Repository navigation
Description
Convergence latency (deterministic)
Workflow duration across 184 completed run(s) in the window — computed by pr_review_health.sh, not the model.
| Workflow | Runs | min | p50 | p95 | max |
|---|---|---|---|---|---|
pr-review-trigger.yml |
184 | 1s | 36s | 5m45s | 19m36s |
Run-outcome mix by triggering event
pr-review run conclusions over the window, split by the event that triggered each run — computed deterministically by pr-review-outcomes.sh, not the model. A high cancelled share means runs were evicted while PENDING (concurrency admits one running + one pending per group); a cancelled run is non-blocking but historically rendered as a failing check (#1421).
| Event | success | cancelled | skipped | failure | total | cancel % |
|---|---|---|---|---|---|---|
check_suite |
16 | 2 | 7 | 0 | 25 | 8% |
pull_request |
7 | 1 | 0 | 0 | 8 | 12% |
pull_request_review |
5 | 4 | 6 | 0 | 15 | 26% |
repository_dispatch |
70 | 33 | 0 | 6 | 109 | 30% |
workflow_dispatch |
26 | 1 | 0 | 0 | 27 | 3% |
| TOTAL | 124 | 41 | 13 | 6 | 184 | 22% |
Sweep hit-rate (deterministic)
How often the pr-review-sweep scheduled backstop actually had to re-review a PR — computed deterministically by pr-review-sweep-metrics.sh, not the model. The scheduled (cron) path re-dispatches only for genuinely un-eventable cases (GitHub-App / cross-repo checks like SonarCloud that emit no workflow_run); the workflow_run fast path handles every eventable case, so a scheduled re-dispatch is always a case the fast path did not cover.
| Metric | Value |
|---|---|
| Scheduled sweep ticks | 8 |
| Ticks that re-dispatched a review (hits) | 8 |
| Sweep hit-rate | 100% |
Interpretation: a low rate confirms the timer stays exception-only (the event fast path is carrying the load); a rising rate is a signal that an eventable case is leaking onto the timer — a hole in the narrowing worth investigating.
PR Review Agent — Health Report
Workflow: pr-review-trigger.yml | Repo: petry-projects/.github-private
Generated: 2026-09-04
1. Executive Summary
Status: WARNING (self-resolved)
Period: 2026-09-03T11:42Z – 2026-09-04T11:10Z
Result: 6 of 184 runs failed (3.3%) — all failures within a 47-minute window on 2026-09-03
Key findings:
- All 6 failures clustered between 11:42Z and 12:29Z on 2026-09-03; zero failures before or after that window
- No failed run logs were captured in the report payload — root cause cannot be confirmed from log content alone
- 41 cancellations are consistent with expected concurrency-group preemption (burst of events triggering rapid successive runs), not errors
Action required: Retrieve and inspect logs for runs #71889, #71905, #71909, #71919, #71923, #71925 to confirm root cause before the failure window recurs.
2. Failure Breakdown
Note: The "Failed Run Logs" section in the source data was empty. The table below is derived from timing patterns and workflow structure; error messages are unavailable without the actual log content.
| Failure Category | Affected Runs | Example Error Message |
|---|---|---|
| Transient infrastructure / runner error | 6 (estimated) | (logs not captured — see note above) |
| Engine rate limit (Claude quota) | Possible secondary | (unconfirmed without logs) |
| GitHub API rate limit | Possible secondary | (unconfirmed without logs) |
| Permission / auth error | Unlikely (runs resumed immediately after) | — |
The 47-minute blast radius and full self-recovery strongly suggest a transient upstream issue (runner infrastructure, reusable-workflow availability, or a brief quota burst) rather than a config regression.
3. Error Patterns
No log content was provided for the 6 failed runs. The following is inferred from timing alone:
Pattern — Tight failure burst, immediate recovery
- Runs #71889 → #71925 all failed within a 47-minute window.
- The run immediately before (#71888, cancelled) and immediately after (#71926, success) show no error signal.
- This rules out a permanent config change as the cause (those would persist past 12:30Z).
- Consistent with: transient runner quota exhaustion, a brief outage in the
pr-review/v1-nextreusable's host ref, or a GitHub API 503 burst.
Pattern — Concurrency cancellations are normal
- Cancelled runs (#71881, #71877, etc.) appear paired with successful runs at identical or near-identical timestamps.
- This is the expected behavior of GitHub Actions concurrency groups: a newer run preempts an older in-flight run. Not an error condition.
Pattern — Skipped runs
- 7 skipped runs (#71934, #71970–#71975, #71949, #71954, #71955, #71983, #71996, #72001) indicate the workflow's own guard conditions fired (e.g.,
if:expressions evaluating false for bots, draft PRs, or non-qualifying event types). Normal.
4. Token Scope Analysis
Scopes granted by pr-review-trigger.yml:
| Scope | Level | Purpose |
|---|---|---|
contents: read |
read | Checkout, reading workflow files |
pull-requests: write |
write | Submitting reviews and comments |
checks: read |
read | Reading check suite status for the check_suite trigger |
Scopes potentially missing or unverifiable (no gh auth status output in logs):
| Scope | Status | Recommendation |
|---|---|---|
statuses: read |
Unverified | If the reusable reads commit statuses (not checks), add statuses: read to the calling job |
actions: read |
Unverified | Required if the reusable inspects other workflow runs; add if 403s appear on Actions API calls |
metadata: read |
Implicit | Always granted; no action needed |
No auth errors are evidenced from the run pattern (failures self-resolved and auth misconfigurations do not self-heal), so the current scope set is likely sufficient.
5. Recommendations
-
Retrieve logs for the 6 failed runs [CRITICAL priority for diagnosis]
- What:
gh run view 71889 --log-failed, repeated for 71905, 71909, 71919, 71923, 71925 - Why: The empty log payload means root cause is currently unknown; if the condition recurs, there is no baseline to diff against
- Expected impact: Confirms or rules out rate limiting, reusable-ref outage, or runner issue
- Urgency: HIGH (failures resolved, but next occurrence is unactionable without this)
- What:
-
Add failure alerting for burst patterns [MEDIUM]
- What: In
.github/workflows/pr-review-trigger.ymlor a companion workflow, add a step that fires a GitHub issue or dispatch event when ≥3 failures occur within 30 minutes on the same workflow - Why: The 47-minute failure window on 2026-09-03 resolved before it could be caught manually; earlier detection would allow faster triage
- Expected impact: Reduces mean-time-to-detection for transient burst failures
- Urgency: MEDIUM
- What: In
-
Verify
pr-review/v1-nextref was stable during 11:42–12:29Z on 2026-09-03 [MEDIUM]- What: Check
scripts/cut-release.shhistory and tag movement onpr-review/v1-nextaround that timestamp (git log --tags --simplify-by-decoration --pretty="format:%d %ai" | grep pr-review/v1-next) - Why: If a bad release was cut to
v1-nextand then rolled back, the failure window maps directly to the bad-ref window — a known risk for ring-0 canaries - Expected impact: Either closes the hypothesis or surfaces the need for a canary health gate before
nexttag moves - Urgency: MEDIUM
- What: Check
-
Document concurrency group configuration [LOW]
- What: Confirm the reusable
pr-review.yml@pr-review/v1-nextdefines aconcurrency:group scoped topr_urlor PR number; if not, add one in this stub - Why: The high cancellation count (41) is expected for event bursts, but without a scoped concurrency key, unrelated PRs can cancel each other's reviews
- Expected impact: Reduces unnecessary cancellations and improves throughput during high-activity windows
- Urgency: LOW
- What: Confirm the reusable
6. Health Score
Health: 8/10 — Workflow is operating normally with a single unexplained 47-minute failure burst on 2026-09-03 that self-resolved; root cause remains unconfirmed due to missing log capture.
Activity