[kyle-sexton]

No related issue: companion to Automation Lanes' fix/verify-fail-label-hold, which has pr-verify label a person's PR pr-verify-failed on a FAIL.

Summary

ci-status enforces pr-verify-failed only in the merge-queue step. On the pull_request run it would deadlock pr-verify, which judges a head only after ci-status passes (.github/workflows/pr-verify.yml:74). So a labeled PR shows a green ci-status, pr-merge arms and enqueues it, and the queue ejects it, in a loop. This PR makes pr-merge hold the label in every place it already holds do-not-merge.

Fix

pr-merge has no config-read skip-label list today. do-not-merge is a constant in two places, and this PR adds pr-verify-failed beside it in both:

• plugins/source-control/skills/babysit-prs/scripts/babysit_merge_policy.py MERGE_HOLD_LABELS. It feeds merge_hold_label, which covers the gate's blockers (no merge or enqueue) and pr-merge's candidate prefilter (babysit_merge_prefilter.py). It also feeds adoption_refusal (adopting a person's idle PR) and entry_hold, the queue sweep that removes a queued entry once it carries a hold.
• .github/scripts/pr-merge/enable-auto-merge.mjs HOLD_LABELS, the mirror read by holdReason. That job does not arm a labeled PR, and it disarms one that is already armed, on scheduled runs and on event runs. It arms only lanes App PRs, so for a person's PR the Python list above is what stops the loop.

Left unchanged on purpose:

• The labels-absent lane filters in docs/conventions/pr-pipeline.yaml: they would also stop the lanes that turn the FAIL into a PASS.
• babysit_merge_block_labels / --block-labels: an additive per-tier config, not the merge hold.
• The draft-recovery skip in babysit_whole_review.py, check-wip-cap and apply-patch: none of them arms or enqueues.
• gate-check.mjs and superseded.mjs: neither has a label skip.

The next such label still takes an edit to both constants.

Verification

The new tests were added first, and all four failed against the unchanged constants:

• enable-auto-merge.test.mjs, "a pr-verify-failed pull request is not armed, and an armed one is disarmed": schedule and event runs.
• test_babysit_merge_policy.py, EntryHold.test_the_pr_verify_failed_label and AdoptIdleOwnerPullRequests.test_the_pr_verify_failed_label_holds.
• test_babysit_merge_prefilter.py, test_a_pr_verify_failed_candidate_is_dropped.

The existing do-not-merge tests are untouched and still pass. After the fix, the .github/scripts/pr-merge node suite passed 608/608 and the babysit-prs Python suite passed 3344.

Related

• Automation Lanes fix/verify-fail-label-hold: pr-verify adds and removes the label, and ci-status enforces it in the merge queue.
• perf(source-control): carry completed review-lane runs across pr-merge runs #10666 and fix(pr-merge): count a stale-dismissed approval and carry it over an exact base merge #10671 touch nearby files. This change avoids their hunks.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T15:10:02.222105Z	0efc4d5	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed f85be444eaa1d8dab142276690a21c57286f901c: 0 finding(s), 0 as inline comments.

Incremental review of the changes since 0efc4d5 found no security issues.

• .github/scripts/pr-merge/enable-auto-merge.mjs: the only change is exporting HOLD_LABELS. Its contents are unchanged (do-not-merge, pr-verify-failed), so no merge hold is weakened or narrowed. The module already exports many constants, and its main() runs only behind the import.meta.url === pathToFileURL(process.argv[1]) guard at line 1567. Importing it from a test therefore has no side effects. The array is Object.freezed, so an importer cannot mutate it.
• plugins/source-control/skills/babysit-prs/scripts/tests/test_merge_holds_sync.py: this is a new test that compares the JS HOLD_LABELS with policy.MERGE_HOLD_LABELS. It spawns node with fixed arguments, and json.dumps(module.as_uri()) builds the module path from the repository root, not from untrusted input. It is test-only code and guards against the two hold lists drifting apart.

No trust-control widening, no manifest changes, and no hold needed. The change spends no billable resources.

critical-findings: 0

[github-actions]

claude-review reviewed f85be444eaa1d8dab142276690a21c57286f901c: 0 finding(s), 0 as inline comments.

Incremental review of the change since 0efc4d5: no findings.

• .github/scripts/pr-merge/enable-auto-merge.mjs (trusted path): HOLD_LABELS is now exported. The value is unchanged and the array is still frozen. The file's main() only runs behind the import.meta.url === pathToFileURL(process.argv[1]) guard, so importing the module from the new test starts no run and has no side effects. The export does not widen any gate.
• plugins/source-control/skills/babysit-prs/scripts/tests/test_merge_holds_sync.py: the new TheHoldLabelsAreTheGates test imports the module under node and compares HOLD_LABELS with policy.MERGE_HOLD_LABELS (("do-not-merge", "pr-verify-failed")). The comparison is by order as well as content. Both lists are in that order today. It follows the sibling tests' skip-if-no-node pattern and closes the "next label takes an edit to both constants" gap the PR description names. I read both sides and did not run the test, so this is a static read.

Spend: no cost hold. The change touches no paid plans or billing.

[github-actions]

Post-merge mutation testing

• Range: 05bd117..f7371d8
• Merged pull request: #10676
• Outcome: reported

Killed 3, survived 3, timeout 0, no coverage 0, invalid 0, not run 2.
Covered-code score 50%; mutation score 50%.

File	Killed	Survived	Timeout	No coverage	Invalid	Not run
.github/scripts/pr-merge/enable-auto-merge.mjs	0	0	0	0	0	1
plugins/source-control/scripts/worktree-claim.sh	3	3	0	0	0	0
plugins/source-control/skills/babysit-prs/scripts/babysit_merge_policy.py	0	0	0	0	0	1

Surviving mutants

Location	Operator	Disposition	Cause
plugins/source-control/scripts/worktree-claim.sh:151	SBR	productive	input-gap
plugins/source-control/scripts/worktree-claim.sh:160	UOI	productive	input-gap
plugins/source-control/scripts/worktree-claim.sh:161	SBR	productive	input-gap

The run's summary has the full report.