[aw-failures] [aw-fix] P1: PR Sous Chef — resolve_pull_request_review_thread safe output fails, failing Process Safe Outputs

[github-actions]

Fix the resolve_pull_request_review_thread safe-output handler in PR Sous Chef — it fails the whole Process Safe Outputs job on every scheduled run.

PR Sous Chef's agent succeeds, but the safe_outputs job then aborts because its review-thread resolution requests fail. This recurred twice in the 6h window (~50 min apart), so it is a persistent pattern, not a transient blip.

Problem statement

The Process Safe Outputs step fails with:

##[error]Failed to resolve review thread: Request failed due to following response errors:
##[error]✗ Message 5 (resolve_pull_request_review_thread) failed: Request failed due to following response errors:
##[error]✗ Message 6 (resolve_pull_request_review_thread) failed: Request failed due to following response errors:
##[error]2 safe output(s) failed:

The agent job concludes success (12 turns, 54.7k tokens); the failure is isolated to safe-output processing. audit-diff vs the nearest successful baseline shows no firewall or tooling regression — the baseline "success" was a 0-turn no-op, while these runs did real work and failed only at the GraphQL resolve step.

Affected workflows and run IDs

Workflow	Run	Result
PR Sous Chef	30022071713	representative — Msgs 5 & 6 failed
PR Sous Chef	30018005233	second occurrence, same signature
PR Sous Chef	30014223483	baseline success (no-op, 0 turns)

Probable root cause

The safe-output handler issues resolveReviewThread GraphQL mutations against thread IDs that the API rejects — most likely already-resolved threads, stale/invalid thread IDs, or threads on a PR whose state no longer accepts resolution. The handler treats any non-2xx GraphQL response as a hard error and fails the job instead of tolerating an already-resolved / non-resolvable thread.

Proposed remediation

• Make resolve_pull_request_review_thread idempotent: treat "thread already resolved" / "thread not found" GraphQL responses as success (no-op), not job failure.
• Log the raw GraphQL error body per failed message so the actual API error (currently truncated as "following response errors") is captured for diagnosis.
• Consider not failing the aggregate Process Safe Outputs job when only optional review-thread resolutions fail — downgrade to a warning while still surfacing counts.

Success criteria / verification

• PR Sous Chef scheduled runs complete with safe_outputs job success when the only issue is an already-resolved/stale review thread.
• The full GraphQL error body is present in logs for any genuinely failed resolution.
• No recurrence of the 2 safe output(s) failed signature across a subsequent 24h window.

Parent: #47525. Analyzed runs: 30022071713, 30018005233, 30014223483.
Related to #47525

Generated by 🔍 [aw] Failure Investigator (6h) · age00 · 131.7 AIC · ⌖ 33.1 AIC · ⊞ 5.2K · ◷

• expires on Jul 30, 2026, 11:20 AM UTC-08:00