## Why
Long Buzz threads were rendered as `[Thread Context (13 of 13
messages)]` because the harness counted only the already-limited query
result. That hid older context and could also hide the agent's own prior
reply in busy threads.
## What
- Fetch one extra thread reply as a sentinel so truncated context is
labeled correctly.
- Use a best-effort `/count` call for improved truncated totals when
available, clamped to the sentinel-proven minimum so racy counts cannot
render impossible labels.
- Keep the `/count` path single-attempt with a short timeout and only
add the root to exact totals when the root was actually fetched.
- Fetch and preserve the agent's newest prior reply when it falls
outside the recent window, with exact event-id matching for the
pin/dedup boundary.
- Add parser and fetch-boundary tests for truncation, exact count,
missing root, count-below-minimum clamping, count failure fallback,
distinct fetched-reply lower bounds, agent-reply dedup/pinning, and
serialized query/count filter semantics.
## Risk Assessment
Low-to-medium — limited to buzz-acp prompt context fetching and a small
RestClient helper. If `/count` fails or times out, the code falls back
to the sentinel-derived minimum total rather than failing the prompt.
The synchronous `/count` happens only for truncated thread contexts and
is bounded to one short best-effort attempt.
## References
- Buzz thread: chotchkies-buzz-bombing-flakes /
`7ef71407f1c7a642382c7e48e0c80fb6ca66948890e04d1eb6f1408c3b7278b1`
- Validation at `c1cfd1b16a04a3ac1d1d0d3cf43e1a08508f3532`:
- `cargo fmt -p buzz-acp` ✅
- `cargo test -p buzz-acp test_fetch_thread_context -- --nocapture` ✅ (6
tests)
- `cargo test -p buzz-acp parse_nostr_thread_response` ✅
- `cargo test -p buzz-acp` ✅ (649 unit + 9 lifecycle tests)
- `git diff --check` ✅
- Push was completed with `--no-verify` after pre-push hooks reached
non-code local environment failures: `flutter` missing for
`mobile-test`; Node.js v20.20.2 too old for pnpm/node:sqlite in
`desktop-check` and `desktop-test`. Earlier hook stages passed:
`check-push-org`, `branch-skew`, `rust-tests`, `test`,
`desktop-tauri-checks`.
- Earlier full `./bin/just ci` at
`622ed7eb8807d64e06209101569b1013414af091` ⚠️ passed Rust/desktop/web
stages, then failed in `mobile-test` on unrelated existing mobile test
`ChannelDetailPage keeps follow mode off while a tall newest message
stays visible`; rerunning that single mobile test reproduced the same
failure without touching mobile code.
Generated with Codex
Signed-off-by: npub1m0vvn9qm5md0a080p27qzkm9uaw49e699ukwfq7fc0756xq0y5zqhzhdk2 <dbd8c9941ba6dafebcef0abc015b65e75d52e7452f2ce483c9c3fd4d180f2504@buzz.block.builderlab.xyz>
Co-authored-by: npub1m0vvn9qm5md0a080p27qzkm9uaw49e699ukwfq7fc0756xq0y5zqhzhdk2 <dbd8c9941ba6dafebcef0abc015b65e75d52e7452f2ce483c9c3fd4d180f2504@buzz.block.builderlab.xyz>
## What
Wires genuine provider-reported `total_tokens` through the full
buzz-agent → buzz-acp publish chain so kind-44200 events carry real
per-turn and cumulative totals for OpenAI-backed models, while
preserving all existing behaviour for Anthropic and external harnesses
(goose, claude-code).
## Why
Live prod data showed 0 of 1,934 archived reports carry `totalTokens`.
Both hardcoded `total_tokens: None` in `pool.rs` and the absent field in
`buzz-agent`'s parser are root causes. This is the backend half of a
two-track fix; the display-fallback half lands in
[#2035](https://github.com/block/buzz/pull/2035).
## Changes
**`crates/buzz-agent/src/types.rs`**
- Added `total_tokens: Option<u64>` to `LlmResponse` with an explicit
doc comment that NIP-AM forbids deriving it.
- Added `TurnTotalState` enum (`Unseen | Exact(u64) | Unknown`) with
`fold()` and `exact_value()` — the tri-state accumulator that
distinguishes not-yet-observed from permanently poisoned.
**`crates/buzz-agent/src/llm.rs`**
- `parse_responses` and `parse_openai`: read `usage.total_tokens` from
OpenAI Chat Completions (including Databricks routes) and the Responses
API via `sum_usage`.
- Anthropic: explicit `total_tokens: None` — no genuine total available;
NIP-AM forbids summing categories.
**`crates/buzz-agent/src/agent.rs`**
- Added `turn_total_state: &'a mut TurnTotalState` to `RunCtx`.
- Fold `response.total_tokens` into the accumulator after each
usage-bearing response; non-usage-bearing responses (keepalive/stream
frames) do not poison.
**`crates/buzz-agent/src/lib.rs`**
- Added `accumulated_total_state: TurnTotalState` to `Session` (default
`Unseen`).
- Per-turn state passed to `RunCtx`, folded into session cumulative
after each turn.
- Emits `accumulatedTotalTokens` in `usage_update` only when cumulative
is `Exact(n)`.
**`crates/buzz-acp/src/usage.rs`**
- Added `accumulated_total_tokens: Option<u64>` (serde default) to
`UsageUpdatePayload` — optional for goose compat.
- Added `last_total: Option<u64>` to `SessionState`.
- Added `turn_total_tokens` and `cumulative_total_tokens` to `TurnUsage`
(field-local — never affect `delta_reliable`).
- Derive turn-total delta only when prev and current are both `Some` and
monotonic; absence, decrease, or no baseline leaves only the total delta
null without touching input/output reliability.
**`crates/buzz-acp/src/pool.rs`**
- Replaced both hardcoded `total_tokens: None` in
`publish_agent_turn_metric` with `usage.turn_total_tokens` and
`usage.cumulative_total_tokens`.
## Tests
20 new tests across the four touched files:
| File | Tests |
|------|-------|
| `types.rs` | `TurnTotalState` fold, accumulation, exact_value, default
(7 tests) |
| `llm.rs` | Chat present/absent, Responses present/absent, Anthropic
always-None (5 tests) |
| `usage.rs` | First turn no baseline, second-turn delta, cumulative
decrease (field-local), current absent, goose-shaped deserialization,
baseline absent (6 tests) |
| `pool.rs` | Exact turn+cumulative mapping, null totals never derived
(2 tests) |
`cargo test -p buzz-acp -p buzz-agent` — all passing, 0 failures.
## Scope
Boundary: `crates/buzz-agent/**` + `crates/buzz-acp/**` only. Desktop
unchanged.
`costUsd` explicitly out of scope.
Related: [#2035](https://github.com/block/buzz/pull/2035)
---------
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: npub1g8493u0xfsjrvflg4n08ezd7vec99mnwzlv0qgwpr9d7gvjwhuzqx59rhw <41ea58f1e64c243627e8acde7c89be667052ee6e17d8f021c1195be4324ebf04@buzz.block.builderlab.xyz>
🤖
## Summary
Agent-authored mentions currently depend on matching visible `@Name`
text to channel profiles. That makes notification delivery ambiguous
when names collide or profiles change, and it encourages an extra
post-send lookup just to confirm that the intended `p` tags were
emitted.
This change makes `buzz messages send` mirror Desktop's existing model:
the message keeps a readable name in its content while the recipient
pubkey is supplied separately.
```bash
buzz messages send \
--channel <UUID> \
--content '@Alice could you review this?' \
--mention <alice-hex-or-npub>
```
`--mention` is repeatable. The CLI normalizes and deduplicates explicit
pubkeys, merges them with any names it can resolve from the channel, and
gives explicit identities priority under the existing 50-mention limit.
Before uploading attachments, signing, or publishing, the command checks
every resulting pubkey against the channel's current membership:
- Members are mentioned normally.
- Non-members stop the send and produce an actionable error.
- `--allow-non-member-mentions` deliberately sends notifying `p` tags
without adding anyone to the channel.
Sending a message never changes membership. On success,
`mention_pubkeys` is read from the exact signed event and returned with
the relay response, so callers can verify the emitted recipients without
another query.
Managed-agent guidance teaches this single-command mention flow. Desktop
mention behavior and the Nostr event schema are unchanged. Forum
guidance is intentionally handled separately in #3596.
### Related issue
None found. This replaces the earlier guidance-only approach in this PR
with the underlying CLI behavior it required.
### Testing
- `cargo test -p buzz-sdk`
- `cargo test -p buzz-cli`
- `cargo test -p buzz-acp`
- `cargo test --manifest-path desktop/src-tauri/Cargo.toml`
---------
Signed-off-by: npub1fdupjvyregj3z2tx7gx5x6py04zw89jm5usef9lyea4f3vcgh8qq9zgkdz <4b78193083ca25112966f20d4368247d44e3965ba7219497e4cf6a98b308b9c0@buzz.block.builderlab.xyz>
Signed-off-by: npub13n66s06epmqf2kc3v373ez8hj65cuzyvxzjf93vwpervxqn2u7jq2qd9je <8cf5a83f590ec0955b11647d1c88f796a98e088c30a492c58e0e46c3026ae7a4@buzz.block.builderlab.xyz>
Co-authored-by: npub1fdupjvyregj3z2tx7gx5x6py04zw89jm5usef9lyea4f3vcgh8qq9zgkdz <4b78193083ca25112966f20d4368247d44e3965ba7219497e4cf6a98b308b9c0@buzz.block.builderlab.xyz>
Co-authored-by: npub13n66s06epmqf2kc3v373ez8hj65cuzyvxzjf93vwpervxqn2u7jq2qd9je <8cf5a83f590ec0955b11647d1c88f796a98e088c30a492c58e0e46c3026ae7a4@buzz.block.builderlab.xyz>
## Summary
[block/buzz#3144](https://github.com/block/buzz/pull/3144) injected
`GOOSE_ACP_SCHEDULER_DISABLED=true` into every `AcpClient::spawn` call
as a forward-compatible no-op, intended to suppress the cron scheduler
in goose ACP children once the matching reader landed in goose. That
reader only ever existed in
[aaif-goose/goose#10738](https://github.com/aaif-goose/goose/pull/10738),
which was closed unmerged.
[goose#10781](https://github.com/aaif-goose/goose/pull/10781) (Lifei
Zhou, merged 2026-07-29) disables the ACP scheduler by default at the
source: `goose acp` now requires `--enable-scheduler` to start a
scheduler. Buzz-spawned children therefore get no scheduler with zero
configuration — making the `GOOSE_ACP_SCHEDULER_DISABLED` injection
permanently dead code.
## What changes
Removes from `crates/buzz-acp/src/acp.rs`:
- `GOOSE_SCHEDULER_DISABLED_ENV` constant
- `cmd.env(GOOSE_SCHEDULER_DISABLED_ENV, "true")` injection in
`AcpClient::spawn`
- `spawn_injects_scheduler_disabled_env_by_default` test
- `spawn_scheduler_disabled_env_overrides_conflicting_extra_env` test
- `spawn_and_read_child_env` helper (unreferenced once the two tests
above are gone)
No other files are affected.
## Why now
Leaving dead code that references an env var no reader will ever consume
misleads future maintainers about the actual scheduler-isolation
mechanism. The isolation is now an upstream default, not a Buzz
injection.
Reverts: [block/buzz#3144](https://github.com/block/buzz/pull/3144)
Related:
[aaif-goose/goose#10781](https://github.com/aaif-goose/goose/pull/10781)
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
## Summary
Makes Gemini — and every other non-Claude, non-GPT-5 model on the
Databricks MLflow route (`databricks_v2`) — usable in an agent loop.
These are the non-`benchmarks/` changes from
`benchmark/harness-accounting-and-solo`, lifted onto a clean base off
`main` so they can land independently while the harness work continues.
Two defects made these models unusable, one fatal and one silent. Both
live only in `openai_body` / `parse_openai`, which is the
least-exercised of the three `databricks_v2` sub-routes — the
`luna`/`sol` conditions run the Responses route and the `opus`
conditions run the Anthropic route, so **this change is inert for every
model already in use** and only lights up the MLflow path.
## Why a third route at all
`databricks_v2_route_for_model` buckets by model family: `claude*` →
Anthropic Messages, `gpt-5`/code-names → OpenAI Responses, **everything
else → MLflow chat-completions**. Gemini, Qwen, gpt-oss, and friends all
fall through to that third pair — and both bugs below live only there.
## D1 — dropped thought signatures (fatal)
Gemini returns a `thoughtSignature` on every tool call and **requires it
echoed back**. `openai_body` reserialized each call as `{id, type,
function}` only, dropping the field, so the next request 400'd:
```
HTTP 400 Function call is missing a thought_signature in functionCall parts.
```
For a coding agent this fires on the **first** tool call, so the model
never completes a single turn.
**Position is load-bearing.** A four-shape replay probe against the live
gateway established that the signature must sit as a *sibling* of
`function` — nesting it inside `function{}` fails with the *same* 400 as
omitting it. A fix that "preserves the field" without preserving its
position passes a unit test and still 400s.
The fix: `ToolCall` gains `provider_extra: Map<String, Value>`.
`parse_openai` captures every top-level wire key except the three we
model (`id`, `type`, `function`); `openai_body` re-emits them beside
`function`. Keeping *whatever we did not model*, rather than naming
`thoughtSignature`, means the next provider with an opaque per-call
token needs no change here. The Responses and Anthropic replay shapes
are fully modelled, so they pass `Default::default()` and stay
**byte-identical** to before.
### D1b — duplicate tool-call ids (same root cause)
Gemini returns the **function name** as the id, so two parallel calls to
one function arrive sharing an id — and that id is what pairs a
`role:"tool"` result back to its call, making two results
indistinguishable. `dedupe_provider_ids` suffixes collisions
(`get_weather`, `get_weather-2`). Safe because both halves of the
pairing (the assistant `tool_calls[].id` and the result's
`tool_call_id`) are re-emitted from this same value; the provider never
sees its original id again.
## D2 — block-array content discarded (silent, worse than a crash)
`parse_openai` read `content` with `as_str()`, which returns `""` for
anything that isn't a JSON string. Gemini (and Qwen35, gpt-oss) send an
array of typed blocks:
```json
"content": [
{"type": "reasoning", "summary": [{"type": "summary_text", "text": "…"}]},
{"type": "text", "text": "391"}
]
```
So the model answered and the answer was thrown away — no error, no
warning, just a turn that looked like the model had said nothing. On a
benchmark this reads as "Gemini is bad at the task" rather than "buzz
dropped the reply."
`openai_content_parts` now accepts either shape — string as before, or a
block array where `text` blocks concatenate into text and `reasoning`
blocks into reasoning (Gemini nests the prose one level down under
`summary`). Message-level `reasoning_content` / `reasoning` still win
when present, so DeepSeek and vLLM-style hosts are unchanged; block
reasoning is the last fallback.
## Also: a turn-start log line (`buzz-acp` `pool.rs`)
Small, independent observability change that also rides in the
non-benchmark delta: `run_prompt_task` now emits a `pool::prompt` "turn
starting" line, labelled by the same `prompt_label` helper as
`log_stop_reason`, so a log reads as start/stop pairs. An unpaired start
is the only durable evidence that a turn was entered and never returned
— without it, a stalled agent and an agent nobody woke leave identical
(zero-completion) logs.
## Interaction with #3538#3538 (already merged) rewrote `databricks_v2_route_for_model` to route
by boundary-aware model-family segments. That change and this one touch
**different functions** in `llm.rs` — routing vs. body/parse — and
compose cleanly; the family routing decides *which* pair runs, and this
fixes the MLflow pair it can now select.
## Testing
- `cargo fmt --all -- --check`, `cargo clippy -p buzz-agent -p buzz-acp
--all-targets -- -D warnings` — clean.
- `cargo test -p buzz-agent -p buzz-acp` — all green (304 + 632 lib
tests plus integration suites, 0 failures). Five new tests cover:
block-array text extraction, plain-string regression, passthrough
capture (and non-duplication of the modelled keys), replay position
(`thoughtSignature` beside `function`, not inside it), and id
de-duplication.
- Wire evidence: the four-shape replay table and the reasoning-effort
probe were run against `block-lakehouse-staging` (recorded in the design
doc).
## Relationship to the benchmark branch
The full design write-up (four-shape replay table, position-matters
analysis, effort verification, and open pricing item) lives in
`docs/08-gemini-provider-fixes.md` on
`benchmark/harness-accounting-and-solo`. The benchmark manifests and
endpoint-config entries that exercise these models are separable and
stay on that branch.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Signed-off-by: Atish Patel <atish@squareup.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
clap renders live env var values in help text by default. Three args
carrying credentials were exposed this way:
- `BUZZ_PRIVATE_KEY` in `buzz-cli` (`crates/buzz-cli/src/lib.rs`)
- `BUZZ_AUTH_TAG` in `buzz-cli`
- `BUZZ_PRIVATE_KEY` in `buzz-acp` (`crates/buzz-acp/src/config.rs`)
Add `hide_env_values = true` to each. Env var names remain visible for
discoverability; only their runtime values are withheld from `--help`
output.
Also adds a regression guard in each crate's test module that walks the
clap command tree (recursing into subcommands for `buzz-cli`) and
asserts every arg whose env var name contains `KEY`, `SECRET`, `TOKEN`,
`PASSWORD`, `CRED`, or `AUTH` has `hide_env_values` set. This prevents
future credential-bearing args from being added without the masking in
place.
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
> On 8 tasks matched by name across the two runs, cost fell $8.36 →
$1.77 (4.71×) and wall-clock 12,423 s → 1,085 s (11.45×).
## Summary
Two independent, self-contained fixes to `buzz-agent`/`buzz-acp`, split
out of the benchmark branch so they can land while the harness work
continues:
1. **Request and surface Anthropic prompt caching.** buzz never sent a
`cache_control` breakpoint, so on the Databricks Anthropic route
`cache_read_input_tokens` was **structurally always 0** and the ~10×
cache-read discount was never claimed. This teaches `anthropic_body()`
to mark the cacheable prefix, and plumbs the cache split end-to-end so
accounting can price it.
2. **Pass proxy + TLS-trust env into MCP tool subprocesses**, so agent
tools on a proxy-only host stop reporting a live network as offline.
## Why the caching gap matters
The Anthropic Messages API does **not** cache unless the request carries
a `cache_control` breakpoint, and the Databricks AI Gateway — a
third-party proxy in front of the model, in the same category as
Bedrock/Vertex — does **not** auto-cache (only the first-party Anthropic
API and Claude-on-AWS do zero-config caching). So every request was
billed cold.
Measured live against the Databricks gateway
(`databricks-claude-opus-5`, 2026-07-28), the same call with and without
a single `cache_control` marker:
| Run | `input_tokens` | `cache_creation` | `cache_read` | latency |
|---|---|---|---|---|
| No `cache_control`, two byte-identical calls | 121,625 | 0 | **0** |
~9.3 s |
| With one marker — cold (write) | 4 | 121,625 | 0 | 9.3 s |
| With one marker — warm (read) | 4 | 0 | **121,625** | **4.5 s** |
One marker moved 121,625 tokens from full-price input to a 0.1× cache
read and roughly halved latency (a clean, isolated ~2.07× prefill
speedup on this single-threaded microbenchmark). The gateway honours
`cache_control`; buzz simply never sent it.
At fleet scale this was a real budget item. Across matched
Terminal-Bench solo sweeps (89 tasks, `-n 20`, before the fix), the two
OpenAI-route models independently landed at ~86–87% cache reads — the
expected shape for an agentic loop, where system + tools + append-only
history repeat every turn — while the Anthropic route returned a hard 0%
on every receipt:
| Condition | Route | Input tokens | Cache reads | Cost | Cost if
uncached | Discount |
|---|---|---|---|---|---|---|
| luna (`gpt-5-6`) | OpenAI | 20,320,818 | **17.7M (87.0%)** | $6.96 |
$22.87 | **3.28×** |
| sol (`gpt-5-6`) | OpenAI | 22,312,290 | **19.2M (85.9%)** | $37.07 |
$123.35 | **3.33×** |
| opus (`claude-opus-5`) | Anthropic | 12,459,822 | **0 (0.0%)** |
$81.31 | $81.31 | **1.00×** |
Applying luna's measured 87% read rate to the opus token counts at list
prices (`input $5/M`, `cached_input $0.5/M`, `output $25/M`) puts the
opus run at **~$32.53 vs the $81.31 actually paid — a ~60% overspend on
those 49 trials (~$89 on a full sweep)**. That is an upper bound (it
prices every cached token at the 0.1× read rate and ignores the 1.25×
write premium), and the opus discount is structurally smaller than
luna/sol's because opus emits ~3.5× more uncacheable output per trial,
which sets a floor on what caching can recover.
There is also a plausible **second-order effect**: Databricks appears to
meter its per-minute rate limit on *uncached* input tokens, so the
missing cache also cost rate-limit headroom — the opus endpoint lost 63%
of its trials to fatal 429s while running alone at one-third of a GPT
endpoint's raw throughput. This is a hypothesis, not a proven mechanism
(the only zero-cache condition is also the only Anthropic endpoint), but
it is the reading that explains the throttling with one rule instead of
two.
## Post-fix results (provisional — first trials of an in-flight re-run)
On 8 tasks matched by name across the two runs, cost fell **$8.36 →
$1.77 (4.71×)** and wall-clock **12,423 s → 1,085 s (11.45×)**.
| Metric | before (`4a955a858`) | after (`3bef1f6a`) |
|---|---|---|
| Cache reads as % of input | **0.0%** | **78.7%** (still climbing
toward the ~86% steady state) |
| `cost_usd_no_cache_discount / cost_usd` | **1.00×** | **2.18×**
(tracking the projected ~2.5×) |
| Trials with a fatal 429 (same `-n 20`) | **63%** | **15–19%** |
To be clear about attribution: **~2× of that is the clean prefill saving
from caching itself**; the rest is second-order — cached requests burn
far less rate-limit budget, so they stall less and redo less destroyed
work. The 11.45× is a system-level result specific to this throttled
workspace, not a caching benchmark. Quality held (7/8 solved in each
run). A controlled low-`-n` A/B (neither arm hitting a 429), which the
`BUZZ_AGENT_PROMPT_CACHING` opt-out exists to enable, is still owed
before this becomes a published claim.
## What changed
### 1. Request caching (`llm.rs`, `config.rs`)
`anthropic_body()` emits ephemeral `cache_control` breakpoints, gated by
`BUZZ_AGENT_PROMPT_CACHING` (**default on**, `=0` to opt out):
- **Static prefix** — marker on the `system` block. Prefix order is
`tools → system → messages`, so this single marker caches **tools +
system** together. Byte-identical on every turn of a run, and survives a
context handoff (system/tools come from cfg/mcp, not `self.history`).
- **Rolling tail + leapfrog** — marker on the last block of the last
**two** messages. The append-only history re-reads the prior turn's
prefix from cache; marking two messages (not one) keeps consecutive
breakpoints inside Anthropic's **20-block lookback window** even as tool
parallelism rises, avoiding a silent full-price miss.
An empty system prompt stays a bare string (Anthropic rejects empty text
blocks), and below-threshold prefixes are silently not cached, so the
flag is safe on by default.
### 2. Surface the cache split end-to-end — the plumbing (`types.rs`,
`llm.rs`, `agent.rs`, `lib.rs`, `usage.rs`, `acp.rs`)
This is the part that makes gaps like the one above **visible** instead
of silent. A consumer that prices all of `input_tokens` at the full rate
can't tell a route that's caching from one that isn't — the total looks
right either way. So:
- `LlmResponse` gains `cached_input_tokens` (a **subset** of
`input_tokens`, never an addition); `parse_anthropic` / `parse_openai` /
`parse_responses` each populate it.
- A `usage_first()` helper reads the cache count wherever a provider
hides it — flat `cache_read_input_tokens` (Anthropic),
`prompt_tokens_details.cached_tokens` (OpenAI chat),
`input_tokens_details.cached_tokens` (Responses) — taking the **first
present value, never a sum**. Reading only flat keys is exactly why the
OpenAI route's nested `cached_tokens` had *also* been going unclaimed:
`prompt_tokens` is already inclusive, so the total looked correct while
the discount silently went unreported.
- The per-turn/per-session accumulators and the goose `usage_update`
payload now carry `accumulatedCachedInputTokens`; `buzz-acp`
deserializes it (`serde` default `0` for goose, which doesn't send it)
and logs `cached=<n>`.
### 3. Fix a Databricks MLflow-route double-count (`llm.rs`)
The Databricks MLflow route reports the flat Anthropic-spelled
`cache_read_input_tokens` *alongside* an already-inclusive
`prompt_tokens`, so the old code summed them and nearly doubled the
count — inflating both the context-budget gate and cost.
`openai_chat_input_tokens()` now reads `prompt_tokens` alone. Verified
on a live `databricks-glm-5-2` response where `prompt_tokens +
completion == total` proves inclusivity. (Anthropic's native route
genuinely *excludes* the cache fields and is still summed — the two
never collide, because `claude*` models route to the Anthropic path.)
### 4. Proxy + TLS-trust passthrough into MCP tools (`mcp.rs`) —
independent fix
`buzz-agent` `env_clear()`s each MCP child, and the allowlist carried no
proxy/TLS vars. On a proxy-only host that doesn't degrade the tools, it
**blinds** them: apt, curl, pip, git connect directly, the egress
firewall resets the socket, and the agent reports "Connection reset by
peer" — indistinguishable from a genuinely offline task. Adds both
spellings of `HTTP(S)_PROXY`/`NO_PROXY`/`ALL_PROXY` (curl/git read
lowercase; Go/Python read uppercase; libcurl ignores uppercase
`HTTP_PROXY`) plus `SSL_CERT_FILE`/`SSL_CERT_DIR` for TLS-terminating
proxies that present their own CA.
## Testing
- `cargo fmt --all -- --check`, `cargo clippy -p buzz-agent -p buzz-acp
--all-targets -- -D warnings` — clean.
- `cargo test -p buzz-agent -p buzz-acp` — **all green** (632 + 299 lib
tests plus integration suites, 0 failures). New tests cover: the three
breakpoints and the disabled/empty-system/single-message edge cases;
nested-vs-flat cache parsing for all three routes; the Databricks
inclusive-`prompt_tokens` fix; wire deserialization of
`accumulatedCachedInputTokens`; and the proxy/TLS passthrough allowlist.
- Pre-push lefthook suite green (branch-skew, rust-tests, test,
desktop-check/test/tauri).
## Relationship to the benchmark branch
These are the non-`benchmarks/` changes from
`benchmark/harness-accounting-and-solo`, lifted onto a clean base off
`main` so they can merge independently.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Signed-off-by: Atish Patel <atish@squareup.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
## Summary
- add a generic per-runtime env-defaults table,
`config::default_agent_env()`, mirroring the existing
`default_agent_args()` / `codex_network_env()` precedent, and merge it
once in `AcpClient::spawn` with the established precedence: **runtime
defaults < persona `extra_env` < inherited parent env**
- first (and only) row: Buzz-owned Hermes processes get
`HERMES_ACP_SKIP_CONFIGURED_MCP=1`, so Hermes does not preload unrelated
profile-configured MCP servers before answering ACP `initialize` (fixes
the 10s model-discovery timeout in #3355 — Buzz supplies session MCP
servers explicitly through `session/new`, per Hermes's documented
host-integration contract for this variable)
- normalize Windows `.cmd`/`.bat` shims alongside `.exe` in
`normalize_agent_command_identity` (npm installs resolve to those
wrappers)
- switch the `extra_env` parent-presence check from `var()` to
`var_os()` so non-UTF-8 parent values are honored
Replaces the runtime-specific approach in #3386: same behavior, but the
mechanism is generic runtime spawn metadata in `config.rs` rather than a
Hermes/ACP special case in `acp.rs`, and the seam covers every launch
path (Desktop spawn, `buzz-acp models`, CLI) because they all funnel
through `AcpClient::spawn`. ~15 lines of production code.
Fixes#3355
## Testing
- `cargo test -p buzz-acp` — **639 passed, 0 failed** (full package,
includes the new `default_agent_env_recognizes_hermes_identities` unit
test and `spawn_applies_runtime_env_defaults_with_extra_env_precedence`
integration test covering default injection, extra_env override, and
non-Hermes exclusion)
- `cargo fmt --all -- --check`, `cargo clippy -p buzz-acp --all-targets
-- -D warnings` — clean
- live-local with real Hermes v0.19.0 (`hermes-acp`): `buzz-acp models`
returned **13 models / currentModelId in 2.6–3.0s** (was a 10.0s timeout
on the first cold run without isolation); a wrapper probe confirmed the
child received `HERMES_ACP_SKIP_CONFIGURED_MCP=1` by default and `0`
when the parent env set it explicitly (operator wins)
- lefthook pre-push suite green: rust-tests, desktop-check,
desktop-test, desktop-tauri-test, mobile-test, branch-skew
No UI changes; subprocess environment behavior only.
Signed-off-by: npub1qyvc0c5kl4gqv2fd97fsk46tu378sqgy35vc83rvgfwne90sel7s0ed67d <011987e296fd5006292d2f930b574be47c7801048d1983c46c425d3c95f0cffd@buzz.block.builderlab.xyz>
Signed-off-by: Tyler Longwell <tlongwell@block.xyz>
Co-authored-by: npub1qyvc0c5kl4gqv2fd97fsk46tu378sqgy35vc83rvgfwne90sel7s0ed67d <011987e296fd5006292d2f930b574be47c7801048d1983c46c425d3c95f0cffd@buzz.block.builderlab.xyz>
Co-authored-by: Tyler Longwell <tlongwell@block.xyz>
Co-authored-by: mr-r0b0t.eth <adam.manning@pro-serveinc.com>
Mid-turn steering was reachable only through goose's
`_goose/unstable/session/steer`, which requires an `expectedRunId`
sourced from `_meta.goose.activeRunId`. claude-agent-acp and codex-acp
never emit a run id, so every mid-turn mention to those harnesses bailed
at the run-id guard before writing a byte and degraded to cancel +
merge, destroying in-flight tool calls.
Both adapters ship `_session/steering` (params `{sessionId, prompt}`,
result `{outcome}`) and advertise it as `_meta.steering.supported` on
the `initialize` response. This adds it as a second steer transport
selected at write time, reusing the existing withhold/release, ack
routing, and cancel+merge fallback machinery unchanged.
## Transport selection
| `active_run_id` | `steering_supported` | Transport |
|---|---|---|
| `Some(run_id)` | any | `_goose/unstable/session/steer` +
`expectedRunId` (unchanged) |
| `None` | `true` | `_session/steering` with `{sessionId, prompt}` |
| `None` | `false` | ack `ExpectedRunIdMissing`, write nothing
(unchanged) |
goose keeps priority when both are present — `expectedRunId` is strictly
more precise about *which* run is being steered.
## Two load-bearing safety properties
**The advertised capability is the only gate — never error-code
probing.** codex-acp's `extMethod` answers unrecognized extension
methods with a bare `{}`, which is a JSON-RPC *success* rather than
`-32601`. Buzz maps a steer success to `queue.remove_event`, so probing
an unknown method would silently delete the user's message with no
error, no fallback, and no log line.
**An `outcome` must be positively recognized.** Only `injected` and
`startedNewTurn` count as delivery. Anything else — codex's `failed`, an
unknown value, or a missing `outcome` entirely — is
`SteerError::OutcomeRejected`, which releases the withheld event and
fires the cancel+merge fallback. This makes the silent-loss path above
unreachable even if an adapter mis-advertises.
`startedNewTurn` acks `Success`, because the message really was
delivered and must not be redelivered, but deliberately does **not**
renew the read loop's hard deadline: the turn Buzz was awaiting had
already settled, and renewing would extend the clock on a finished turn.
## Notes for reviewers
- `SteerError::OutcomeRejected` needs no new arm in the
`PoolEvent::SteerAck` match — the existing catch-all
`Ok(SteerAck::Err(_)) => (true, false, true)` already gives release +
fallback, and the two `AgentError` arms above it match that variant
specifically, so they do not shadow it.
- Comments that described the old goose-only "try-and-tolerate" `-32601`
behavior are corrected; that assumption was never valid for codex-acp.
- No CI job runs `buzz-acp` tests. The full package suite was run
locally: **617 passing, 0 failing**.
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
A Buzz install with a scheduled goose recipe fires each cron entry once
per `goose acp` child instead of once, because every child
unconditionally starts its own cron scheduler over the shared
`~/.local/share/goose/schedule.json`. With a pool of N children per
harness and multiple harnesses, one scheduled recipe fans out to N ×
harness_count executions — each running under the managed agent's
identity rather than the operator's, and racing the operator's own
standalone goose over the same schedule file.
This injects `GOOSE_ACP_SCHEDULER_DISABLED=true` into every child
spawned by `AcpClient::spawn`, so a managed agent never owns the
operator's cron schedule.
## Placement
The `cmd.env` call is set last — after the `extra_env` operator-wins
loop and after the `CODEX_CONFIG` merge — deliberately with no escape
hatch. Managed children not running the operator's schedule is a
correctness invariant rather than an operator-tunable default, so the
injection must beat both a conflicting persona `extra_env` entry and any
value inherited from the parent process.
It is injected for all agents, not just goose. Agent builds that don't
recognize the variable ignore it.
## Sequencing
The goose-side flag that reads this variable and skips scheduler startup
lands separately (repo TBD). Until it does, this change is a
forward-compatible no-op: it sets an environment variable nothing
currently reads. Merging it first means no coordinated release is needed
— the fix takes effect as soon as the goose side ships.
Related: https://github.com/aaif-goose/goose/pull/10738
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
## Summary
Fixes#2794. Related: #2692.
`resolve_model_switch_method()` reads the `configId` key from each
`session/new` `configOptions` entry and skips entries that lack it.
`claude-agent-acp` (v0.61.0) keys its entries with `id`, so every
model-category entry was skipped, the desired model never matched, and
Claude Code sessions fell back to the CLI default from the user's
`~/.claude/settings.json`. The only trace was a `pool::model` WARN that
never reaches the per-agent log files.
This is the ACP-side half of the symptom reported in #2692. The open
desktop-side PRs (#2695, #2701, #2696) inject `ANTHROPIC_MODEL` at
spawn, which masks the problem for spawn-time selection but leaves the
config-option switch path broken.
## Changes
- `resolve_model_switch_method()` accepts either `configId` or `id` when
extracting the config id. The set request is unchanged: the ACP SDK
schema takes `configId` as the request param and the adapter resolves it
against its `id`-keyed entries, so only the read side needed fixing.
- Regression test with an `id`-keyed `configOptions` payload mirroring
the real adapter response (including `models: null`, so the unstable
fallback path cannot rescue the match).
- Doc comment on `extract_model_config_options()` notes the key drift.
## Testing
`cargo test -p buzz-acp --lib`: 599 passed, 0 failed. The new test fails
on main and passes with this change.
Verified against the real adapter: a stdio JSON-RPC probe of the bundled
`claude-agent-acp` 0.61.0 confirms `session/new` returns `id`-keyed
config options with `opus[1m]` present as a value, and the SDK's
`SetSessionConfigOptionRequest` schema accepts `{sessionId, configId,
value}` as sent by `session_set_config_option()`.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Signed-off-by: chillerno1 <gh.chiller@pm.me>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
ACP harnesses that name a session from the first text they receive all
land in the same place: every managed Buzz agent opens with the
identical `[Base] You are operating inside the Buzz platform…` framing,
so the harness session list shows a wall of indistinguishable rows.
Because sessions are keyed per channel, one agent active in several
channels produces several of them.
This sends the name out of band instead. `session/new` carries
`_meta.sessionTitle` with `Agent · #channel`, composed from the agent's
`display_name` (or its unique `name` handle) and the channel it is
serving. The prompt is untouched — no tokens spent, no perturbation of
the prompt contract, and nothing new for the desktop observer's section
parsing to handle.
The mechanism is harness-agnostic: Buzz sends the field on every ACP
`session/new` regardless of which harness is behind it, and adapters
that don't read it ignore it per spec.
## Inert until a consuming adapter ships
ACP adapters ignore `_meta` members they do not recognize, so against an
adapter with no reader a Buzz session gets no title and nothing else
changes. Three adapter halves consume it — Codex, Goose, and Claude Code
(linked below); this half and each reader are only useful together, and
each reader lands independently.
No version floor is added. `codex_adapter_is_outdated_with_path` already
gates codex-acp on major version `>= 1`
(`desktop/src-tauri/src/managed_agents/discovery.rs:1276-1284`) and this
feature needs nothing above that — an older adapter is not broken by the
extra member, it simply ignores it.
## What changes
**`crates/buzz-acp`** owns sanitization and composition.
`sanitize_session_title` collapses whitespace, drops control characters,
and caps at `SESSION_TITLE_MAX_CHARS` (80) by character, not byte, so a
multi-byte character cannot be split. `compose_session_title` truncates
only the channel part against that cap, so the agent name always
survives; when the agent name alone fills the cap the channel is dropped
rather than the name. `session_new_full` sets `_meta.sessionTitle` when
a title exists and omits `_meta` entirely when it does not, since an
adapter may distinguish an absent member from a null one.
**`desktop/src-tauri`** only resolves and exports.
`resolve_session_title` picks `display_name` or falls back to `name`,
and `spawn_agent_child` writes it to `BUZZ_ACP_SESSION_TITLE` — or
removes the variable when neither candidate yields anything printable.
DMs, unresolved channels, and heartbeat sessions get the bare agent name
with no channel suffix.
## Four properties that are easy to remove by accident
**Control characters are stripped at the desktop boundary, not in the
harness.** An interior NUL cannot cross the environment boundary at all
— `Command::env` fails the entire spawn rather than passing it through.
Deferring the strip to `buzz-acp` would let a corrupted display name
turn display chrome into a spawn failure. A display name that is *only*
control characters falls back to `name`.
**The title is hashed into `spawn_config_hash`.** Without it, renaming
an agent left the running process with a stale title and no restart
badge. The hash runs the same `resolve_session_title` the spawn writes,
and skips it when a user env override shadows `BUZZ_ACP_SESSION_TITLE` —
spawn writes the title *before* the layered user env, so the override is
what actually runs, and it already reaches the hash through
`descriptor.env`. Hashing the record-derived value under an override
would badge a rename that changes nothing.
**One channel resolve serves both consumers.**
`resolve_new_session_channel_context` returns `(is_dm, title_channel)`
from a single metadata lookup, feeding both the canvas block's DM check
and the title. `ChannelInfoResolver` caches only `Some`, so two
independent calls against an unresolvable channel pay the full
`fetch_channel_info` retry sequence twice — two timeouts plus a retry
delay each — directly in front of `session/new`, precisely when the
relay is already degraded.
**The `"unknown"` channel name is treated as absent.**
`fetch_channel_info` substitutes the literal `"unknown"` for a metadata
event with no `name` tag. Composing that sentinel would title every
unnamed channel `Agent · #unknown`, reintroducing the exact collision
the suffix exists to remove while naming a channel something it isn't.
The startup cache already refuses `channel_type == "unknown"` for the
same reason.
Closes#2334
Related — the adapter halves that consume `_meta.sessionTitle`:
-
[codex-acp#338](https://github.com/agentclientprotocol/codex-acp/pull/338)
— Codex
-
[aaif-goose/goose#10712](https://github.com/aaif-goose/goose/pull/10712)
— Goose
-
[claude-agent-acp#920](https://github.com/agentclientprotocol/claude-agent-acp/pull/920)
— Claude Code
---------
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Agent commits were authored by a raw 63-character npub, which makes `git
log`, `git blame`, and GitHub's author column effectively unreadable.
This uses the agent's display name for `user.name` instead, while
leaving the pubkey where it does real work.
## What changes
`build_git_env` in `crates/buzz-dev-mcp/src/shim.rs` now reads
`BUZZ_ACP_DISPLAY_NAME`, sanitizes it, and uses the result as
`user.name`. When the variable is absent or unusable it falls back to
`info.npub` — byte-identical to today's behavior.
`user.email`, `user.signingkey`, and the whole credential/signing block
are untouched. The pubkey is what NIP-98 auth, NIP-GS signing, and
contributor matching key on, and it stays in the email verbatim.
`crates/buzz-acp/src/lib.rs` forwards the variable into the dev-mcp
server's declared env, mirroring the existing `BUZZ_AUTH_TAG` block. It
reads `std::env::var` directly rather than going through `Config`, so
the variable is picked up whenever the process has it.
`crates/buzz-agent/src/mcp.rs` adds one `PASSTHROUGH_ENV` entry so ACP
clients that spawn `buzz-agent` without declaring the variable on the
wire still propagate it.
## Why a dedicated variable
`BUZZ_ACP_DISPLAY_NAME` is its own contract rather than a reuse of the
ACP session title. Commits outlive sessions: a session title is
per-session UI chrome and may be composed downstream into `Agent ·
#channel`, and if that composed form ever reached the env var, git
attribution would change silently with no test able to catch it. Git
identity gets a variable whose contract is "bare agent display name,
never channel-qualified."
Nothing writes it yet — a one-line Desktop write lands as a follow-up.
Until then `std::env::var` returns `Err`, the npub fallback fires, and
behavior is byte-for-byte current `main`.
## Sanitizing
Strip control characters, Unicode format characters, and angle brackets;
collapse whitespace runs, trim, cap at 80 characters (by `chars()`, so a
multi-byte name is never split mid-UTF-8).
Angle brackets go because git drops them silently rather than erroring:
`Duncan <evil@x.com>` renders as `Duncan evil@x.com <hex@relay>`. It
forges nothing, but it reads as though it might.
The empty result also has to cover more than literal emptiness. git's
`ident.c` treats a set of characters as "crud" — stripped from both
ends, and fatal when a name is *nothing but* those characters:
```
$ git -c user.name=';;' commit -m t
fatal: name consists only of disallowed characters: ;;
```
Verified against git 2.54.0 by committing with each ASCII byte 32..=126
as the entire `user.name`: exactly space, `"`, `'`, `,`, `:`, `;`, `<`,
`>`, `\` abort, plus all control characters (the predicate is `c <=
32`). `.` is not crud in this version, despite older lore. Names that
merely *contain* crud are fine — `O'Brien` and `Smith, Jr.` both commit
cleanly — so the check is "at least one non-crud character survives,"
not "no crud present." Without it, a display name of `;;` or `""` would
abort every commit that agent makes.
## Unicode format characters
`char::is_control` covers only category `Cc`. Category `Cf` — zero-width
spaces and joiners, bidi embedding and override marks, invisible math
operators, tag characters — is neither control, nor whitespace, nor git
crud, so those characters survived every one of the checks above. A
display name of nothing but U+200B ZERO WIDTH SPACE therefore satisfied
"at least one non-crud character survives" and git accepted the commit
with a visually blank author:
```
# pre-fix, BUZZ_ACP_DISPLAY_NAME set to two U+200B
$ git log -1 --format='%an' | xxd -p
e2808be2808b0a
```
Embedded marks were the other half: a trailing U+202E RIGHT-TO-LEFT
OVERRIDE reorders everything after it, so a stored author line renders
as something other than what it stores — the same confusion class the
angle-bracket filtering exists to prevent.
`is_unicode_format` rejects the whole `Cf` category rather than the
known-bad marks, because the boundary that matters is "invisible or
reorders text", not "the codepoint someone thought of". The 21 ranges
come from the UCD's `DerivedGeneralCategory.txt` (17.0.0), cross-checked
against Python's `unicodedata` (16.0.0); both yield exactly the same
set. They are inlined as a `matches!` rather than pulling in a
Unicode-tables crate for one predicate, and a test asserts both
endpoints of every range plus the codepoints immediately outside them —
including U+2065, which sits inside the U+2060 block but is unassigned
rather than `Cf`.
Filtering happens inside the existing per-word filter, so a format-only
name collapses to empty and falls out through the same `None` → npub
path as a crud-only name. No new fallback logic. And because filtering
precedes truncation, invisible padding cannot eat the 80-character
budget.
## NUL is handled one layer up
An interior NUL is a sibling constraint that cannot be fixed here: it
makes `Command::env` fail the entire spawn before this code runs, so it
has to die at the writer. #3028 establishes that pattern for the session
title in `resolve_session_title` via `filter(|c| !c.is_control())`, and
the Desktop follow-up that writes `BUZZ_ACP_DISPLAY_NAME` inherits it.
The shim sanitizer is a second line of defense for values that arrive
from somewhere other than Desktop.
## Verified end to end
Driving the real `buzz-dev-mcp` binary over stdio MCP and committing
inside its shimmed environment:
```
# BUZZ_ACP_DISPLAY_NAME="Duncan Idaho"
Duncan Idaho <dcfd242e...0f95@buzz.block.builderlab.xyz>
verify_exit=0
# BUZZ_ACP_DISPLAY_NAME unset
npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e...0f95@buzz.block.builderlab.xyz>
verify_exit=0
# BUZZ_ACP_DISPLAY_NAME=";;" (crud-only; would otherwise be fatal)
npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e...0f95@buzz.block.builderlab.xyz>
verify_exit=0
# BUZZ_ACP_DISPLAY_NAME=U+200B U+200B (format-only; would otherwise be blank)
npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e...0f95@buzz.block.builderlab.xyz>
verify_exit=0
# BUZZ_ACP_DISPLAY_NAME="Duncan" + U+202E (bidi override stripped)
Duncan <dcfd242e...0f95@buzz.block.builderlab.xyz>
verify_exit=0
# BUZZ_ACP_DISPLAY_NAME="Dun" + U+200B + "can" (zero-width removed, word not split)
Duncan <dcfd242e...0f95@buzz.block.builderlab.xyz>
verify_exit=0
```
Signature verification passes in every case — the signing identity is
unchanged.
`Related: #3028` — it establishes the Desktop-side env plumbing this
builds beside; the one-line Desktop follow-up that writes
`BUZZ_ACP_DISPLAY_NAME` alongside the session title ships after it
merges. Not a dependency: with the variable absent, `std::env::var`
returns `Err` and the npub fallback keeps current behavior exactly.
---------
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Signed-off-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
Co-authored-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
## Problem
Auth-class errors (expired OAuth token, HTTP 401) are non-retryable: the
token won't self-repair between attempts. Today, `PromptOutcome::Error`
for an application-class error falls into the generic `queue.requeue()`
path, burning up to 10 retry slots over a long backoff window before
dead-lettering. Will's canary run observed the 401 message being retried
repeatedly.
## Solution
Add `is_auth_error()` that classifies `AcpError::AgentError` messages
matching two narrow patterns observed in the field:
- `"Re-authenticate"` — emitted by the Claude CLI for expired OAuth
tokens
- `"API Error: 401"` — present in Claude/Codex HTTP-401 responses
Conservative matching is intentional: a false positive (misclassifying a
transient error as non-retryable) silently drops a user message, which
is worse than a false negative (extra retries on an auth error).
In `handle_prompt_result`, a new branch intercepts the failing batch
before `queue.requeue()` for auth-class errors and dead-letters
immediately, posting a user-visible notice to re-authenticate the CLI
(e.g. `claude /login` or `codex login`).
The transport/application split in `PromptOutcome::Error` is untouched —
this only changes batch fate after an application-class auth error.
## Tests
6 new tests in `error_outcome_emission_tests`:
1. `is_auth_error` matches `Re-authenticate` message
2. `is_auth_error` matches `API Error: 401` message
3. `is_auth_error` rejects other `AgentError` messages (usage credits,
etc.)
4. `is_auth_error` rejects transport errors (I/O, WriteTimeout)
5. Auth error dead-letters immediately — 0 pending channels after
`handle_prompt_result`
6. Non-auth application error still requeued — 1 pending channel after
`handle_prompt_result`
Full `cargo test -p buzz-acp`: 598/598 passing.
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Co-authored-by: npub1mn7jgtj4w2pd0g0zeuhxsa6jy6p0rewxz4kujt98my82ahfmp72sxjexk7 <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
