[CamSoper]

Proposed changes

Fixes a race in claude-code-review.yml where a push plus a @claude #new-review lands a PR on review:error with no review.

How it happened (#21172 / #21210, 2026-10-08):

1. A push fires claude-triage.yml, which runs on synchronize, and it chains a workflow_run into claude-review.
2. The #new-review dispatch starts first and claims the claude-review-<PR> concurrency group (Add blog post: AI Agent Guardrails for Infrastructure #21172: 01:08:11).
3. Triage finishes and the chained run queues in the same group. When it gets a runner it cancels the in-flight dispatch through cancel-in-progress (01:09:37). That chained run was always going to skip: it's the workflow_run path on a PR that already has a review-state label.
4. The cancel lands in a pre-step (Layer B claim extraction), so the Opus step's outcome is skipped rather than cancelled. The classify step only treats cancelled as superseded, so superseded=false. The error trap then sets review:error.
5. The chained run's pr-context reads review:error and skips with "already-reviewed" (01:10:27). There's no review, and the label says error.

The other five PRs in that batch got lucky on ordering: their triage-chained run queued first and the dispatch cancelled it.

Fix (two layers):

• autofire-gate job. A cheap job that only runs on the workflow_run path. It reads draft status and the review-state labels, which are the same signals pr-context's workflow_run skip uses. When the run is certain to skip, claude-review moves into a per-run concurrency group (claude-review-<PR>-noop-<run_id>), so it can't cancel anything. claude-review still runs and publishes its skipped check-run as before. The gate fails open: a lookup error puts the run in the shared group, same as today. claude-review gets needs: autofire-gate with a !cancelled() wrapper on its if, so the gate being skipped on workflow_dispatch doesn't skip the review.
• Classify review outcome. If the Opus step outcome is skipped but job.status is cancelled, the run counts as superseded (or as timed out if past the job budget). The error trap then leaves the labels to the newer run. On its own this would just leave review:in-progress behind instead of review:error, which is why the first layer is the real fix.

Validation:

• make test-review-pipeline: passes.
• test_workflow_sanity.py: 137 passed.
• actionlint 1.7.7: clean, before and after.
• The jq filter was checked against label/draft fixtures.

Two things to watch after merge:

• job.status reading cancelled inside an always() step after a concurrency cancel is documented GitHub behavior, but I couldn't exercise it here.
• needs in a job-level concurrency expression is an allowed context, but this is its first use in the repo.

A cheap live check is to push and immediately comment #new-review on a reviewed PR, then confirm the triage-chained run logs autofire-gate: … noop=true and the dispatch survives.

Related issues (optional)

Seen on #21172 and #21210 after the v3-scripts merge batch. Same family as the guard comment in the error trap (a superseded run's review:error blocking its replacement).

🤖 Generated with Claude Code

https://claude.ai/code/session_015EbMbem5iFw1cErRVNJXzH

---

Generated by Claude Code

[github-actions]

Sentinel — merge gate status (preview)

Warning

⚠️ Preview mode — this check is NOT blocking your merge

The Sentinel is in preview (report-only) mode: its check-run concludes neutral whatever the gates below say, so nothing here gates this PR. The rows are informational and safe to ignore for now. It would have concluded failure.

This will be enforced in the near future. Once it is, every row below has to be green before this PR can merge. The rows are the sign-offs a merge needs — some yours, some a reviewer's or a deploy's. Tell us in #docs if one looks wrong.

1 of 5 gates would block this merge once the Sentinel is enforced.

	Gate	What it needs
✅	G1 review-ran	Nothing to do: review current at 1df0012c1.
✅	G2 findings-answered	Nothing to do: every finding answered.
🔴	G3 right-approver	Needs an approval from anyone with write access to this repo — no qualifying human approval at this head (bot approvals count only on the automated lanes bot_approvers lists).
➖	G4 infra-evidence	Doesn't apply to this PR: no changed path affects the deploy.
➖	G5 oversized-ack	Doesn't apply to this PR: not oversized.

Break glass: a member of a routing team (.github/review-routing.yml teams:) can apply review:waived, which is logged.

_{Evaluated at head 1df0012c1.}

[pulumi-bot]

Your site preview for commit 1df0012 is ready! 🎉

http://www-testing-pulumi-docs-origin-pr-22182-1df0012c.s3-website.us-west-2.amazonaws.com

[github-actions]

Reviewer's guide v1 — not for the author

Tip

Check the ⚠️ items, then approve. Your approval covers only those; links, shortcodes, metadata, and verified claims were checked by machine (receipts on the evidence page). Code samples are read, not compiled.

PR author: nothing here is yours. Your to-do list is the "Author action guide" comment.

Review requested from: @pulumi/docs-tools — an approval from anyone with write access satisfies the merge gate.

Note

What this PR changes:

• New autofire-gate job: on workflow_run, reads the PR's draft state and review-state labels and outputs noop.
• claude-review now needs: autofire-gate, adds a !cancelled() guard, and moves a noop run into its own per-run concurrency group.
• Outcome step: a cancel that lands in a pre-step (Opus step skipped, job.status cancelled) now counts as superseded or timed out instead of review:error.

The changes would be wrong if the gate drifted from pr-context's skip guard, or if the needs wiring skipped claude-review on workflow_dispatch. The review compared the gate's label list with the already-reviewed guard (they match, and the gate fails open) and traced the !cancelled() condition through the dispatch path.

Review confidence:

Dimension	Level	Notes
mechanics	HIGH
facts	HIGH

⚠️ Check these before approving

Nothing needs a human eye beyond the rubber-stamp list below.

✅ What you can rubber-stamp

• Facts: no factual claims found in the changed lines.
• Mechanics: no internal links added.
• Style: 0 advisory suggestions left with the author; never blocking.

Pre-existing issues in touched files: 0 — details on the evidence page.

Full evidence: verification trail, investigation log, review history.

_{Review v1 · updated 2026-10-08T14:59:23Z · head commit 1df0012}

---

For the reviewer: something look wrong? Comment @claude <your point> #update-review and the review re-checks with your input.

[github-actions]

Author action guide v1 — nothing blocks merge

Note

Nothing here needs an answer from you. A human reviewer still approves the merge.

This PR adds an autofire-gate job so an auto-fire run that is certain to skip can't cancel an in-flight review, and treats a pre-step cancel as superseded rather than an error; the review checked the gate's skip signals against pr-context's already-reviewed guard and the needs/if wiring on workflow_dispatch.

Full evidence: verification trail, investigation log, review history.

_{Review v1 · updated 2026-10-08T14:59:23Z · head commit 1df0012}

---

Please don't edit or delete this comment; it's the review's record.