[aw-failures] dismiss_pull_request_review 404 on stale review_id hard-fails safe_outputs job #49733

Add dismiss_pull_request_review to the safe-outputs report-only failure set — right now a stale review ID hard-fails the whole safe_outputs job. Why: ##[error]2 safe output(s) failed on run 30731801766 came from two dismiss_pull_request_review messages hitting 404 Not Found on pulls.getReview, and that failure propagates as a fatal job error instead of a no-op.

Problem statement

PR Sous Chef (schedule-triggered) failed at job safe_outputs, step "Process Safe Outputs", on run 30731801766 (2026-08-02T04:06Z). Two safe-output messages of type dismiss_pull_request_review (messages 5 and 6) both errored:

##[error]✗ Message 5 (dismiss_pull_request_review) failed: Not Found - https://docs.github.com/rest/pulls/reviews#get-a-review-for-a-pull-request
##[error]✗ Message 6 (dismiss_pull_request_review) failed: Not Found - https://docs.github.com/rest/pulls/reviews#get-a-review-for-a-pull-request
##[error]2 safe output(s) failed:

audit-diff against the nearest successful cohort run (30729572519) shows no firewall anomalies — the only material delta is GitHub core quota consumption (+314%, 58 vs 14 calls) from the extra review lookups, confirming this is an application-level 404, not an infra/auth issue.

Affected workflows and run IDs

• PR Sous Chef (.github/workflows/pr-sous-chef.lock.yml) — run 30731801766
• Handler: actions/setup/js/dismiss_pull_request_review.cjs

Probable root cause

actions/setup/js/dismiss_pull_request_review.cjs:228 calls githubClient.rest.pulls.getReview({ ... review_id: reviewId }) for the explicit (non-auto) review_id path. When that review no longer exists — already dismissed, deleted, or a stale ID captured earlier in the same agent turn — the call 404s and is returned as {success: false, error: ...}.

That failure is then treated as fatal because dismiss_pull_request_review is absent from REPORT_ONLY_FAILURE_TYPES in actions/setup/js/safe_output_handler_manager.cjs:671 (currently only assign_to_agent and upload_artifact). This is the same class of problem PR #49648 just fixed for stale/resolved review threads (resolve_pull_request_review_thread) — but the fix wasn't extended to stale/already-dismissed reviews referenced by explicit ID in dismiss_pull_request_review.

Proposed remediation

1. In dismiss_pull_request_review.cjs, catch the getReview 404 specifically and return a soft no-op ({success: true, skipped: true, reason: "review no longer exists"} or similar) instead of a hard error — mirroring the stale-thread no-op pattern added in fix(safe-outputs): treat stale/resolved review threads as no-ops; allow bot review dismissal #49648.
2. Alternatively/additionally, add dismiss_pull_request_review to REPORT_ONLY_FAILURE_TYPES in safe_outputs_handler_manager.cjs:671 so a single stale-review 404 can't fail the whole safe_outputs job for unrelated safe outputs in the same run.
3. Add a regression test in actions/setup/js/dismiss_pull_request_review.test.cjs covering a 404 from pulls.getReview on the explicit review_id path.

Success criteria / verification

• A dismiss_pull_request_review message targeting a review_id that 404s no longer fails the safe_outputs job; it is logged as a skipped/no-op result.
• dismiss_pull_request_review.test.cjs includes a case for getReview returning 404 and asserts a non-fatal result.
• Next PR Sous Chef scheduled run completes with Process Safe Outputs green even if a targeted review was already dismissed/deleted.

Parent: #49245 (rolling failure-investigator report). Analyzed run: 30731801766 (audit + audit-diff vs baseline 30729572519).
Related to #49245

Generated by 🔍 [aw] Failure Investigator (6h) · agent · 109 AIC · ⌖ 33.3 AIC · ⊞ 5.3K · ◷

• expires on Aug 8, 2026, 11:47 PM UTC-08:00