PR Review Agent — failures detected 2026-07-23 #1361

1. Executive Summary

Status: HEALTHY
Period: 2026-07-22T23:44:20Z – 2026-07-23T07:40:10Z (~8 hours)
Result: 1 of 100 runs failed (1.0%) — 57 succeeded, 19 cancelled, 23 skipped

Key findings:

• Single failure (#58056) has no recoverable log output; root cause is unconfirmed
• 19 cancellations are concurrency-driven auto-cancels from rapid push bursts (03:06–03:51 UTC cluster), not errors
• 23 skips are expected conditional exits (trigger fired but job if: conditions not met — normal for check_suite / pull_request_review events on non-PR commits)

Action required: Retrieve and inspect logs for run #58056 to rule out a transient vs. recurring fault before closing.

2. Failure Breakdown

3. Error Patterns

Run #58056 — failure @ 2026-07-23T00:57:42Z

No log content was included in the report for this run. The run is immediately adjacent to a cancelled run (#58057, same second) and a skipped run (#58058, 1 s later), which suggests it fired during the same push burst at ~00:57 UTC. Possible causes in order of likelihood:

1. Transient GitHub infrastructure error — the run started and was interrupted before the called workflow (pr-review.yml@pr-review/next) could initialise; no repeating pattern in surrounding runs supports a code-level fault.
2. Reusable workflow ref not found — if pr-review/next tag was absent or mid-move at that instant, the uses: line would fail immediately with Could not find uses reference.
3. Secret resolution failure — secrets: inherit propagation occasionally fails transiently if the secret backend is temporarily unavailable.

To confirm: run gh run view 58056 --log against petry-projects/.github-private.

Cancellation cluster (03:06–03:51 UTC)

18 of 19 cancellations occur in a tight burst. This is the expected behaviour of GitHub's built-in concurrency cancellation when many pull_request: synchronize or check_suite: completed events fire in rapid succession (e.g., a force-push or a stack of dependent PRs landing). No action needed unless the cancellations are causing reviews to be silently skipped on the final commit — the last run in each burst shows success, confirming the terminal run does complete.

4. Token Scope Analysis

Scopes granted in workflow source (permissions: block):

Scopes absent / not assessed (no gh auth status output in logs):

No scope-related errors are present in available logs. The permissions block mirrors the documented requirements in pr-review.yml, so no changes are indicated at this time.

5. Recommendations

1. Retrieve and inspect failed-run logs for #58056

   • Command: gh run view 58056 --log --repo petry-projects/.github-private
   • Why: The single failure is unclassified; without log content it cannot be confirmed as transient. If the error is a reusable-ref resolution failure it would indicate a gap in the pr-review/next tag-move procedure.
   • Expected impact: Identifies whether a fix is needed or the failure can be closed as a one-off.
   • Urgency: MEDIUM — failure rate is 1%; no evidence of recurrence.

2. Verify pr-review/next tag integrity after every cut-release.sh run

   • File: scripts/cut-release.sh
   • Why: The failure occurred at the same timestamp as a cancel/skip burst, which can coincide with a tag move. A lightweight post-move check (gh api /repos/petry-projects/.github-private/git/ref/tags/pr-review/next) would catch a missing or mis-pointed tag before the next real PR triggers it.
   • Expected impact: Eliminates ref-not-found failures during release cuts.
   • Urgency: LOW — no confirmed occurrence, purely preventive.

3. Consider adding a concurrency: group to pr-review-trigger.yml with cancel-in-progress: true (if not already in pr-review.yml)

   • Why: The cancellation burst confirms rapid-fire triggering. Making the cancellation policy explicit in the stub (scoped to pr-${{ github.event.pull_request.number }}) prevents silent coverage gaps where neither the cancelled nor the successor run reviews a given commit.
   • Expected impact: Reduces noise cancellations and guarantees exactly one review per head SHA.
   • Urgency: LOW — current behaviour appears to converge correctly (last run in each burst succeeds).

6. Health Score

Health: 9/10 — one undiagnosed isolated failure in an otherwise clean 8-hour window; no recurring errors, cancellations and skips are structurally expected.