mac: Alt+arrows do not match (that is the win/linux chord)", () => { + assert.equal( + matchBackForwardChord(chord({ altKey: true, key: "ArrowLeft" }), true), + null, + ); +}); + +test("mac: ⌘←/⌘→ never match — they are line start/end in text editing", () => { + // Deliberately unbound: editable targets must keep receiving ⌘←/⌘→ so + // line-start/line-end editing still works. Only ⌘[ / ⌘] navigate. + assert.equal( + matchBackForwardChord(chord({ key: "ArrowLeft", metaKey: true }), true), + null, + ); + assert.equal( + matchBackForwardChord(chord({ key: "ArrowRight", metaKey: true }), true), + null, + ); +}); + +// ── Windows/Linux: Alt+← / Alt+→ ───────────────────────────────────────────── + +test("win/linux: Alt+ArrowLeft matches back", () => { + assert.equal( + matchBackForwardChord(chord({ altKey: true, key: "ArrowLeft" }), false), + "back", + ); +}); + +test("win/linux: Alt+ArrowRight matches forward", () => { + assert.equal( + matchBackForwardChord(chord({ altKey: true, key: "ArrowRight" }), false), + "forward", + ); +}); + +test("win/linux: requires alt", () => { + assert.equal(matchBackForwardChord(chord({ key: "ArrowLeft" }), false), null); +}); + +test("win/linux: rejects extra modifiers", () => { + for (const extra of [ + { ctrlKey: true }, + { metaKey: true }, + { shiftKey: true }, + ]) { + assert.equal( + matchBackForwardChord( + chord({ altKey: true, key: "ArrowLeft", ...extra }), + false, + ), + null, + ); + } +}); + +test("win/linux: ⌘[ does not match (that is the mac chord)", () => { + assert.equal( + matchBackForwardChord(chord({ key: "[", metaKey: true }), false), + null, + ); +}); + +// ── Non-chord keys never match ──────────────────────────────────────────────── + +test("plain bracket / arrow keys without the platform modifier never match", () => { + for (const isMac of [true, false]) { + for (const key of ["[", "]", "ArrowLeft", "ArrowRight", "a", "Enter"]) { + assert.equal(matchBackForwardChord(chord({ key }), isMac), null); + } + } +}); diff --git a/desktop/src/app/navigation/backForwardChords.ts b/desktop/src/app/navigation/backForwardChords.ts new file mode 100644 index 0000000000..ac81f52ea7 --- /dev/null +++ b/desktop/src/app/navigation/backForwardChords.ts @@ -0,0 +1,51 @@ +/** + * Global back/forward navigation chords. + * + * macOS: ⌘[ / ⌘] — matching Safari, Chrome, Finder, and Slack. + * Windows/Linux: Alt+← / Alt+→ — matching browsers and Slack. + * + * Kept pure (no DOM access) so chord matching can be unit tested; the + * window listener wiring lives in `useBackForwardControls`. + */ + +export type BackForwardDirection = "back" | "forward"; + +export type BackForwardChordEvent = Pick< + KeyboardEvent, + "altKey" | "code" | "ctrlKey" | "key" | "metaKey" | "shiftKey" +>; + +export function matchBackForwardChord( + event: BackForwardChordEvent, + isMac: boolean, +): BackForwardDirection | null { + if (isMac) { + if (!event.metaKey || event.ctrlKey || event.altKey || event.shiftKey) { + return null; + } + + if (event.key === "[" || event.code === "BracketLeft") { + return "back"; + } + + if (event.key === "]" || event.code === "BracketRight") { + return "forward"; + } + + return null; + } + + if (!event.altKey || event.metaKey || event.ctrlKey || event.shiftKey) { + return null; + } + + if (event.key === "ArrowLeft") { + return "back"; + } + + if (event.key === "ArrowRight") { + return "forward"; + } + + return null; +} diff --git a/desktop/src/app/navigation/useBackForwardControls.ts b/desktop/src/app/navigation/useBackForwardControls.ts index 7f7d84f6d7..e5513247d5 100644 --- a/desktop/src/app/navigation/useBackForwardControls.ts +++ b/desktop/src/app/navigation/useBackForwardControls.ts @@ -4,7 +4,10 @@ import { useRouter, useRouterState, } from "@tanstack/react-router"; +import { isTauri } from "@tauri-apps/api/core"; +import { listen } from "@tauri-apps/api/event"; +import { matchBackForwardChord } from "@/app/navigation/backForwardChords"; import { isMacPlatform } from "@/shared/lib/platform"; import { trimMapToSize } from "@/shared/lib/trimMapToSize"; @@ -14,19 +17,6 @@ type RouterHistoryState = { key?: string; }; -function isEditableTarget(target: EventTarget | null): boolean { - if (!(target instanceof HTMLElement)) { - return false; - } - - return ( - target.isContentEditable || - target.closest( - 'input, textarea, select, [contenteditable=""], [contenteditable="true"]', - ) !== null - ); -} - export function useBackForwardControls() { const router = useRouter(); const canGoBack = useCanGoBack(); @@ -81,46 +71,38 @@ export function useBackForwardControls() { }, [canGoForward, router.history]); const handleKeyDown = React.useEffectEvent((event: KeyboardEvent) => { - if (isEditableTarget(event.target)) { - return; - } + // Note: the chords deliberately fire even when focus is inside an + // editable element. The composer autofocuses on every channel switch + // (`useComposerAutofocus`), so in steady state focus almost always + // lives in a contenteditable — an editable-target guard here made the + // shortcuts effectively dead (#3775). Safe because neither ⌘[ / ⌘] + // (macOS) nor Alt+←/→ (Windows/Linux) carry text-editing semantics, + // and the TipTap editor binds no conflicting shortcuts. + const direction = matchBackForwardChord(event, isMacPlatform()); - const isMac = isMacPlatform(); - const isBackShortcut = isMac - ? event.metaKey && - !event.ctrlKey && - !event.altKey && - !event.shiftKey && - (event.key === "[" || event.code === "BracketLeft") - : event.altKey && - !event.metaKey && - !event.ctrlKey && - !event.shiftKey && - event.key === "ArrowLeft"; - const isForwardShortcut = isMac - ? event.metaKey && - !event.ctrlKey && - !event.altKey && - !event.shiftKey && - (event.key === "]" || event.code === "BracketRight") - : event.altKey && - !event.metaKey && - !event.ctrlKey && - !event.shiftKey && - event.key === "ArrowRight"; - - if (isBackShortcut) { + if (direction === "back") { event.preventDefault(); goBack(); return; } - if (isForwardShortcut) { + if (direction === "forward") { event.preventDefault(); goForward(); } }); + const handleMouseNav = React.useEffectEvent((direction: string) => { + if (direction === "back") { + goBack(); + return; + } + + if (direction === "forward") { + goForward(); + } + }); + React.useEffect(() => { window.addEventListener("keydown", handleKeyDown); return () => { @@ -128,6 +110,23 @@ export function useBackForwardControls() { }; }, []); + // macOS: WKWebView never delivers X1/X2 button events or horizontal + // swipe gestures to the DOM, so the native layer catches them + // (`mouse_nav.rs`) and forwards them as a Tauri event. + React.useEffect(() => { + if (!isTauri()) { + return; + } + + const unlistenPromise = listen("mouse-nav", (event) => { + handleMouseNav(event.payload); + }); + + return () => { + void unlistenPromise.then((unlisten) => unlisten()); + }; + }, []); + return { canGoBack, canGoForward, diff --git a/desktop/src/app/useChannelActivityProjection.test.mjs b/desktop/src/app/useChannelActivityProjection.test.mjs new file mode 100644 index 0000000000..6b54f32246 --- /dev/null +++ b/desktop/src/app/useChannelActivityProjection.test.mjs @@ -0,0 +1,30 @@ +import assert from "node:assert/strict"; +import test from "node:test"; + +import { resolveChannelActivityFeedItemReadAt } from "./useChannelActivityProjection.ts"; + +test("channel activity read state folds the item's own message and channel markers", () => { + const markers = new Map([ + ["msg:reply-general", 100], + ["general", 200], + ["random", 500], + ]); + + assert.equal( + resolveChannelActivityFeedItemReadAt( + { id: "reply-general", channelId: "general" }, + (contextId) => markers.get(contextId) ?? null, + ), + 200, + ); +}); + +test("channel activity read state honors a channel marker without a message marker", () => { + assert.equal( + resolveChannelActivityFeedItemReadAt( + { id: "reply-general", channelId: "general" }, + (contextId) => (contextId === "general" ? 300 : null), + ), + 300, + ); +}); diff --git a/desktop/src/app/useChannelActivityProjection.ts b/desktop/src/app/useChannelActivityProjection.ts new file mode 100644 index 0000000000..b1961318ce --- /dev/null +++ b/desktop/src/app/useChannelActivityProjection.ts @@ -0,0 +1,145 @@ +import * as React from "react"; + +import { useThreadActivityFeedItems } from "@/app/useThreadActivityFeedItems"; +import { + maxReadAt, + msgContextKey, +} from "@/features/channels/readState/readStateFormat"; +import type { ThreadActivityItem } from "@/features/channels/useUnreadChannels"; +import { isThreadReply } from "@/features/messages/lib/threading"; +import type { Channel, FeedItem, HomeFeed } from "@/shared/api/types"; + +type ReadTimestamp = (contextKey: string) => number | null; +type MarkChannelRead = ( + contextKey: string, + readAt: string | null | undefined, + options?: { topLevelOnly?: boolean }, +) => void; + +type UseChannelActivityProjectionOptions = { + channels: Channel[]; + feed: HomeFeed | undefined; + unreadFeedItemIds: ReadonlySet; + getChannelReadAt: ReadTimestamp; + getOwnReadAt: ReadTimestamp; + markChannelRead: MarkChannelRead; + readStateVersion: number; + threadActivityItems: ThreadActivityItem[]; + mutedRootIds: ReadonlySet; +}; + +export function resolveChannelActivityFeedItemReadAt( + item: Pick, + getOwnReadAt: ReadTimestamp, +): number | null { + return maxReadAt( + getOwnReadAt(msgContextKey(item.id)), + item.channelId ? getOwnReadAt(item.channelId) : null, + ); +} + +export function useChannelActivityProjection({ + channels, + feed, + unreadFeedItemIds, + getChannelReadAt, + getOwnReadAt, + markChannelRead, + readStateVersion, + threadActivityItems, + mutedRootIds, +}: UseChannelActivityProjectionOptions) { + const getThreadReadAt = React.useCallback( + (rootId: string, channelId?: string | null) => { + const threadReadAt = getOwnReadAt(`thread:${rootId}`); + if (!channelId) return threadReadAt; + + const channelReadAt = getChannelReadAt(channelId); + if (threadReadAt === null) return channelReadAt; + if (channelReadAt === null) return threadReadAt; + return Math.max(threadReadAt, channelReadAt); + }, + [getChannelReadAt, getOwnReadAt], + ); + const markThreadRead = React.useCallback( + (rootId: string, timestamp: number) => + markChannelRead( + `thread:${rootId}`, + new Date(timestamp * 1_000).toISOString(), + ), + [markChannelRead], + ); + const getMessageReadAt = React.useCallback( + (messageId: string) => getChannelReadAt(msgContextKey(messageId)), + [getChannelReadAt], + ); + const getChannelActivityItemReadAt = React.useCallback( + (item: Pick) => + resolveChannelActivityFeedItemReadAt(item, getOwnReadAt), + [getOwnReadAt], + ); + const markMessageRead = React.useCallback( + (messageId: string, timestamp: number) => + markChannelRead( + msgContextKey(messageId), + new Date(timestamp * 1_000).toISOString(), + ), + [markChannelRead], + ); + const threadActivityFeedItems = useThreadActivityFeedItems( + threadActivityItems, + mutedRootIds, + channels, + ); + const locallyUnreadFeedItems = React.useMemo(() => { + if (!feed || unreadFeedItemIds.size === 0) return []; + return [ + ...feed.mentions, + ...feed.needsAction, + ...feed.activity, + ...feed.agentActivity, + ].filter((item) => unreadFeedItemIds.has(item.id)); + }, [feed, unreadFeedItemIds]); + const unreadThreadFeedItems = React.useMemo(() => { + void readStateVersion; + const candidatesById = new Map( + threadActivityFeedItems.map((item) => [item.id, item]), + ); + for (const item of locallyUnreadFeedItems) + candidatesById.set(item.id, item); + + return [...candidatesById.values()].filter( + (item) => + isThreadReply(item.tags) && + (unreadFeedItemIds.has(item.id) || + item.createdAt > (getChannelActivityItemReadAt(item) ?? 0)), + ); + }, [ + getChannelActivityItemReadAt, + locallyUnreadFeedItems, + readStateVersion, + threadActivityFeedItems, + unreadFeedItemIds, + ]); + const unreadThreadChannelIds = React.useMemo( + () => + new Set( + unreadThreadFeedItems.flatMap((item) => + item.channelId ? [item.channelId] : [], + ), + ) as ReadonlySet, + [unreadThreadFeedItems], + ); + + return { + getThreadReadAt, + markThreadRead, + getMessageReadAt, + getChannelActivityItemReadAt, + markMessageRead, + threadActivityFeedItems, + locallyUnreadFeedItems, + unreadThreadFeedItems, + unreadThreadChannelIds, + }; +} diff --git a/desktop/src/features/agents/cardMintStore.test.mjs b/desktop/src/features/agents/cardMintStore.test.mjs new file mode 100644 index 0000000000..6c2f4522ab --- /dev/null +++ b/desktop/src/features/agents/cardMintStore.test.mjs @@ -0,0 +1,219 @@ +import assert from "node:assert/strict"; +import { beforeEach, describe, it } from "node:test"; + +// cardMintStore drives the non-blocking mint flow: the dialog dispatches a +// job and closes; the composer chip, completion toast, and viewer all read +// this store. These tests exercise the job lifecycle with an injected mintFn +// (never the real Tauri command). +// +// The store imports sonner (toast); calling toast outside a mounted +// only queues, so no DOM is required. + +import { + dismissCardMintJob, + getCardGalleryOpen, + getCardMintJobs, + getCardViewerState, + closeCardViewer, + openCardViewer, + resetCardMintStore, + runCardMintJob, + setCardGalleryOpen, + subscribeCardMintStore, + viewMintedCardJob, +} from "./cardMintStore.ts"; + +const CARD = { + cardPngBase64: "aGVsbG8=", + fileName: "eva.agent.png", + designerNotes: "notes", + locked: false, + memoryLevel: "none", +}; + +const INPUT = { agentId: "agent-1", agentName: "Eva" }; + +describe("cardMintStore", () => { + beforeEach(() => { + resetCardMintStore(); + }); + + it("forwards the mint input — including memoryLevel — to mintFn", async () => { + const seen = []; + await runCardMintJob( + { ...INPUT, styleNotes: "stormy", lock: true, memoryLevel: "core" }, + (...args) => { + seen.push(args); + return Promise.resolve({ ...CARD, memoryLevel: "core" }); + }, + ); + assert.deepEqual(seen, [["agent-1", "stormy", true, "core"]]); + + // Omitted memoryLevel stays undefined so Rust applies its "none" default. + await runCardMintJob(INPUT, (...args) => { + seen.push(args); + return Promise.resolve(CARD); + }); + assert.deepEqual(seen[1], ["agent-1", undefined, undefined, undefined]); + }); + + it("tracks a successful mint through minting → done", async () => { + let resolveMint; + const pending = new Promise((resolve) => { + resolveMint = resolve; + }); + const run = runCardMintJob(INPUT, () => pending); + + let jobs = getCardMintJobs(); + assert.equal(jobs.length, 1); + assert.equal(jobs[0].phase, "minting"); + assert.equal(jobs[0].input.agentName, "Eva"); + assert.equal(jobs[0].card, null); + + resolveMint(CARD); + await run; + + jobs = getCardMintJobs(); + assert.equal(jobs.length, 1); + assert.equal(jobs[0].phase, "done"); + assert.deepEqual(jobs[0].card, CARD); + assert.equal(jobs[0].error, null); + }); + + it("records the error message on a failed mint", async () => { + await runCardMintJob(INPUT, () => Promise.reject(new Error("boom"))); + const jobs = getCardMintJobs(); + assert.equal(jobs.length, 1); + assert.equal(jobs[0].phase, "error"); + assert.equal(jobs[0].error, "boom"); + assert.equal(jobs[0].card, null); + }); + + it("strips the NO_OPENAI_KEY wire prefix from mint errors", async () => { + await runCardMintJob(INPUT, () => + Promise.reject(new Error("NO_OPENAI_KEY: No OPENAI_API_KEY found.")), + ); + assert.equal(getCardMintJobs()[0].error, "No OPENAI_API_KEY found."); + }); + + it("replaces a 401 HTTP error with an actionable update-key message", async () => { + await runCardMintJob(INPUT, () => + Promise.reject( + new Error( + "Card mint failed (HTTP 401 Unauthorized): Incorrect API key provided: sk-proj-***", + ), + ), + ); + const { error } = getCardMintJobs()[0]; + assert.ok( + error?.includes("invalid or expired"), + `expected 'invalid or expired' in: ${error}`, + ); + assert.ok( + error?.includes("Update API key"), + `expected 'Update API key' in: ${error}`, + ); + }); + + it("replaces an 'Incorrect API key' error without an HTTP status code", async () => { + await runCardMintJob(INPUT, () => + Promise.reject(new Error("Incorrect API key provided: sk-proj-***")), + ); + const { error } = getCardMintJobs()[0]; + assert.ok( + error?.includes("invalid or expired"), + `expected 'invalid or expired' in: ${error}`, + ); + }); + + it("does not apply the 401 branch to generic non-auth errors", async () => { + await runCardMintJob(INPUT, () => + Promise.reject(new Error("Connection timeout")), + ); + assert.equal(getCardMintJobs()[0].error, "Connection timeout"); + }); + + it("does not rewrite avatar fetch 401 as an API key error", async () => { + // Avatar fetch failures have a different error prefix — rewriting them + // would send the user down a path that cannot fix the avatar failure. + const avatarError = "Avatar fetch failed: HTTP 401 Unauthorized"; + await runCardMintJob(INPUT, () => Promise.reject(new Error(avatarError))); + assert.equal( + getCardMintJobs()[0].error, + avatarError, + "avatar 401 must pass through unchanged", + ); + }); + + it("viewMintedCardJob moves a done job into the viewer and clears the chip", async () => { + await runCardMintJob(INPUT, () => Promise.resolve(CARD)); + const jobId = getCardMintJobs()[0].jobId; + + viewMintedCardJob(jobId); + + assert.equal(getCardMintJobs().length, 0); + const viewer = getCardViewerState(); + assert.ok(viewer); + assert.equal(viewer.agentName, "Eva"); + assert.deepEqual(viewer.card, CARD); + // Fresh mints keep their input so the viewer can reroll. + assert.deepEqual(viewer.remint, INPUT); + }); + + it("viewMintedCardJob ignores jobs that are still minting", () => { + let resolveMint; + void runCardMintJob( + INPUT, + () => new Promise((resolve) => (resolveMint = resolve)), + ); + const jobId = getCardMintJobs()[0].jobId; + + viewMintedCardJob(jobId); + + assert.equal(getCardMintJobs().length, 1); + assert.equal(getCardViewerState(), null); + resolveMint(CARD); // avoid a dangling promise + }); + + it("dismissCardMintJob removes only the named job", async () => { + await runCardMintJob(INPUT, () => Promise.reject(new Error("a"))); + await runCardMintJob({ agentId: "agent-2", agentName: "Wren" }, () => + Promise.reject(new Error("b")), + ); + const [first, second] = getCardMintJobs(); + + dismissCardMintJob(first.jobId); + + const jobs = getCardMintJobs(); + assert.equal(jobs.length, 1); + assert.equal(jobs[0].jobId, second.jobId); + }); + + it("openCardViewer/closeCardViewer manage archive views without remint", () => { + openCardViewer({ card: CARD, agentName: "Eva", remint: null }); + assert.equal(getCardViewerState()?.remint, null); + closeCardViewer(); + assert.equal(getCardViewerState(), null); + }); + + it("gallery open flag toggles and notifies subscribers", () => { + let notified = 0; + const unsubscribe = subscribeCardMintStore(() => { + notified += 1; + }); + setCardGalleryOpen(true); + assert.equal(getCardGalleryOpen(), true); + setCardGalleryOpen(true); // no-op must not notify + setCardGalleryOpen(false); + assert.equal(getCardGalleryOpen(), false); + assert.equal(notified, 2); + unsubscribe(); + }); + + it("concurrent jobs keep distinct snapshots (referential updates)", async () => { + const before = getCardMintJobs(); + await runCardMintJob(INPUT, () => Promise.resolve(CARD)); + const after = getCardMintJobs(); + assert.notEqual(before, after, "snapshot identity must change on update"); + }); +}); diff --git a/desktop/src/features/agents/cardMintStore.ts b/desktop/src/features/agents/cardMintStore.ts new file mode 100644 index 0000000000..c38bf393cc --- /dev/null +++ b/desktop/src/features/agents/cardMintStore.ts @@ -0,0 +1,236 @@ +import * as React from "react"; +import { toast } from "sonner"; + +import { + mintAgentCard, + NO_OPENAI_KEY_PREFIX, + type MintedAgentCard, + type SnapshotMemoryLevel, +} from "@/shared/api/tauriPersonas"; + +/** + * Module store for agent-card mints (`useSyncExternalStore` pattern, same as + * `avatarPresentationStore`). + * + * A mint is one stateless ~2–3 minute Rust call. Owning the in-flight promise + * here — instead of inside the mint dialog — is what makes the dialog + * non-blocking: it dispatches and closes, the composer activity rail shows a + * live "Minting card…" chip, and completion lands as a clickable toast plus a + * persistent "card ready" chip, none of which need the dialog mounted. + */ + +/** Everything needed to run (or re-run) one mint. */ +export type CardMintInput = { + agentId: string; + agentName: string; + styleNotes?: string; + lock?: boolean; + /** Memory to embed in the card's snapshot. Omitted = "none". */ + memoryLevel?: SnapshotMemoryLevel; +}; + +export type CardMintJob = { + jobId: string; + input: CardMintInput; + phase: "minting" | "done" | "error"; + /** Populated when phase is "done". */ + card: MintedAgentCard | null; + /** Populated when phase is "error". */ + error: string | null; + startedAt: number; +}; + +/** Card content shown by the global viewer dialog. */ +export type CardViewerState = { + card: MintedAgentCard; + agentName: string; + /** + * Present when the card can be rerolled (fresh mints carry their input; + * archive views do not — the original style notes are gone). + */ + remint: CardMintInput | null; + /** + * Monotonic per-open sequence, assigned by the store. The viewer keys its + * content on this so switching cards remounts (resetting recipients/menu + * state) — card bytes can't serve as the key because every card PNG shares + * the same header prefix and dimensions. + */ + viewerSeq: number; +}; + +let jobs: CardMintJob[] = []; +let viewer: CardViewerState | null = null; +let galleryOpen = false; +const listeners = new Set<() => void>(); +let nextJobId = 1; +let nextViewerSeq = 1; + +function emitChange(): void { + for (const listener of listeners) listener(); +} + +function updateJob(jobId: string, patch: Partial): void { + jobs = jobs.map((job) => (job.jobId === jobId ? { ...job, ...patch } : job)); + emitChange(); +} + +/** + * Run one mint as a background job. `mintFn` is injectable for tests; the + * public `startCardMint` binds the real Tauri command. + */ +export async function runCardMintJob( + input: CardMintInput, + mintFn: ( + id: string, + styleNotes?: string, + lock?: boolean, + memoryLevel?: SnapshotMemoryLevel, + ) => Promise, +): Promise { + const jobId = `card-mint-${nextJobId++}`; + jobs = [ + ...jobs, + { + jobId, + input, + phase: "minting", + card: null, + error: null, + startedAt: Date.now(), + }, + ]; + emitChange(); + + try { + const card = await mintFn( + input.agentId, + input.styleNotes, + input.lock, + input.memoryLevel, + ); + updateJob(jobId, { phase: "done", card }); + toast.success(`${input.agentName}'s card is ready`, { + action: { + label: "View card", + onClick: () => viewMintedCardJob(jobId), + }, + duration: 10_000, + }); + } catch (error) { + let message = error instanceof Error ? error.message : "Card mint failed."; + if (message.startsWith(NO_OPENAI_KEY_PREFIX)) { + // The dialog pre-checks the key, so this only happens when the key was + // removed between dialog-open and mint. The dialog's key-setup panel is + // long gone — surface a plain instruction instead of the wire prefix. + message = message.slice(NO_OPENAI_KEY_PREFIX.length).trim(); + } else if ( + message.startsWith("Card mint failed (HTTP 401 ") || + message.includes("Incorrect API key") + ) { + // The saved OpenAI key is invalid or expired. Only match the OpenAI-call + // envelope prefix and the specific Incorrect-API-key message to avoid + // rewriting unrelated 401s (e.g. "Avatar fetch failed: HTTP 401 …"). + message = + 'The OpenAI API key is invalid or expired. Open the mint dialog and use "Update API key" to replace it.'; + } + updateJob(jobId, { phase: "error", error: message }); + toast.error(`Minting ${input.agentName}'s card failed`, { + description: message, + }); + } +} + +/** Start a mint in the background. Fire-and-forget; state flows via the store. */ +export function startCardMint(input: CardMintInput): void { + void runCardMintJob(input, mintAgentCard); +} + +/** Open the finished card of a job in the viewer and clear its rail chip. */ +export function viewMintedCardJob(jobId: string): void { + const job = jobs.find((candidate) => candidate.jobId === jobId); + if (job?.phase !== "done" || !job.card) return; + viewer = { + card: job.card, + agentName: job.input.agentName, + remint: job.input, + viewerSeq: nextViewerSeq++, + }; + jobs = jobs.filter((candidate) => candidate.jobId !== jobId); + emitChange(); +} + +/** Remove a job chip (used for error dismissal). */ +export function dismissCardMintJob(jobId: string): void { + const next = jobs.filter((candidate) => candidate.jobId !== jobId); + if (next.length === jobs.length) return; + jobs = next; + emitChange(); +} + +/** Open the viewer on an arbitrary card (e.g. one loaded from the archive). */ +export function openCardViewer( + state: Omit, +): void { + viewer = { ...state, viewerSeq: nextViewerSeq++ }; + emitChange(); +} + +export function closeCardViewer(): void { + if (!viewer) return; + viewer = null; + emitChange(); +} + +export function setCardGalleryOpen(open: boolean): void { + if (galleryOpen === open) return; + galleryOpen = open; + emitChange(); +} + +export function resetCardMintStore(): void { + jobs = []; + viewer = null; + galleryOpen = false; + emitChange(); +} + +export function getCardMintJobs(): CardMintJob[] { + return jobs; +} + +export function getCardViewerState(): CardViewerState | null { + return viewer; +} + +export function getCardGalleryOpen(): boolean { + return galleryOpen; +} + +export function subscribeCardMintStore(listener: () => void): () => void { + listeners.add(listener); + return () => listeners.delete(listener); +} + +export function useCardMintJobs(): CardMintJob[] { + return React.useSyncExternalStore( + subscribeCardMintStore, + getCardMintJobs, + getCardMintJobs, + ); +} + +export function useCardViewerState(): CardViewerState | null { + return React.useSyncExternalStore( + subscribeCardMintStore, + getCardViewerState, + getCardViewerState, + ); +} + +export function useCardGalleryOpen(): boolean { + return React.useSyncExternalStore( + subscribeCardMintStore, + getCardGalleryOpen, + getCardGalleryOpen, + ); +} diff --git a/desktop/src/features/agents/lib/agentCardGalleryState.test.mjs b/desktop/src/features/agents/lib/agentCardGalleryState.test.mjs new file mode 100644 index 0000000000..ebc7caa8f4 --- /dev/null +++ b/desktop/src/features/agents/lib/agentCardGalleryState.test.mjs @@ -0,0 +1,96 @@ +import assert from "node:assert/strict"; +import { describe, it } from "node:test"; + +// Pins the gallery's query-state boundary: a rejected `list_agent_cards` +// MUST surface as an error state — never as "No cards yet". `data` falls +// back to `[]` at the call site, so an emptiness-first branch would render +// a permissions/IPC failure as a false empty archive of paid cards. + +import { agentCardGalleryViewState } from "./agentCardGalleryState.ts"; + +describe("agentCardGalleryViewState", () => { + it("maps a rejected query to an error state with the message, not empty", () => { + const state = agentCardGalleryViewState({ + isLoading: false, + isError: true, + error: new Error("cards directory unreadable"), + data: undefined, + }); + assert.deepEqual(state, { + kind: "error", + message: "cards directory unreadable", + }); + }); + + it("error wins even when data has fallen back to an empty array", () => { + // react-query keeps `data: undefined` on first failure, but the call + // site coalesces to []. Guard the exact shape that caused the bug. + const state = agentCardGalleryViewState({ + isLoading: false, + isError: true, + error: new Error("ipc failure"), + data: [], + }); + assert.equal(state.kind, "error"); + }); + + it("stringifies non-Error rejections", () => { + const state = agentCardGalleryViewState({ + isLoading: false, + isError: true, + error: "permission denied", + data: undefined, + }); + assert.deepEqual(state, { kind: "error", message: "permission denied" }); + }); + + it("null/undefined rejections still produce a message", () => { + const state = agentCardGalleryViewState({ + isLoading: false, + isError: true, + error: null, + data: undefined, + }); + assert.deepEqual(state, { kind: "error", message: "unknown error" }); + }); + + it("loading while fetching", () => { + const state = agentCardGalleryViewState({ + isLoading: true, + isError: false, + error: null, + data: undefined, + }); + assert.deepEqual(state, { kind: "loading" }); + }); + + it("no data yet (not loading, not error) stays loading, not empty", () => { + const state = agentCardGalleryViewState({ + isLoading: false, + isError: false, + error: null, + data: undefined, + }); + assert.deepEqual(state, { kind: "loading" }); + }); + + it("resolved and empty is the only path to the empty state", () => { + const state = agentCardGalleryViewState({ + isLoading: false, + isError: false, + error: null, + data: [], + }); + assert.deepEqual(state, { kind: "empty" }); + }); + + it("resolved with cards renders cards", () => { + const state = agentCardGalleryViewState({ + isLoading: false, + isError: false, + error: null, + data: [{ storedFileName: "a.png" }], + }); + assert.deepEqual(state, { kind: "cards" }); + }); +}); diff --git a/desktop/src/features/agents/lib/agentCardGalleryState.ts b/desktop/src/features/agents/lib/agentCardGalleryState.ts new file mode 100644 index 0000000000..8f6cf4269e --- /dev/null +++ b/desktop/src/features/agents/lib/agentCardGalleryState.ts @@ -0,0 +1,35 @@ +/** + * View-state selection for the minted-card gallery. + * + * Kept as a pure function so the error boundary is testable without a DOM: + * a rejected `list_agent_cards` query MUST surface as an error state, never + * as an empty archive — "No cards yet" on a permissions/IPC failure tells + * someone their paid, persisted cards do not exist. + */ + +export type AgentCardGalleryViewState = + | { kind: "loading" } + | { kind: "error"; message: string } + | { kind: "empty" } + | { kind: "cards" }; + +export function agentCardGalleryViewState(query: { + isLoading: boolean; + isError: boolean; + error: unknown; + data: readonly unknown[] | undefined; +}): AgentCardGalleryViewState { + // Error wins over everything: `data` falls back to `[]` at the call site, + // so checking emptiness first would render a failure as a false empty. + if (query.isError) { + const message = + query.error instanceof Error + ? query.error.message + : String(query.error ?? "unknown error"); + return { kind: "error", message }; + } + if (query.isLoading || query.data === undefined) { + return { kind: "loading" }; + } + return query.data.length === 0 ? { kind: "empty" } : { kind: "cards" }; +} diff --git a/desktop/src/features/agents/ui/AgentCardMintDialog.test.mjs b/desktop/src/features/agents/ui/AgentCardMintDialog.test.mjs new file mode 100644 index 0000000000..ff3efaec92 --- /dev/null +++ b/desktop/src/features/agents/ui/AgentCardMintDialog.test.mjs @@ -0,0 +1,257 @@ +import assert from "node:assert/strict"; +import { describe, it } from "node:test"; + +// Tests for the key-panel visibility derivations that AgentCardMintDialog +// imports from cardMintKeyUtils. These tests exercise the exact production +// module — changes to any exported function will cause failures here. + +import { + isReadOnlyLayer, + isWritableLayer, + keyPanelTitle, + showCancelButton, + showKeyPanel, + showKeyStatusRow, + showReadOnlyRow, +} from "./cardMintKeyUtils.ts"; + +describe("cardMintKeyUtils — key panel derivations", () => { + // ── isWritableLayer ──────────────────────────────────────────────────────── + + it("isWritableLayer_none_true", () => { + assert.equal(isWritableLayer("none"), true); + }); + + it("isWritableLayer_global_true", () => { + assert.equal(isWritableLayer("global"), true); + }); + + it("isWritableLayer_agent_false", () => { + assert.equal(isWritableLayer("agent"), false); + }); + + it("isWritableLayer_persona_false", () => { + assert.equal(isWritableLayer("persona"), false); + }); + + it("isWritableLayer_process_false", () => { + assert.equal(isWritableLayer("process"), false); + }); + + it("isWritableLayer_undefined_false", () => { + // Unknown (pending/error) — don't offer a write path + assert.equal(isWritableLayer(undefined), false); + }); + + // ── isReadOnlyLayer ──────────────────────────────────────────────────────── + + it("isReadOnlyLayer_agent_true", () => { + assert.equal(isReadOnlyLayer("agent"), true); + }); + + it("isReadOnlyLayer_persona_true", () => { + assert.equal(isReadOnlyLayer("persona"), true); + }); + + it("isReadOnlyLayer_process_true", () => { + assert.equal(isReadOnlyLayer("process"), true); + }); + + it("isReadOnlyLayer_global_false", () => { + assert.equal(isReadOnlyLayer("global"), false); + }); + + it("isReadOnlyLayer_none_false", () => { + assert.equal(isReadOnlyLayer("none"), false); + }); + + it("isReadOnlyLayer_undefined_false", () => { + assert.equal(isReadOnlyLayer(undefined), false); + }); + + // ── showKeyPanel ─────────────────────────────────────────────────────────── + // Key panel replaces the mint form ONLY for first-time setup (none) or + // user-initiated editing (editingKey). Read-only layers do NOT replace the + // mint form — they show an inline status row instead. + + it("showKeyPanel_none_notEditing_shows", () => { + // First-time user: key not set → show setup panel + assert.equal(showKeyPanel("none", false), true); + }); + + it("showKeyPanel_global_notEditing_hides", () => { + // Normal state: key in global defaults, not editing → show mint form + assert.equal(showKeyPanel("global", false), false); + }); + + it("showKeyPanel_global_editing_shows", () => { + // User clicked Update → show the update panel + assert.equal(showKeyPanel("global", true), true); + }); + + it("showKeyPanel_agent_notEditing_hides", () => { + // Read-only layer: mint form stays visible; inline row shown instead + assert.equal(showKeyPanel("agent", false), false); + }); + + it("showKeyPanel_persona_notEditing_hides", () => { + assert.equal(showKeyPanel("persona", false), false); + }); + + it("showKeyPanel_process_notEditing_hides", () => { + assert.equal(showKeyPanel("process", false), false); + }); + + it("showKeyPanel_agent_editing_shows", () => { + // User clicked Why? on a read-only row → show the redirect panel + assert.equal(showKeyPanel("agent", true), true); + }); + + it("showKeyPanel_persona_editing_shows", () => { + assert.equal(showKeyPanel("persona", true), true); + }); + + it("showKeyPanel_process_editing_shows", () => { + assert.equal(showKeyPanel("process", true), true); + }); + + it("showKeyPanel_undefined_notEditing_hides", () => { + // Query pending/error → show mint form (fail-open, no panel claim) + assert.equal(showKeyPanel(undefined, false), false); + }); + + // ── showCancelButton ─────────────────────────────────────────────────────── + + it("showCancelButton_global_editing_shows", () => { + // Update mode for a global key: Cancel returns to the mint form + assert.equal(showCancelButton("global", true), true); + }); + + it("showCancelButton_none_editing_hides", () => { + // First-time setup: no cancel (no mint form to return to) + assert.equal(showCancelButton("none", true), false); + }); + + it("showCancelButton_global_notEditing_hides", () => { + assert.equal(showCancelButton("global", false), false); + }); + + it("showCancelButton_agent_editing_shows", () => { + // Read-only layer + user clicked Why?: Cancel returns to the mint form + assert.equal(showCancelButton("agent", true), true); + }); + + it("showCancelButton_persona_editing_shows", () => { + assert.equal(showCancelButton("persona", true), true); + }); + + it("showCancelButton_process_editing_shows", () => { + assert.equal(showCancelButton("process", true), true); + }); + + // ── showKeyStatusRow ─────────────────────────────────────────────────────── + + it("showKeyStatusRow_global_notEditing_shows", () => { + // Confirmed writable key: show "Using your saved OpenAI key · Update" + assert.equal(showKeyStatusRow("global", false), true); + }); + + it("showKeyStatusRow_global_editing_hides", () => { + // In update panel: status row is redundant while editing + assert.equal(showKeyStatusRow("global", true), false); + }); + + it("showKeyStatusRow_none_notEditing_hides", () => { + // No key: show setup panel, not status row + assert.equal(showKeyStatusRow("none", false), false); + }); + + it("showKeyStatusRow_agent_notEditing_hides", () => { + // Read-only layer: use showReadOnlyRow instead + assert.equal(showKeyStatusRow("agent", false), false); + }); + + it("showKeyStatusRow_undefined_notEditing_hides", () => { + // Query pending/error: do not assert key existence + assert.equal(showKeyStatusRow(undefined, false), false); + }); + + // ── showReadOnlyRow ──────────────────────────────────────────────────────── + // Inline provenance row on the mint form for keys the dialog cannot update. + + it("showReadOnlyRow_agent_notEditing_shows", () => { + assert.equal(showReadOnlyRow("agent", false), true); + }); + + it("showReadOnlyRow_persona_notEditing_shows", () => { + assert.equal(showReadOnlyRow("persona", false), true); + }); + + it("showReadOnlyRow_process_notEditing_shows", () => { + assert.equal(showReadOnlyRow("process", false), true); + }); + + it("showReadOnlyRow_agent_editing_hides", () => { + // User clicked Why? → redirect panel shown; row hidden + assert.equal(showReadOnlyRow("agent", true), false); + }); + + it("showReadOnlyRow_global_notEditing_hides", () => { + // Global key uses showKeyStatusRow instead + assert.equal(showReadOnlyRow("global", false), false); + }); + + it("showReadOnlyRow_none_hides", () => { + assert.equal(showReadOnlyRow("none", false), false); + }); + + it("showReadOnlyRow_undefined_hides", () => { + assert.equal(showReadOnlyRow(undefined, false), false); + }); + + // ── Mint-reachability invariant ──────────────────────────────────────────── + // The mint form (and Mint button) must be reachable whenever a key resolves. + // showKeyPanel returns true only for setup (none) or user-initiated edit. + + it("mintReachable_global_noEdit", () => { + assert.equal(showKeyPanel("global", false), false); + }); + + it("mintReachable_agent_noEdit", () => { + assert.equal(showKeyPanel("agent", false), false); + }); + + it("mintReachable_persona_noEdit", () => { + assert.equal(showKeyPanel("persona", false), false); + }); + + it("mintReachable_process_noEdit", () => { + assert.equal(showKeyPanel("process", false), false); + }); + + it("mintBlocked_none_noEdit", () => { + // Only when no key is set at all does the panel gate minting + assert.equal(showKeyPanel("none", false), true); + }); + + // ── keyPanelTitle ────────────────────────────────────────────────────────── + + it("keyPanelTitle_none_firstTimeSetup", () => { + assert.equal( + keyPanelTitle("none", false), + "One-time setup: OpenAI API key", + ); + }); + + it("keyPanelTitle_global_editing_update", () => { + assert.equal(keyPanelTitle("global", true), "Update OpenAI API key"); + }); + + it("keyPanelTitle_agent_readOnly", () => { + assert.equal(keyPanelTitle("agent", false), "OpenAI API key"); + }); + + it("keyPanelTitle_persona_readOnly", () => { + assert.equal(keyPanelTitle("persona", true), "OpenAI API key"); + }); +}); diff --git a/desktop/src/features/agents/ui/AgentCardMintDialog.tsx b/desktop/src/features/agents/ui/AgentCardMintDialog.tsx new file mode 100644 index 0000000000..b2bffc00df --- /dev/null +++ b/desktop/src/features/agents/ui/AgentCardMintDialog.tsx @@ -0,0 +1,463 @@ +import * as React from "react"; +import { + AlertCircle, + Brain, + ExternalLink, + GalleryVerticalEnd, + KeyRound, + Lock, + Sparkles, +} from "lucide-react"; +import { useMutation, useQuery, useQueryClient } from "@tanstack/react-query"; +import { openUrl } from "@tauri-apps/plugin-opener"; +import { toast } from "sonner"; + +import { + setCardGalleryOpen, + startCardMint, +} from "@/features/agents/cardMintStore"; +import { globalAgentConfigQueryKey } from "@/features/agents/useGlobalAgentConfig"; +import { + cardMintKeyStatus, + cardMintSaveOpenaiKey, + type CardMintKeyLayer, + type SnapshotMemoryLevel, +} from "@/shared/api/tauriPersonas"; +import { Button } from "@/shared/ui/button"; +import { + Dialog, + DialogContent, + DialogDescription, + DialogHeader, + DialogTitle, +} from "@/shared/ui/dialog"; +import { Input } from "@/shared/ui/input"; +import { Switch } from "@/shared/ui/switch"; +import { Textarea } from "@/shared/ui/textarea"; +import { SnapshotOptionMenu } from "./SnapshotOptionMenu"; +import { + isReadOnlyLayer, + keyPanelTitle, + showCancelButton, + showKeyPanel, + showKeyStatusRow, + showReadOnlyRow, +} from "./cardMintKeyUtils"; + +const OPENAI_KEYS_URL = "https://platform.openai.com/api-keys"; + +/** Same three levels as snapshot export; "Agent only" is the safe default. */ +const MEMORY_LEVELS: { value: SnapshotMemoryLevel; label: string }[] = [ + { value: "none", label: "Agent only" }, + { value: "core", label: "Agent + core memory" }, + { value: "everything", label: "Agent + all memories" }, +]; + +/** + * The free alternative, as an action: ordinary snapshot export shares the + * same importable agent without card art or API spend. Rendered in both the + * key-setup panel and the normal pre-mint form (the cost disclosure and its + * escape hatch must be visible BEFORE any spend, not only during onboarding). + */ +function FreeSharePathRow({ + disabled, + onExportInstead, +}: { + disabled: boolean; + onExportInstead?: () => void; +}) { + return ( +
+

+ Don’t want to spend money? Ordinary export shares the same importable + agent — free, just without the card art. +

+ {onExportInstead ? ( + + ) : null} +
+ ); +} + +/** + * Mint-a-trading-card dialog — the pre-mint half only: key setup (when + * needed) → optional style notes → "Mint card". Minting itself runs as a + * background job in `cardMintStore`: this dialog dispatches and closes, the + * composer activity rail shows live status, and the finished card opens in + * the global `AgentCardViewerDialog` (preview, reroll, save, share). + * + * The saved PNG carries the agent's `buzz_agent_snapshot` chunk, so sharing + * the card shares an importable agent (fresh identity, never secrets; memory + * only when the owner opts in below — plaintext unless the card is locked). + * All snapshot construction and verification happens in Rust. + */ +export function AgentCardMintDialog({ + agentId, + agentName, + canLock, + onExportInstead, + onOpenChange, +}: { + /** Instance pubkey or definition slug — same resolution as snapshot export. */ + agentId: string; + agentName: string; + /** + * True when the agent has a linked instance (a keypair to lock to). + * Locking is disabled — with an explanation — for bare definitions. + */ + canLock: boolean; + /** + * Free alternative: close this dialog and open the ordinary snapshot + * export flow (no API spend). Omitted = the action is not rendered. + */ + onExportInstead?: () => void; + onOpenChange: (open: boolean) => void; +}) { + const [styleNotes, setStyleNotes] = React.useState(""); + const [lockCard, setLockCard] = React.useState(false); + const [memoryLevel, setMemoryLevel] = + React.useState("none"); + const [keyDraft, setKeyDraft] = React.useState(""); + const [editingKey, setEditingKey] = React.useState(false); + + const queryClient = useQueryClient(); + + const effectiveLock = canLock && lockCard; + // Embedded memory is plaintext in an unlocked card — and unlocked cards + // are meant to be shared. A locked card encrypts the whole manifest to the + // (owner, agent) pair, so the plaintext warning would be false there. + const showMemoryWarning = memoryLevel !== "none" && !effectiveLock; + + // Whether a key already resolves through the agent's env layering, and from + // which layer. While unknown (loading/error) we treat as if no verified key + // exists — mint still works fail-open, but we don't assert a key is present. + const keyStatusQuery = useQuery({ + queryKey: ["cardMintKeyStatus", agentId], + queryFn: () => cardMintKeyStatus(agentId), + }); + const keyLayer: CardMintKeyLayer | undefined = keyStatusQuery.data; + // True when the key resolves from a layer this dialog cannot update. + const keyIsReadOnly = isReadOnlyLayer(keyLayer); + + // Save the pasted key into the global Agent Defaults env — the same single + // source of truth every agent inherits. Narrow Rust seam: validated + // single-key merge, never restarts running agents (the mint re-reads + // config per call, so no restart is needed for minting). + const saveKeyMutation = useMutation({ + mutationFn: (key: string) => cardMintSaveOpenaiKey(key), + onSuccess: () => { + // The key now lives in global defaults — update the cached layer so the + // status row shows correctly without waiting for a refetch. + queryClient.setQueryData( + ["cardMintKeyStatus", agentId], + "global", + ); + // The Agent Defaults editor caches the whole config — refetch it so a + // later-opened settings view shows the key we just wrote. + void queryClient.invalidateQueries({ + queryKey: globalAgentConfigQueryKey, + }); + setKeyDraft(""); + setEditingKey(false); + toast.success( + "API key saved to your agent defaults. Running agents pick it up on their next restart.", + ); + }, + onError: (error) => + toast.error(typeof error === "string" ? error : "Couldn't save the key."), + }); + + function beginMint() { + // Dispatch to the background store and close: the composer rail shows + // "Minting card…" and the completion toast opens the viewer. + startCardMint({ + agentId, + agentName, + styleNotes: styleNotes.trim() || undefined, + lock: effectiveLock, + memoryLevel: canLock ? memoryLevel : "none", + }); + onOpenChange(false); + } + + return ( + + + + + + {`Create ${agentName}'s card`} + + + Mint a collectible trading card that doubles as a shareable, + importable copy of this agent. + + + + {showKeyPanel(keyLayer, editingKey) ? ( +
+
+ + + {keyPanelTitle(keyLayer, editingKey)} + + {keyIsReadOnly ? ( + // Key resolves from a layer the dialog cannot write to — show + // a read-only redirect instead of an input that would be + // shadowed by the higher-priority layer. +

+ {keyLayer === "agent" + ? "This agent's OpenAI key is set in its own agent settings — update it there." + : keyLayer === "persona" + ? "This agent's OpenAI key comes from its linked persona settings — update it there." + : "This agent's OpenAI key is set in the process environment — update it in your shell or launch config."} +

+ ) : ( + <> +

+ Minting a card costs money — it generates the art and card + text through the OpenAI API with your key (typically well + under a dollar per mint, billed by OpenAI). The key is saved + as OPENAI_API_KEY in your + agent defaults env — that's the row to update in Settings if + you ever need to change it there. +

+ + setKeyDraft(e.target.value)} + placeholder="sk-…" + type="password" + value={keyDraft} + /> + + )} +
+ +
+ {showCancelButton(keyLayer, editingKey) ? ( + + ) : null} + {!keyIsReadOnly ? ( + + ) : null} +
+
+ ) : ( +
+
+