An inbound private managed-agent (PMA) tombstone could be resurrected
locally. The deleted branch of reconcile_inbound_private_managed_agent
persisted nothing when no local record existed, and the generation guards
only run when a record IS present. So a delayed stale active head at the
tombstone's generation (or lower) — a fresh-device hydration/live race, or a
post-erase replay — found no record, skipped every guard, and hit the
no-match insert path, reconstructing the deleted agent including its nsec.
Fix: give every coordinate a DURABLE generation floor, keyed by the retained
managed_agent_aggregates rows (the MAX generation across active AND deleted
rows). The deleted branch now seeds a confirmed tombstone row even when no
local record exists, and both branches gate on the floor before touching the
store: an inbound head at <= the floor generation is dropped, while a genuine
re-creation at a strictly higher generation still passes.
- seed_confirmed_managed_agent_tombstone: idempotent (INSERT OR IGNORE)
confirmed deleted-row seed; never downgrades a flush-authored row.
- reconcile_inbound_private_managed_agent: open retention once for both
branches, read the floor, reject stale heads, seed the tombstone floor
before erasing the local record.
Regression: the exact active-N / deleted-N+1 / replay-active-N interleaving
stays floored (agent absent), a re-created N+2 clears the floor, plus the
seed's idempotency and synced-deleted-head contract. Full desktop/src-tauri
workspace: 2310 passed / 0 failed / 13 ignored; fmt and clippy -D warnings
clean.
Addresses PR #4940 review thread discussion_r3724146156.
Co-authored-by: Mongo <5c25403eab7271f9f94ddd4f2b270e8cac2c92e2c830c51877cca6ec974ffb3f@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
**Category:** fix
**User Impact:** Custom emoji with valid 64-character names can now be
used as reactions without errors.
**Problem:** Buzz accepted 64-character custom emoji names during
registration, but rejected them as reactions after the required
surrounding colons made the payload 66 characters. Validation also
differed between desktop, SDK, relay, and storage boundaries.
<img width="554" height="47" alt="image"
src="https://github.com/user-attachments/assets/4013452f-210e-4dd3-9003-f45ff3b28dc8"
/>
**Solution:** Keep the product limit at 64 ASCII characters for custom
emoji names, enforce it consistently when emoji sets are registered, and
allow only valid matching custom reaction payloads up to 66 characters.
Widen the reaction projection to preserve the wrapped payload while
retaining the existing 64-character limit for ordinary reactions.
<details>
<summary>File changes</summary>
**crates/buzz-sdk/src/builders.rs**
Defines the shared custom emoji boundaries and covers accepted
64-character and rejected 65-character shortcodes.
**crates/buzz-relay/src/handlers/ingest.rs**
Validates emoji-set shortcodes and permits 66-character reactions only
when they are valid colon-wrapped custom emoji with a matching tag.
**crates/buzz-db/src/event.rs**
Adds storage regression coverage for maximum-length custom emoji
reactions.
**crates/buzz-db/src/migration.rs**
Verifies the reaction column migration is applied correctly.
**desktop/src/shared/api/customEmoji.ts**
Enforces the existing 64-character shortcode maximum during desktop
normalization and registration/import.
**desktop/src/shared/api/customEmoji.test.mjs**
Covers the desktop shortcode boundary.
**migrations/0027_long_reaction_payloads.sql**
Widens stored reaction payloads to 66 characters for the two required
surrounding colons.
**schema/schema.sql**
Keeps the desired schema aligned with the migration.
</details>
## Reproduction Steps
1. Register or import a custom emoji whose ASCII shortcode is exactly 64
characters.
2. Select that emoji as a reaction to a message.
3. Confirm the reaction publishes, persists, and renders without an
error.
4. Attempt to register a 65-character shortcode and confirm it is
rejected.
5. Publish an ordinary or malformed reaction over 64 characters and
confirm the relay rejects it.
## Verification
- `cargo test -p buzz-sdk`: 243 passed
- `cargo test -p buzz-db`: 94 passed, 152 Postgres-required tests
ignored
- `pnpm test` in `desktop`: 3,859 passed
- `cargo test -p buzz-relay`: 795 passed, 9 existing
Postgres-unavailable failures, 35 ignored; new reaction boundary tests
pass directly
- `cargo fmt --all -- --check`
- `git diff --check`
Originating Buzz channel: `f2ec9671-d78e-4cde-894c-9f4c458c7f1f`
---------
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
Co-authored-by: npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6w <52a228d6edf316ec6812ac3c9fc0d696ab59fc7954d77e7be31eedcddf91335b@buzz.block.builderlab.xyz>
**Category:** improvement
**User Impact:** People can leave their final Buzz community and return
to **Join or create a community** without losing their signed-in
identity.
**Problem:** Buzz Desktop blocked people from leaving when only one
community remained. Its existing remove action also changed local
configuration without ending relay membership.
**Solution:** Allow the final community to be left. Buzz now asks the
relay to end membership, removes the community locally only after
acceptance, and returns the person to the community selector while
keeping their identity signed in. If other communities remain, Buzz
switches to one of them. Relay rejection or timeout keeps the community
in place and shows an actionable retry error.
<details>
<summary>File changes</summary>
**desktop/src/features/communities/leaveCommunity.ts**
Adds signed kind 28936 publishing for active and inactive community
relays with actionable timeout handling.
**desktop/src/features/communities/leaveCommunity.test.mjs**
Covers event shape, relay selection, acceptance gating, rejection,
timeout messaging, and cleanup.
**desktop/src/features/communities/useCommunities.tsx**
Allows final-community removal and clears community-specific storage
without touching identity.
**desktop/src/features/communities/resolveCommunityRemoval.test.mjs**
Covers final, active, and inactive community removal state transitions.
**desktop/src/app/useCommunityNavigationTransitions.ts**
Gates local removal on relay acceptance and routes to a fallback
community or setup selector.
**desktop/src/app/AppShell.tsx**
Passes the asynchronous leave operation through shell entry points.
**desktop/src/features/communities/ui/EditCommunityDialog.tsx**
Replaces the local-only remove action with a pending-aware Leave
Community action that retains actionable errors.
**desktop/src/features/communities/ui/CommunitySwitcher.tsx**
Enables leaving the final community and carries the asynchronous
callback.
**desktop/src/features/sidebar/ui/AppSidebar.tsx**
Carries the asynchronous leave callback through sidebar props.
**desktop/src/features/sidebar/ui/CommunityRail.tsx**
Enables leaving the final community from rail settings.
**desktop/src/features/sidebar/ui/SidebarProfileCard.tsx**
Carries the asynchronous leave callback through profile community
settings.
**desktop/src/testing/e2eBridge.ts**
Teaches the mock relay to accept NIP-43 leave events.
**desktop/tests/e2e/community-rail.spec.ts**
Updates leave interactions and verifies final-community setup
navigation, storage cleanup, and identity preservation.
</details>
### Reproduction steps
1. Run Buzz Desktop with a signed-in identity and one joined community.
2. Open Community settings and choose **Leave Community**.
3. Confirm the app shows **Join or create a community** and the existing
identity remains signed in.
4. Repeat with two communities and confirm leaving the active one
switches cleanly to the remaining community.
5. Reject or withhold the relay `OK` response and confirm the community
remains configured with an actionable error in the dialog.
### Test plan
- `pnpm check`
- `pnpm build`
- `pnpm test` (3,913 passing)
- `pnpm build:e2e && pnpm exec playwright test
tests/e2e/community-rail.spec.ts --grep "final community"`
<img width="557" height="316" alt="image"
src="https://github.com/user-attachments/assets/b628182f-cba5-451d-ae4b-bee8d8dd19aa"
/>
---------
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: npub14ndfusear8wdpe4kss8h7juc7wjk78atnqzf63zvppcpneknv4sq6x9370 <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>
Co-authored-by: npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6w <52a228d6edf316ec6812ac3c9fc0d696ab59fc7954d77e7be31eedcddf91335b@buzz.block.builderlab.xyz>
Co-authored-by: npub14ndfusear8wdpe4kss8h7juc7wjk78atnqzf63zvppcpneknv4sq6x9370 <acda9e433d19dcd0e6b6840f7f4b98f3a56f1fab98049d444c087019e6d36560@buzz.block.builderlab.xyz>
Provider `context_length_exceeded` 400s permanently wedged agent
sessions: the turn errored, the oversized history persisted in the
in-memory session, and the usage baseline stayed frozen at the last
successful sub-threshold reading (failed requests report no usage), so
the preflight handoff gate never fired again — every later prompt failed
identically until an agent restart. The byte-truncation fallback never
intervened because it is a request-body limiter (`estimated_bytes`), not
a context-window defence; at context-window scale it is a measured
no-op.
This adds the reactive recovery path:
- **Typed classification.** `AgentError::LlmContextExceeded` is
classified at both non-success provider terminals — the shared `post()`
(Anthropic, OpenAI, Databricks) and `openrouter_post()` — on `status ==
400` plus a context-window body match, so ordinary 400s stay terminal.
- **Forced handoff.** A context-400 forces a summarize-handoff that
bypasses `should_handoff()` and `BUZZ_AGENT_MAX_HANDOFFS`, bounded by
its own per-turn budget (`MAX_CONTEXT_RECOVERIES_PER_RUN = 3`).
- **Shrink ladder.** The summarize prompt budget halves from the
observed rejected history size — not from `max_context_tokens`, the
number the provider just contradicted — rung to rung, with a 4096-byte
floor. A summarize call rejected for the same reason takes the next rung
instead of re-sticking. At the floor (overflow dominated by unshrinkable
frame: system prompt, tool schemas, live prompt) recovery is refused and
the provider error surfaces clearly instead of self-healing.
- **Baseline reset.** The stale usage baseline is cleared when a request
fails, so the preflight gate cannot stay frozen sub-threshold on
retries.
Named behavior changes:
1. **Anthropic and OpenRouter errors now carry the `(model)` stamp.**
Provider arms return their `Result` into the central error mapper
instead of early-returning past it, making the code match its documented
single-convergence contract at that mapper.
2. **`max_rounds` now counts completions the loop acts on.** A request
rejected with a context-400 that is then successfully recovered refunds
its round before the retry, paired 1:1 with a consumed recovery rung, so
the round cap is neither weakened nor able to drop a recovered turn
unanswered.
Related: #4805 — the complementary proactive fix (per-session
handoff-cap kill switch that let sessions grow to the provider wall).
#4805 prevents reaching the wall; this PR recovers at it.
---------
Co-authored-by: npub17jjz49l9jjmhhk7cac63j8yt9z555n9cw8vk7v5jz4vzw4ppld5qgj57cc <f4a42a97e594b77bdbd8ee35191c8b28a94a4cb871d96f32921558275421fb68@buzz.block.builderlab.xyz>
Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
## Summary
The desktop app no longer imports persona packs the way the docs
described. `PERSONA_PACK_SPEC.md` and the `meadow-core` example still
pointed users at a `.zip` import through "My Teams / My Agents → Import"
and a future "Install Pack" button — none of that exists anymore. The
app only imports agent/team **snapshots** (`.agent.json`/`.agent.png`,
`.team.json`/`.team.png`), and a persona-pack `.zip` is explicitly
rejected.
## Fix
Updated both docs to describe the current import paths (Agents / Agent
teams sections, snapshot files only) and added a note that persona packs
and desktop snapshots are separate, non-interchangeable formats today.
Fixes#4468
---------
Signed-off-by: SomSamantray <92726151+SomSamantray@users.noreply.github.com>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Co-authored-by: Wes <wesbillman@users.noreply.github.com>
Co-authored-by: Carl <c7ebe626f000404285d3686e1dc74cc07cc60a9754a150041ba132e14bd3e2ec@buzz.block.builderlab.xyz>
## Summary
- deliver macOS notifications through `UNUserNotificationCenter`
- route notification clicks to the referenced channel or thread through
the existing frontend activation path
- preserve click targets across cold startup and frontend remounts with
a small process-wide activation queue
- keep Linux notification activation behavior unchanged
## Architecture
A single `UNUserNotificationCenterDelegate` is installed during Tauri
setup. Each notification stores its navigation target in `userInfo`. On
click, Rust queues the target before emitting a wake-up event; the
frontend atomically drains the queue and dispatches the existing
notification action. The queue is the source of truth, which prevents
cold-start loss and duplicate delivery.
## Validation
Verified at `a81241611d617becf7640bee6fe56b5cdb4d0fab`:
- Biome format/check and lint
- TypeScript typecheck
- repository and desktop-Tauri `cargo fmt --check`
- repository and desktop-Tauri Clippy with `-D warnings`
- full pre-push desktop tests and Tauri workspace checks/tests
- desktop production build
Manual macOS validation passed: after explicitly ad-hoc signing the
local bundle with `xyz.block.buzz.app`, the operator confirmed real
Notification Center delivery and click navigation.
<details>
<summary>Local macOS test procedure</summary>
Tauri's generated ad-hoc signing identifier is not accepted by
`UNUserNotificationCenter`. Re-sign the local bundle with its bundle
identifier and keep other Buzz copies closed:
```bash
just desktop-release-build
APP="$HOME/.cache/cargo-target/aarch64-apple-darwin/release/bundle/macos/Buzz.app"
codesign --force --deep --sign - \
--identifier xyz.block.buzz.app \
--entitlements desktop/src-tauri/Entitlements.plist \
"$APP"
codesign --verify --deep --strict --verbose=2 "$APP"
pkill -x buzz-desktop || true
open -n "$APP"
```
</details>
Buzz channel: `55e2bfca-1b38-48fb-9dc2-584d400501f3`
---------
Signed-off-by: npub1dccv64krpcpse5cmkzfeh998cftungyatw3djt8jwdw6g43f7fyqzzmrf7 <6e30cd56c30e030cd31bb0939b94a7c257c9a09d5ba2d92cf2735da45629f248@buzz.block.builderlab.xyz>
Signed-off-by: am <6e30cd56c30e030cd31bb0939b94a7c257c9a09d5ba2d92cf2735da45629f248@buzz.block.builderlab.xyz>
Co-authored-by: npub1dccv64krpcpse5cmkzfeh998cftungyatw3djt8jwdw6g43f7fyqzzmrf7 <6e30cd56c30e030cd31bb0939b94a7c257c9a09d5ba2d92cf2735da45629f248@buzz.block.builderlab.xyz>
**Category:** new-feature
**User Impact:** Mobile now keeps each community’s appearance in sync
with desktop, including theme, accent, and system-mode preference.
**Problem:** Appearance choices were device-local, so the same account
could look different between desktop and mobile. Live sync could also
stop after the relay closed a subscription.
**Solution:** Store each community’s encrypted appearance preference on
its relay using the shared desktop wire contract, restore it from a
local identity-scoped cache, and apply replacement events live. Closed
subscriptions now recover with guarded backoff and fetch the latest
preference so no update is lost during the gap.
<details>
<summary>File changes</summary>
**mobile/lib/app.dart**
Connects community appearance state to the authenticated app lifecycle.
**mobile/lib/features/settings/accent_picker_page.dart**
Aligns mobile accent choices and selection behavior with the shared
catalog.
**mobile/lib/features/settings/settings_page/appearance_section.dart**
Clarifies the active appearance and hides accent controls when the Buzz
theme owns its neutral accent.
**mobile/lib/features/settings/theme_picker_page.dart**
Persists catalog theme choices through the community-scoped provider.
**mobile/lib/shared/theme/accent_colors.dart**
Matches desktop’s accent catalog and wire values.
**mobile/lib/shared/theme/buzz_theme.dart**
Keeps Buzz visually neutral without discarding the user’s stored accent
for other themes.
**mobile/lib/shared/theme/community_theme_preference.dart**
Defines and validates the versioned desktop-compatible appearance
payload.
**mobile/lib/shared/theme/community_theme_provider.dart**
Coordinates cache-first appearance loading with account and community
changes.
**mobile/lib/shared/theme/community_theme_sync.dart**
Adds encrypted NIP-78 relay persistence, live replacement handling,
deterministic ordering, safe seeding, and resilient subscription
recovery.
**mobile/lib/shared/theme/theme.dart**
Exports the community appearance modules.
**mobile/test/features/settings/theme_picker_page_test.dart**
Covers the updated settings behavior.
**mobile/test/shared/crypto/nip44_interop_test.dart**
Proves Dart decrypts a desktop-produced nostr-rs NIP-44 v2 preference.
**mobile/test/shared/theme/buzz_theme_test.dart**
Covers Buzz’s neutral rendering and stored-accent restoration.
**mobile/test/shared/theme/community_theme_preference_test.dart**
Covers wire parsing, validation, migration, and future-version handling.
**mobile/test/shared/theme/community_theme_sync_test.dart**
Covers cache/relay lifecycle, replacement ordering, switching races,
absence-only seeding, and closed-subscription recovery.
</details>
## Reproduction steps
1. Sign into desktop and mobile with the same account and join the same
community relay.
2. On desktop, choose a distinctive non-Buzz theme and accent; mobile
should update without a local toggle.
3. Restart mobile and confirm it restores the same appearance.
4. Change the mobile theme and accent and confirm desktop follows.
5. Leave mobile idle or backgrounded through a relay reconnect, then
change desktop again; mobile should resubscribe and catch up
automatically.
6. Switch communities and confirm each community restores only its own
appearance.
---------
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
Co-authored-by: npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6w <52a228d6edf316ec6812ac3c9fc0d696ab59fc7954d77e7be31eedcddf91335b@buzz.block.builderlab.xyz>
**Category:** new-feature
**User Impact:** Users can keep a distinct Appearance scheme for each
community and restore it on another desktop signed in with the same
identity.
**Problem:** A single global theme makes it harder to distinguish among
communities, and local-only preferences do not follow a user to another
device. **Solution:** Save each community's stable theme, accent, and
system-following selection as private encrypted relay state, backed by a
responsive local cache and guarded against switch races, invalid future
records, and relay failures.
<details>
<summary>File changes</summary>
**desktop/src/app/App.tsx**
Mounts the community-scoped theme controller inside the active community
lifecycle.
**desktop/src/features/settings/lib/appearanceScopeCopy.test.mjs**
Covers active-community and fallback labels used to explain Appearance
scope.
**desktop/src/features/settings/lib/appearanceScopeCopy.ts**
Builds a safe, trimmed label for the currently active community.
**desktop/src/features/settings/ui/SettingsPanels.tsx**
Clarifies which Appearance controls are per-community and which apply
globally when multiple communities exist.
**desktop/src/shared/constants/kinds.ts**
Defines the NIP-78 application-data event kind used for theme
preferences.
**desktop/src/shared/theme/CommunityThemeController.tsx**
Coordinates cached appearance, encrypted relay retrieval, live updates,
reconnect behavior, and safe community switching.
**desktop/src/shared/theme/ThemeProvider.tsx**
Exposes a single appearance application path so synchronized preferences
use the existing renderer and persistence behavior.
**desktop/src/shared/theme/communityThemePreference.test.mjs**
Covers contract validation, user/relay isolation, malformed records,
cache failures, and switch-race decisions.
**desktop/src/shared/theme/communityThemePreference.ts**
Defines the versioned stable preference contract, safe defaults, local
cache keys, and persistence guards.
**desktop/src/shared/theme/communityThemeSync.test.mjs**
Covers relay absence, unreadable records, unavailability, seeding
safety, and teardown of pending writes.
**desktop/src/shared/theme/communityThemeSync.ts**
Encrypts theme preferences to the user, publishes and retrieves NIP-78
state, and handles ordering and lifecycle safety.
</details>
### Reproduction steps
1. Join at least two communities and open **Settings → Appearance**.
2. Choose a different theme, accent, or system-following mode in each
community.
3. Switch between the communities and verify each one restores its own
scheme without overwriting the other.
4. Sign in on another desktop with the same Nostr identity, join the
same community, and verify its saved scheme is restored from that
community's relay.
5. Disconnect the relay, change Appearance, and verify the UI remains
responsive and the local fallback is retained.
### Screenshots / demos
<img width="1733" height="948" alt="image"
src="https://github.com/user-attachments/assets/5afeabaa-0def-482c-9b87-8a880ee0a467"
/>
<img width="700" height="412" alt="Screen Recording 2026-07-29 at 4 39
52 PM"
src="https://github.com/user-attachments/assets/d58da329-aec5-4324-a4b2-cbcb2702a81a"
/>
---------
Signed-off-by: Taylor Ho <taylorkmho@gmail.com>
Signed-off-by: npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6w <52a228d6edf316ec6812ac3c9fc0d696ab59fc7954d77e7be31eedcddf91335b@buzz.block.builderlab.xyz>
Co-authored-by: npub1223z34hd7vtwc6qj4s7flsxkj644nlre2nthu7lrrmkumhu3xddsrx9r6w <52a228d6edf316ec6812ac3c9fc0d696ab59fc7954d77e7be31eedcddf91335b@buzz.block.builderlab.xyz>
OpenClaw connects to a single shared Gateway daemon. Spawning the
default 10 ACP workers per agent is both resource-expensive and
architecturally wrong — each worker opens a separate gateway connection.
Tyler's ruling: cap at 5, lower if needed.
## Contract
Store the requested value (1–32) verbatim at every persistence and wire
boundary. Apply `effective = min(requested, harness_cap)` only at the
four enforcement points:
| Boundary | Implementation |
|---|---|
| Local spawn | `BUZZ_ACP_AGENTS` env var in child `Command` |
| Remote deploy | `launch.policy_env["BUZZ_ACP_AGENTS"]` + legacy
`parallelism` field |
| Restart badge | `SpawnConfigSnapshot.parallelism` stores effective
value; the diff surface displays what actually runs |
| UI copy | Amber hint when requested > cap; no `max` attribute, no
save-path clamp |
`BUZZ_ACP_AGENTS` is added to `RESERVED_ENV_KEYS` — the Desktop resolves
the effective value into `policy_env`; a user-supplied override in `env`
would bypass the cap and is silently stripped.
## Changes
**`managed_agents/parallelism.rs`** (new) — policy core:
- `OPENCLAW_MAX_PARALLELISM = 5`
- `harness_max_parallelism(command)` — keyed on
`normalize_command_identity` so path prefixes, `.exe` suffixes, and
other cosmetic differences are ignored
- `effective_parallelism(command, value)` — identity for uncapped
harnesses
- `acp_agents_value(command, parallelism)` — `env("BUZZ_ACP_AGENTS", …)`
helper
**`runtime.rs`** — spawn clamp: `BUZZ_ACP_AGENTS =
acp_agents_value(effective_command, record.parallelism)`
**`agents_deploy.rs`** — deploy egress clamp: `build_deploy_payload`
resolves `effective_parallelism` once from `descriptor.command`; both
`launch.policy_env["BUZZ_ACP_AGENTS"]` and the legacy top-level
`parallelism` field use that value — the two are always consistent
regardless of stale `record.agent_command` pins
**`spawn_snapshot.rs`** — `from_inputs` stores
`effective_parallelism(&descriptor.command, record.parallelism)` in the
`parallelism` field. Over-cap edits that don't change the pool (e.g. 10
→ 8, both clamp to 5 on OpenClaw) produce equal snapshots; cap crossings
(8 → 3) produce different snapshots.
**`AcpRuntimeCatalogEntry.max_parallelism: Option<u32>`** — derived from
the static definition command, not the probed `entry.command` (which may
be `null` for unavailable entries), so unavailable OpenClaw entries
still carry the cap. Propagated through all four catalog constructors
(builtin discovery, preset catalog construction, custom discovery,
custom-save response), IPC types
(`RawAcpRuntimeCatalogEntry.max_parallelism`), and the frontend catalog
type.
**UI** — `EditAgentAdvancedFields` and `PersonaAdvancedFields` show an
amber hint when `selectedRuntime.maxParallelism` is set and the current
value exceeds it. Cap and label come from the catalog entry — no
hardcoded 5 in TS. No `max` attribute on inputs; the input stays
`type="text"` with 1–32 copy.
**Docs** — `docs/remote-agents.md`: `BUZZ_ACP_AGENTS` moved from the
deliberately-non-reserved section to reserved; new contract documented.
`desktop/src/features/agents/AGENTS.md`: command-keyed execution policy
documented as the sanctioned second metadata source feeding the catalog
projection.
## Tests
**Rust** (`parallelism.rs`):
- `policy_table` — `harness_max_parallelism` and `effective_parallelism`
across all openclaw variants and uncapped harnesses
- `acp_agents_value_openclaw_above_cap_is_capped` — spawn-env seam
- `override_direction_*` — both override directions (openclaw runtime +
goose override; goose runtime + openclaw override)
- `summary_persona_inherited_*` — live persona wins over stale
`agent_command`
- `snapshot_export_carries_requested_definition_parallelism` — requested
value travels wire/sync unchanged
**Rust** (`spawn_snapshot/tests.rs`):
- `openclaw_above_cap_parallelism_snapshots_equal` — stored 10 vs 8,
both clamp to 5 → snapshots equal
- `openclaw_cap_crossing_parallelism_snapshots_differ` — 8 (clamps to 5)
vs 3 → snapshots differ
**Rust** (`discovery/presets.rs`):
- `openclaw_preset_unavailable_carries_max_parallelism` /
`openclaw_preset_available_carries_max_parallelism` — catalog metadata
present with `command: null` and with a resolved path
**Rust** (`agents_deploy.rs`):
- `launch_block_openclaw_over_cap_policy_env_is_capped` — direct
`launch.policy_env` seam
-
`deploy_payload_json_stale_goose_record_live_openclaw_descriptor_both_capped`
— stale `record.agent_command=goose`, live descriptor=openclaw: both
fields cap to 5
-
`deploy_payload_json_stale_openclaw_record_live_goose_descriptor_both_uncapped`
— stale `record.agent_command=openclaw`, live descriptor=goose: both
fields pass through requested
- `deploy_payload_json_explicit_openclaw_override_both_capped` —
explicit `agent_command_override=openclaw`: both fields cap to 5
**Rust** (`persona_events/stale_pin_tests.rs`):
- `apply_persona_snapshot_goose_to_custom_harness_drops_stale_goose_pin`
— custom-direction stale-pin drop (builtin pin → loaded custom harness
via `update_loaded_harness_registry`)
**TypeScript** (`agentParallelism.test.mjs`):
- `parallelismCapHint` — at/below cap (null), above cap (hint includes
label and cap value), singular form for cap=1, uncapped harness (null)
**TypeScript** (`tauri.test.mjs`):
- `fromRawAcpRuntimeCatalogEntry` round-trips `max_parallelism` →
`maxParallelism`; absent when `undefined`
---------
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
`BUZZ_AGENT_MAX_HANDOFFS` compared against the session-cumulative
`handoff_count` (persisted across prompts). After N handoffs a
long-lived session hit the cap permanently: `maybe_handoff()` returned
`Skipped` on every subsequent prompt, the 16 MiB byte-truncation
fallback never bound before a 1M-token provider wall, and the session
wedged on the first 400 with no recovery path. Thufir's session log
shows 8 days of cap-forced truncation before the first
`context_length_exceeded` 400.
The fix replaces the session-level cap comparison with a local
`handoff_attempts` counter constructed at the start of `run()` and
passed into `maybe_handoff()`. The counter resets on every
`session/prompt` turn so `BUZZ_AGENT_MAX_HANDOFFS` caps compaction loops
within a single turn while allowing unbounded compactions across a
session's lifetime. The session-cumulative `handoff_count` is retained
for log context only and is not reset. Steer-driven rounds share the
per-turn budget automatically since steers inject into the running
`run()` loop, not a new call.
- Move `handoff_attempts` increment to before `summarize()` so failed,
empty, and cancelled summarize calls each consume one budget slot — the
cap cannot be bypassed by a repeatedly-failing summarizer
- Upgrade cap-forced `Skipped` from `INFO` to `WARN`; add structured
fields for `session_id`, attempt count, projected tokens, and threshold
so the cap→wall pairing is attributable per session
- Document `max_handoffs` in `config.rs` as a per-`session/prompt`-turn
bound
- Three new behavioral regression tests: per-turn reset proven across
two separate turns; within-turn cap proven via multi-round tool-call
turn; failed summarize proven to burn the attempt budget
Note: this is the proactive half of the context-window fix. The reactive
`context_length_exceeded` 400 recovery path is owned by Sami's branch
(`buzz-ctxfix-sami`, Tyler's crew); this PR is intended to land after
that one.
---------
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
Add `managed_agents::migration::driver`, the transport half of the
relay-canonical migration seam. It is the only code that moves a retained
aggregate across the wire: it consumes one immutable row from
`managed_agent_aggregates`, POSTs the row's `request_json` verbatim to
`/api/managed-agents/aggregate` with fresh owner-signed NIP-98 auth,
strict-deserializes the corrected `migration::AggregateResponse`, verifies it
through `verify_promotion`, and flips the durable row to synced only after a
fully verified read-back.
Design invariants:
- Exact bytes on the wire. The body is `request_json` verbatim; the driver
never re-serializes it, so the stored request and the request the relay
authenticates are the same bytes. The verification source
(`MigrationCandidate`) is reconstructed from those same bytes — parse the
three signed events, decrypt the head under the owner key — so the candidate
and the wire request cannot diverge and no build-side crypto is duplicated.
- Fresh owner NIP-98 per attempt via `build_nip98_auth_header_for_keys` (unique
nonce), so a retry never replays a stale token. Mirrors the `/events`
submit path: egress guard, rate-limit wait, `classify_request_error` /
`relay_error_message` / `parse_json_response`.
- Only the exact retained (generation, private_event_id) is marked synced.
Beyond `verify_promotion`'s response-vs-candidate checks, the driver binds
the *durable row's* coordinate to the verified evidence, so a row whose
columns disagree with its own request_json (and thus the verified head) is
never confirmed.
- Errors preserve retry. Every failure path — transport, non-2xx, malformed
body, reconstruction, verification, coordinate mismatch — persists an
exact-attempt diagnostic via `record_managed_agent_aggregate_error` and leaves
`pending_sync = 1`. Nothing here clears retry except a verified read-back.
Transport tests use a loopback axum stub in the shape of the aggregate route
(6, all green): faithful read-back marks the exact generation synced and posts
verbatim; conflict status preserves retry; malformed success body preserves
retry; tampered read-back fails verification and preserves retry; row
coordinate disagreeing with the verified head is not synced; no pending row is
a no-op with no network submission.
Pure transport confined to the new module + one `mod` line; no delete/tombstone,
docs, or relay-crate changes. `managed_agents::` suite: 1026 passed, 0 failed;
`cargo check --lib` clean.
Co-authored-by: Mongo <5c25403eab7271f9f94ddd4f2b270e8cac2c92e2c830c51877cca6ec974ffb3f@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Add `managed_agents::migration`, the first real consumer of the slice-1/2
authority contract. It converts one hydrated `ManagedAgentRecord` + owner/agent
keys + exact signed kind:30175/30177 projections + CAS coordinate into the
shared `buzz_core::private_managed_agent::Payload` and a signed, inert
kind:30179 aggregate candidate, and verifies the relay read-back byte-exact
before yielding promotion evidence.
Pure by construction: no HTTP, no persistence, no authority mutation. The
transport/driver lane (sibling) owns submission, retry, durable state, and the
actual RelayAuthoritative write. This module is the gate those lanes pass
through, so every "is the relay head a faithful owner-signed copy?" check lives
here behind an adversarial fixture matrix.
Builder (`build_migration_candidate`):
- `PrivateConfig.backend` carries the versioned `VersionedBackend` envelope, not
a bare `BackendKind`, so a shape change is an explicit version bump.
- `auth_tag` is RE-MINTED unconditionally via `compute_auth_tag(owner, agent,
"")`; the stored record string is never copied or trusted (a re-mint cannot
byte-match it, and the codec requires an unconditional attestation).
- Readiness gating derives the ACTUAL largest inserted codec `Value` (the two
projection recovery envelopes dominate) plus the encrypted payload size and
feeds them to `assess_migration_readiness` with the relay's NIP-11 tokens —
never a caller-supplied number.
- Fails fast on inconsistent keys/projections; `build_event` re-runs full codec
validation so any semantic slip surfaces locally, not at the relay.
Verifier (`verify_promotion`): treats the aggregate response as a LOCAL typed
fixture matching the documented route JSON, not a buzz-db type. Checks, in
order: `validate_and_decrypt` of the returned head; owner/agent/generation/
predecessor/state equality; binding event-id + content-hash + recovery equality
against the returned signed projections (and that the served standalone
projection matches its binding, so a swapped public event is caught); definition
revision equality; and source-vs-roundtrip equality for CARRIED classes only
(the payload type structurally carries only PrivateConfig + identity, i.e. the
carried classes — derived/transient/bookkeeping fields never enter it). Returns
`PromotionEvidence` only after every check passes.
Adversarial fixtures (14, all green): happy exact round-trip; re-mint-not-copy;
versioned backend envelope; agent-key/record-pubkey mismatch; projection not
owner-signed; wrong-kind projection; capability-absent block; oversize codec
value block; tampered head ciphertext; head signed by wrong owner; swapped
definition projection; swapped instance projection; wrong definition revision;
stale-generation head. Uses shared codec kind/size constants directly.
Foundation API is `#[allow(dead_code)]` with a named sibling consumer (the
transport lane); prod `cargo check --lib` is warning-free. `managed_agents::`
suite: 1014 passed, 0 failed.
Co-authored-by: Mongo <5c25403eab7271f9f94ddd4f2b270e8cac2c92e2c830c51877cca6ec974ffb3f@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Persist each exact signed kind:30179 aggregate request by generation so
retries reuse immutable bytes and newer tombstones do not erase older
in-flight attempts. Enforce contiguous generations and exact per-attempt
compare-and-clear/error updates.
Add focused coverage for immutable generations, stale acknowledgements, and
diagnostic scoping.
Co-authored-by: Carl <c7ebe626f000404285d3686e1dc74cc07cc60a9754a150041ba132e14bd3e2ec@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Add the migration-readiness contract on top of the authority foundation
(FIX#2, FIX#5). Types + pure gate only — no migration enablement, no
destructive local mutation. The migration driver (sibling lane) is the
consumer.
- Reclassify `auth_tag` PrivateCanonical -> DerivedNotCarried: it is
re-minted at spawn from owner keys + agent pubkey (compute_auth_tag),
never carried in a projection and never diffed at verification. Carrying a
stale tag would pin an agent to legacy forever.
- Add a `VersionedBackend` envelope with `BACKEND_PAYLOAD_VERSION = 1` and a
stable serialized shape; deny-unknown-fields so a newer envelope is
rejected rather than silently truncated. Round-trip + reject-unknown
fixtures.
- Add the `MigrationBlock` failure taxonomy and the pure
`assess_migration_readiness` gate (FIX#5): capability FIRST, then payload
size, then per-value size. Capability is the exact NIP-11 aggregate token
`nip-pma-aggregate-v1` (NIP_PMA_AGGREGATE_TOKEN) matched over the advertised
token set — not a boolean "kind accepted". Size boundaries are
inclusive-safe; exactly-at-limit passes, strictly-over blocks. On any block
the caller MUST stay LegacyOnly and NOT retry until an input changes (no
timer/retry loop).
- `MIGRATION_MAX_VALUE_BYTES = 32_768` tracks the relay per-`Value` codec
limit (`private_managed_agent::MAX_VALUE_BYTES`); it is NOT a keyring floor
or a combined recovery blob. `MigrationBlock::reason()` renders a diagnostic
string for logging/surfacing — the variant is the signal; durable
persistence is owned by the migration driver record, not this gate.
These are foundation contract types; the next slice wires them into a real
payload builder / verification consumer so they prevent drift rather than
sit beside `PrivateConfig.backend: Value` decoratively.
Co-authored-by: Mongo <5c25403eab7271f9f94ddd4f2b270e8cac2c92e2c830c51877cca6ec974ffb3f@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Introduce the Desktop safety foundation for the relay-canonical managed-agent
migration. State + classification only — no migration enablement, no
destructive local mutation. Every existing record defaults to LegacyOnly, so an
upgrade that predates any migration deserializes unchanged and keeps local JSON
+ keyring authoritative.
- Add `managed_agents::authority` with a single internally-tagged
`RelayAuthority` enum (`LegacyOnly` | `RelayAuthoritative { evidence }`), so
impossible states are unrepresentable: authoritative structurally carries a
verified head, legacy structurally cannot. No `{"state":{"state":...}}`
nesting — persisted shape is `{"authority":"legacy_only"}` /
`{"authority":"relay_authoritative","evidence":{...}}`, asserted exactly.
- Add `#[serde(default)] relay_authority: RelayAuthority` to
`ManagedAgentRecord`; a store predating the field deserializes as LegacyOnly.
- Add an exhaustive field→codec-home classification table
(`FIELD_CLASSIFICATIONS`) as a single source of truth, verified against the
ACTUAL PersonaEventContent (30175) / ManagedAgentEventContent (30177)
builders and the 30177 exclusion asserts — not intent:
* start_on_app_launch, auto_restart_on_config_change, provider_binary_path,
persona_team_dir carry through NO codec slot -> TransientLocal.
* avatar_url IS carried by 30175 -> stays DefinitionProjected.
- Tighten the exhaustiveness test to exact set equality: every record field has
a class AND every class key still names a live field (no stale drift).
- Populate `relay_authority` explicitly (`RelayAuthority::legacy()`) at every
production/test constructor rather than relying on a struct-wide default.
Foundation-only classification/authority API is consumed by the not-yet-landed
reconcile/migration lanes; suppression is per-item with a named consumer, not
module-wide.
Co-authored-by: Mongo <5c25403eab7271f9f94ddd4f2b270e8cac2c92e2c830c51877cca6ec974ffb3f@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Introduce an authenticated aggregate endpoint and serializable authority
transaction for encrypted private managed-agent heads and their exact public
projections. Preserve immutable signed recovery bytes, enforce generation and
definition revision CAS, reactivate submitted projections, retire prior heads,
and return a writer-consistent signed aggregate snapshot.
Keep kind 30179 author-only and out of full-text search, reject generic writes,
and advertise the aggregate capability through NIP-11.
Co-authored-by: Princess Donut <68157ebd23b3897c1991015c3038658ea916200c67d3a54620b0754d1b92f6e0@buzz.block.builderlab.xyz>
Signed-off-by: Wes <wesbillman@users.noreply.github.com>
