ci(cursor-review-self): pin workflows#125 and move to the new trigger set - #127
Conversation
… set The trigger change was deliberately held back from #125 so it would not ship `labeled` against the old callable, which had no label filter and would have launched a full review on every label add. This is the caller-wave commit that pin says to make. A push no longer re-reviews; a lane adds `review-now`. The concurrency group mirrors the callable's job `if:`. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Cursor review of |
There was a problem hiding this comment.
Cursor review · grok-4.6 (effort xhigh, fast false) · class P0 · agent bc-d9572030-291d-4b77-aa07-86e036a922fe · 5m 57s · head 389685341c6b
This retargets the repo's own cursor-review caller onto the workflows#125 contract: drop synchronize, listen for labeled, pin 246013d, and put the documented review/other concurrency discriminator on the caller so a stray labeled event cannot cancel-then-skip a live waiter. The new trigger set, group expression, and job grant match the README caller shape and the callable's job if: (minus inputs.enabled, which this caller always sets true); the pin is origin/main's 40-character SHA, and the review-now/p0/p1/p2 and no-ai-review labels already exist here. actionlint 1.7.12, lint_callables, and the cursor-review unit tests are green. Nothing in the diff needs a change before merge.
Findings: 0 blocking, 0 consider, 0 nit (0 posted inline).
Checks the reviewer ran:
- /tmp/actionlint -no-color -oneline .github/workflows/*.yml (v1.7.12, checksum-verified linux_amd64) -> exit 0, no findings
- python3 scripts/lint_callables.py -> 12 file(s) checked, 0 finding(s)
- python3 scripts/test_cursor_review.py -> 64 tests, OK
How to respond: fix and push, or reply in the thread with disposition: <accept|reject|defer> - <reason> and resolve it. A push no longer re-reviews on its own — add the review-now label when the new head needs another review. A blocking finding requests changes; the merge gate honours GitHub's review decision.
What
Retargets this repository's
cursor-reviewcaller onto the reviewer contract that landed in #125 (246013d4284ba0c15040c333e169b90a66334d1f). Workflow file only; nothing else in this repo changes.synchronizeout,labeledin. A push no longer re-reviews. A lane that wants the new head reviewed runsgh pr edit <n> --add-label review-now; the callable launches and then clears the label, so the next re-request is a fresh event. Every other label addition skips.labeledevent (bot-inbox, a triage label) would cancel a live review and leave the PR unreviewed. The callable carries the same allow-list in its jobif:; both halves are needed.852e0f0c) to246013d4284ba0c15040c333e169b90a66334d1f, which is what actually bringsfast: false, the label filter, and the P0/P1/P2 class rule.On 2026-09-19 push-driven re-reviews put the estate at 165
cursor-reviewruns over 80 heads and exhausted the Cursor Models pool for seven and a half hours.The class rule
Logan chose it over a numeric cap, so nothing counts launches. P0 —
.github/workflows/**,.github/actions/**,docs/agents/**,AGENTS.md/CLAUDE.md, orreview-p0— always reviewed. P1 — ordinary code — reviewed once per open/reopen/ready, again onreview-now. P2 — an automation author, achangeset-release/*head, a metadata-only diff, orreview-p2— launches no agent; the orchestrating session self-reviews.review-nowoverrides every P2 signal, and P0 is decided first, so dependabot bumping a pinned action inside.github/workflows/— or anyone labelling a workflow changereview-p2— is still reviewed. Labelsreview-now,review-p0,review-p1andreview-p2were created in this repo for that purpose.Launch ledger:
gh run list --repo narduk-enterprises/workflows --workflow cursor-review.yml.Authorization
Merge default: company-hq D-ORG-1 (g), 2026-09-02.
Logan's verbatim answers, 2026-09-19:
Validation
actionlinton the changed workflow, plus this repo's own required checks on the PR head.🤖 Generated with Claude Code