Repository navigation
ci: let pr readiness accept pre-rename ai review lane names - #16377
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS After merge, re-dispatch PR Readiness for #15593, #15029 and #14969 using their full head SHAs. The sweep will not re-fire them on its own. [DESIGN-REVIEWED] 1cb07f7 Description read[DESCRIPTION-READ] 21a2c8544b8914b4ae8b345841a4f85cf51f1aad739d396b40b5fb00dee64a90 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — 🟡 CONCERNSPremise-level review of First-Principles-Verdict: CONCERNS The Not justified as shipped
What this change shipsInventory (5 items) — 4 justifiedIntent: let fork PRs reviewed before the lane rename reach their real readiness verdict without a re-push. FIX.
[FIRST-PRINCIPLES-REVIEWED] 1cb07f7 Description read[DESCRIPTION-READ] 18a521c3a7b6b915016e0cc6a8ddcd441133b1cfa847402f2c3bdafd49ba2dd3 That sha256 names the PR description this verdict read. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
#16238 renamed the Opus and GPT review lanes. The publish step's fork re-check binds a Stage-2 row by name as well as external_id, so a fork head reviewed before the rename scored green and then re-checked as "Opus 5.5 Review (not started), GPT 6.1 Review (not started)" forever, with only a push able to clear it. Add one LEGACY_LANE_NAMES table (old name -> new name, renamed lanes only). An old-name row answers for its lane only when the head has no bound row under the current name, and it is read exactly like one: red stays red, running holds, and the external_id binding still applies. The hold step and MONITORED_LANES key on run ids and workflow files, so they need no alias. Removal is tracked in #16373.
0aef1a0 to
1cb07f7
Compare
iamwhatever
left a comment
There was a problem hiding this comment.
Reviewed (read-only check of diff, tests and CI). Security looks fine. The permissions and the pull_request_target trigger were already there, and this PR changes neither. No PR code is checked out, and the alias table is fixed in the workflow. An old-name row only counts if it is bound to this PR, this head SHA and the current Fast Gate attempt. Red and running rows still block. So a PR cannot pass readiness without a real AI review of that SHA. Approving. First Principles suggests dropping the name filter in place of the alias table. That is worth a follow-up, but it does not block this PR. After merge, re-dispatch PR Readiness for #15593, #15029 and #14969 using their full 40-character head SHAs.
I only read: gh pr view, gh pr diff, both workflow files at their head SHAs, the bot review comments, and the check lists. I posted or changed nothing. On both PRs, every check that ran passed and the Opus and GPT reviews report no blocking findings. The @kirocrew-computer MCP server was not available in this session; this review didn't need it.
Problem / Motivation
Goal: a fork PR whose AI review lanes ran before #16238 renamed them reaches its real readiness verdict without a re-push.
#16238 renamed
Opus 5 ReviewtoOpus 5.5 ReviewandGPT 5.6 ReviewtoGPT 6.1 Review.pr-readiness.ymlruns from the base branch, so every fork head whose Stage-2 lanes posted before the rename now sits atpendingwith "readiness check(s) changed since evaluation; re-checking Opus 5.5 Review (not started), GPT 6.1 Review (not started)", even with every check green. Example: #15593 at634caf55af. Its rows areOpus 5 Review/GPT 5.6 Review, success,external_idopus-pr-15593-37091252134-2/gpt-pr-15593-37091252134-2. Nothing carries the new names. Open fork PRs stuck this way right now: #15593, #15029, #14969. Others hit the same wall the next time their verdict recomputes to success.Why it matters
PR Readinessis the only required status, so these PRs cannot merge. The only way out today is a force-push to re-run every lane, which is costly while CI is saturated.Not a goal
/ai-review overridere-run lookup inai-review-human-override.yml, which also finds the fork run by its current name (see Residuals).What changed (motivation → approach → change)
pendingby the publish step's pre-success re-check.external_idalone, so it accepts the old-name row. The publish step's fork re-check binds by name andexternal_id, so it finds noOpus 5.5 Reviewrow and reports "(not started)" on every recompute.LEGACY_LANE_NAMESjob-env table, mapping old name to new name for the two renamed lanes only.external_idbinding still applies, so a stale attempt still reads "(not started)".jq --argjsonfail, which publishes the[read-failed]pending.MONITORED_LANESon workflow files, so neither needs the alias.Alternative considered and rejected: drop the name filter, so the re-check binds by
external_idalone like the verdict step (or picks the newest bound row whatever its name). The cleanup review rated this the more general root fix, and that is a fair reading: the verdict step already trustsexternal_idalone, and future renames would cost nothing. I rejected it for three reasons:external_id. That is effectively unreachable. A current-name row on that attempt means the lane already ran post-rename, and the override re-run locates the lane by its current name, so it re-runs the post-rename run.The table is temporary and has a named removal issue, so the rejected generalisation can still be proposed on its own.
Backwards compatibility
Compatible: the re-check output for every lane without an alias is byte-identical. The reviewer diffed old and new jq over 400 random check lists. The only new outcome is an old-name bound row answering where "(not started)" was published before.
Removes nothing: the change only adds an accepted row shape to one read. No key, path, record or validator is taken away.
Tests
test/test_pr_readiness_publish.py, which runs the publish step's real shell against aghstub:success;failureunder the current lane names;pending, and not as "(not started)";main's workflow, withpendingpublished instead of the verdict.test/test_ai_review_workflows.py: the table holds exactly the two ci(review): move every AI review lane to Opus 5.5 and GPT 6.1 Sol #16238 renames. Each new name is a live fork lane, no old name is a live lane, and no workflow other thanpr-readiness.ymlstill posts an old name.test_pr_readiness_publish.py,test_pr_readiness_evaluate.py,test_pr_readiness_sweep.py,test_pr_readiness_dispositions.py,test_ci_surface_tests.py,test_ai_review_workflows.py: 1489 passed. Also passing: flake8, isort, the black and subprocess-encoding gates,scripts/docs-lint.sh.scripts/local-gate.pybackend lane: 2020 passed. Its frontend lane could not run because of a host npm auth error (E401); this diff touches no frontend file. actionlint is not installed on this host.Manual verification
|Opus 5.5 Review (not started), GPT 6.1 Review (not started), the live pending. New:|, which publishes success.Related Issues
Part of #16373
Pattern harvest
Rule candidate: review-prompt
Pattern: a check-run rename must be paired with an audit of every reader that binds by name. Here the publish re-check bound by name while the verdict step did not, and the rename stranded heads that already had the old name.
Residuals (out of this PR's goal):
ai-review-human-override.yml's forkrerun_reviewerfinds the Stage-2 run by the current check name. On a pre-rename fork head,/ai-review override opus|gptreports no run to re-run. Readiness is not stuck by this, because a red old-name row publishes red. The way out is a manual re-run of the fork run or a push.workflow_runtrigger list matches workflow names. A re-run of a same-repo lane run created before the rename may not fire thein_progresshold. The publish re-check, which matches by workflow file path, and the sweep still cover that window.Checklist
/code-review maxand/simplify); code follows project style guidelinesdocs/ci/ci-and-reviews.md)Contribution License Agreement