[safe-output-health] 🏥 Safe Output Health Report - 2026-08-10

[github-actions[bot]]

Executive Summary

• Period: Last 24 hours (2026-08-09 → 2026-08-10)
• Runs analyzed: 202
• Workflows active: ~130 distinct workflows across 9 engines (copilot: 90, pi: 74, claude: 26, codex: 3, aider: 2, kiro: 2, crush/goose/opencode: 1 each)
• safe_outputs job invocations: 202 (184 success, 15 skipped — no items to process, 3 failures)
• Success rate (of attempted jobs): 98.4% (184/187)
• Error clusters identified: 1 (already resolved)
• Adjacent write jobs checked (push_repo_memory, push_evals_state, push_experiments_state, update_cache_memory, upload_assets, unlock): 0 failures

Scope note: this audit covers only safe_outputs job health. The 10 agent-job (driver_exit/agent_logic) failures seen in the same window are out of scope — those are handled by other monitors.

Safe Output Job Statistics

Job Type	Total Executions	Failures	Success Rate
safe_outputs (all types combined)	202	3	98.5%*

* 15 of the 202 runs had no safe-output items to process (job skipped); of the 187 runs that actually attempted to process items, 184 succeeded (98.4%).

Per-type breakdown isn't separately trackable from job-level status (all safe-output types for a run are processed in one safe_outputs job), but the 3 failures were all specifically caused by one output type: resolve_pull_request_review_thread.

Error Clusters

Cluster 1: Stale review-thread node ID in resolve_pull_request_review_thread

• Count: 3 run failures (17 individual failed items across those 3 runs: 1 + 7 + 9)
• Affected job: safe_outputs → step "Process Safe Outputs"
• Affected workflow: "PR Sous Chef" (all 3 occurrences, engine: pi)
• Sample error:

  ##[error]Failed to resolve review thread: Request failed due to following response errors:
   - Could not resolve to a node with the global id of 'PRRT_kwDOPc1QR87fJc0o'.
  ##[error]✗ Message 4 (resolve_pull_request_review_thread) failed: Request failed due to following response errors:
   - Could not resolve to a node with the global id of 'PRRT_kwDOPc1QR87fJc0o'.
  

• Root cause: resolve_pull_request_review_thread called the GraphQL resolveReviewThread mutation using a thread node ID that no longer resolved by the time the safe_outputs job ran (thread already resolved/deleted, or ID gone stale between agent run and safe-outputs processing). The processor treated this as fatal and failed the whole message/job instead of skipping just that one stale item.
• Impact: Each occurrence failed the run's safe_outputs job (conclusion: failure) even though other safe-output items in the same run (comments, issues) succeeded independently.
• Timing: All 3 occurrences happened on 2026-08-09 at 15:00, 15:56, and 22:21 UTC — i.e., runs §31319713324, §31322214642, and §31338952148.

This is already fixed. Commit 59c283e ("Skip stale review-thread node IDs in resolve_pull_request_review_thread instead of failing safe_outputs", PR #51630) merged 2026-08-10T00:26:15Z — after all 3 observed failures and before the rest of this audit window. Checked every "PR Sous Chef" safe_outputs run after that timestamp in this window: all succeeded, with zero recurrences of this error signature.

Root Cause Analysis

API-Related Issues

The one real failure cluster (above) was a GitHub GraphQL "could not resolve to a node" response for an already-resolved review thread — not a rate limit, auth, or permission problem. No other API-related safe_outputs failures were found (no rate-limit or auth errors in any safe_outputs job log this period).

Data Validation / Parsing Issues

None found in safe_outputs job logs this period.

Permission Issues

None found. All safe_outputs jobs that ran had the tokens/scopes they needed; the only related warnings were non-fatal update pull request branch from base warnings ("There are no new commits on the base branch" / workflow-scope timeout) which are expected, already logged as warnings (not errors), and don't fail the job.

Other Issues

None found in safe_outputs job logs. Zero failures in adjacent write jobs (push_evals_state, push_experiments_state, upload_assets, etc.).

Recommendations

Critical Issues (Immediate Action Required)

None. The one cluster found this period is already fixed on main.

Bug Fixes Required

None outstanding — the only bug found (stale review-thread node ID crashing the whole safe_outputs job) was already fixed in commit 59c283e / PR #51630 before this audit ran.

Configuration Changes

None needed.

Process Improvements

1. Non-fatal warnings still worth tracking over time
   • Current state: update pull request branch from base failures ("no new commits on base branch", workflow-scope timeouts) are correctly logged as warnings and don't fail the job.
   • Proposed: No code change needed now, but if these warnings start recurring frequently for the same PRs, it may indicate branch-protection or workflows scope gaps worth a follow-up look. Flagging for future-audit awareness only.

Work Item Plans

No new work items — the single defect found this period is already resolved. Recommend closing out tracking of the resolve-review-thread-stale-node-id pattern in the next audit (2026-08-11) once confirmed there are zero further occurrences.

Historical Context

Compared with the 2026-08-09 audit (cache memory: clean, 0 failures out of 201 executions, 100% success):

Trends

• Error rate trend: Increasing then resolving — went from 0 failures (2026-08-09) to 3 failures (2026-08-10), but all 3 predate a same-day fix that's now merged and verified working.
• Most common recurring issue: none carried over from the prior audit; this is a newly-detected-and-closed pattern.
• Improvement since last audit: the underlying defect that caused today's 3 failures no longer exists on main as of 59c283e.

Metrics and KPIs

• Overall safe_outputs success rate (of attempted): 98.4% (184/187)
• Most reliable job type: adjacent write jobs (push_evals_state, push_experiments_state, upload_assets, etc.) — 100% across all runs checked
• Most problematic job type: safe_outputs for the "PR Sous Chef" workflow specifically, driven entirely by the now-fixed resolve_pull_request_review_thread bug
• Average time to failure: N/A (all failures were terminal within the same safe_outputs job run, no retries observed for this mutation)

Next Steps

• Confirm in the 2026-08-11 audit that zero resolve_pull_request_review_thread stale-node-ID failures recur (verifying commit 59c283e fully resolves the issue in production)
• No other action required this cycle

References:

• §31322214642 — PR Sous Chef, resolve_pull_request_review_thread failure (1 item)
• §31338952148 — PR Sous Chef, resolve_pull_request_review_thread failure (9 items)
• §31319713324 — PR Sous Chef, resolve_pull_request_review_thread failure (7 items)

Generated by 🔒 Safe Output Health Monitor · agent · 269.9 AIC · ⌖ 34.1 AIC · ⊞ 9.7K · ◷

• expires on Aug 10, 2026, 9:02 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Safe Output Health Monitor.

A newer discussion is available at Discussion #51935.