[Fleet Monitor] petry-projects/TalkTerm — .github/workflows/pr-review.yml

[don-petry]

🟡 WARNING: petry-projects/TalkTerm / .github/workflows/pr-review.yml

Metric	Value
Failure Rate	16.7% (1 / 6 runs)
Status	WARNING
p50 Duration	534s
p95 Duration	2048s
Successful Runs	4
Cancelled Runs	1

Workflow: .github/workflows/pr-review.yml

Last updated by Fleet Monitor run on 2026-07-18.
Threshold: failure rate > 10% over the monitored window.

[don-petry]

Dev-Lead Implementation Plan

Issue: #374 — [Fleet Monitor] petry-projects/TalkTerm — .github/workflows/pr-review.yml

Scope

Add a per-head-SHA concurrency block to .github/workflows/pr-review.yml so redundant triggers for the same commit no longer launch parallel duplicate review runs — the root cause of the inflated p95 duration (2048s) and the cancelled/failed runs behind the 16.7% failure rate.

Root-Cause Analysis

pr-review.yml fires on check_suite:completed, pull_request:synchronize, and pull_request_review — all of which fire for the same head commit. It is the only event-driven caller in this repo with no concurrency: block (ci-failure-analyst.yml, add-to-project.yml, auto-rebase.yml, dependabot-rebase.yml, initiative-driver.yml, apply-repo-settings.yml, feature-ideation.yml all have one). Without it, those overlapping triggers run multiple long reviews of identical code in parallel, wasting minutes and producing cancellations. The fix mirrors ci-failure-analyst.yml's convention: group by head SHA, cancel-in-progress: false (never interrupt an in-flight review; the reusable's per-SHA idempotency no-ops the queued duplicate).

Implementation Checklist

• .github/workflows/pr-review.yml — add top-level concurrency: block grouped by head SHA with cross-event fallbacks, cancel-in-progress: false

Tests to Write

• scripts/test-pr-review-workflow.sh — regression guard (follows existing scripts/test-copilot-setup-steps-workflow.sh pattern) asserting: file exists, a concurrency: block is present, the group key references head.sha/head_sha, and cancel-in-progress: false is set. Confirm red before implementing.

Test Command

bash scripts/test-pr-review-workflow.sh (plus bash .dev-lead/scripts/dev-lead-lint.sh for shellcheck)

[don-petry]

Dev-Lead: Implementation Complete

Plan Execution

• Root-caused the WARNING: pr-review.yml was the only event-driven caller with no concurrency: block, so overlapping triggers (check_suite:completed + pull_request:synchronize + pull_request_review) launched parallel duplicate reviews of the same commit — inflating p95 duration and producing cancelled runs.
• Wrote failing regression guard scripts/test-pr-review-workflow.sh (red).
• Added a per-head-SHA concurrency block with cancel-in-progress: false (green).
• Verified: new test green, YAML parses, shellcheck clean, dev-lead-lint passes, sibling workflow tests unaffected.

Test Results

=== test-pr-review-workflow ===
PASS: .github/workflows/pr-review.yml exists
PASS: 'concurrency:' block present in .github/workflows/pr-review.yml
PASS: concurrency group keyed on head SHA in .github/workflows/pr-review.yml
PASS: 'cancel-in-progress: false' present in .github/workflows/pr-review.yml

All checks passed.

Repo lint (.dev-lead/scripts/dev-lead-lint.sh): all checks passed. shellcheck --severity=warning on the new test: clean. All scripts/test-*.sh + *.test.sh pass.

Files Changed

• .github/workflows/pr-review.yml: added a top-level concurrency block grouping runs by PR head SHA (github.event.pull_request.head.sha || github.event.check_suite.head_sha || github.event.inputs.pr_url || github.run_id) with cancel-in-progress: false. Redundant same-commit events now share a lane and queue instead of running duplicate reviews in parallel; a new commit still starts its own run immediately. The pinned uses: channel ref and permissions: block (both off-limits per the caller header) are untouched.
• scripts/test-pr-review-workflow.sh: new regression guard (follows the existing scripts/test-copilot-setup-steps-workflow.sh pattern) asserting the concurrency block exists, is keyed on head SHA, and uses cancel-in-progress: false.

Notes

• cancel-in-progress: false (not true) is deliberate and matches ci-failure-analyst.yml: an in-flight review is never interrupted; the reusable's per-head-SHA idempotency no-ops the queued duplicate.
• Edge cases across all five trigger types are covered by the fallback chain: workflow_dispatch groups by pr_url; repository_dispatch (mention) falls back to the unique run_id so mention-triggered reviews are never blocked; a synchronize on a new commit gets a distinct group and runs immediately.
• No change to the reusable's own logic was needed — this is a caller-side serialization fix only.