Repository navigation
dev-lead claim retraction does not cover issue-comment dispositions: a fixed reply for a commit that never landed stays up #2156
Description
Problem
When a dev-lead pass answers a PR issue comment it posts a reply marked disposition=fixed that cites a commit. If that commit never lands, the reply stays up saying "fixed". The claim-retraction sweep from #2013 does not look at these replies: retract_unlanded_claims (scripts/dev-lead-fix-reviews.sh) lists only pulls/{pr}/comments, the review-thread replies.
The verified-disposition minimizer (#1813) does check the cited SHA, so the original comment is not wrongly resolved. The gap is the reply's text: a person or a review bot reading the PR sees "fixed" for a change that is not on the branch.
Evidence
PR #2080, 2026-10-09 at 02:42Z: a pass posted a review-thread claim (comment 4226119528) and an issue-comment disposition marked disposition=fixed (comment 6073169970), both citing 9313650. That commit never reached the branch: an auto-rebase merge of main had landed at about 02:40Z and the runs on the branch were cancelled. The review-thread claim was retracted about 15 minutes later by a later pass. Nothing retracts the issue-comment reply.
Acceptance criteria
- Issue-comment
fixeddispositions are swept. Adisposition=fixedreply posted as a PR issue comment whose cited commit is not on the remote head is retracted the same way as a review-thread claim: the disposition marker is removed, a retraction notice is added, and the original text is kept as a quote. - Both sweeps cover it. It applies to the end-of-pass sweep (
retract_unlanded_claims) and to the start-of-pass sweep of earlier passes added for dev-lead claim retraction: replies from cancelled runs are never swept, and a clean rebase retracts claims for fixes that landed (follow-up to #2013) #2032. - Other dispositions are untouched.
invalid,answered,informationalandout-of-scopereplies cite no commit and are never retracted by this. - Tests. bats, with a
ghstub: afixedissue-comment reply citing an unlanded commit is retracted; one citing a landed commit is kept; a non-fixeddisposition is left alone.
Sequencing
Build this on the sweep functions from #2032 ("dev-lead claim retraction: replies from cancelled runs are never swept, and a clean rebase retracts claims for fixes that landed"). It is blocked by that issue.
Activity
github-actions commentedon Oct 9, 2026
Thanks for the thorough report, and for including the PR #2080 evidence with the comment IDs. This looks like a clear gap:
retract_unlanded_claimsonly listspulls/{pr}/comments, sodisposition=fixedreplies posted as issue comments are never swept. One thing to confirm before work starts: since this is blocked by #2032, should issue-comment replies be identified purely by thedisposition=fixedmarker plus a cited SHA, or is there another marker the sweep should also match to avoid retracting replies written by humans?don-petry commentedon Oct 10, 2026
Dev-Lead Implementation Plan
Issue: #2156 — dev-lead claim retraction does not cover issue-comment dispositions: a fixed reply for a commit that never landed stays up
Scope
Extend both claim sweeps (end-of-pass
retract_unlanded_claimsand start-of-passsweep_earlier_claims) to also list the PR's issue comments and retract ourdisposition=fixedreplies whose cited commit is not on the remote head.Implementation Checklist
scripts/lib/claim-landing.sh: purecl_select_fixed_dispositions <comments_json> <bot_user> <since_iso> <pass|earlier>(ourdisposition=fixedissue-comment replies in the time window →<id>\t<sha>; a malformedfixedmarker yields an empty sha = unverifiable) andcl_retract_disposition_body <body> <reason>(strips every disposition marker, prepends a retraction notice, quotes the original, appends<!-- dev-lead:retracted reason=… -->). Other dispositions are never selected.scripts/dev-lead-fix-reviews.sh: newsweep_fixed_dispositions <ref> <pass|earlier>that listsissues/{pr}/comments, checks each cited SHA against the remote head, and PATCHesissues/comments/{id}. In the end-of-pass mode a SHA the push guard rebased is re-pointed at its landed successor instead of retracted (as for thread claims, dev-lead claim retraction: replies from cancelled runs are never swept, and a clean rebase retracts claims for fixes that landed (follow-up to #2013) #2032); in the start-of-pass mode a commit GitHub knows (pushed then rewritten) or one missing from a shallow checkout is left alone, as for thread claims.retract_unlanded_claimsandsweep_earlier_claims(failures fold into their return code).comment-disposition-verify.shbeforeclaim-landing.shso the single disposition parser is reused.Tests to Write
tests/dev-lead/unit/test_claim_sweep.bats: earlierfixedissue-comment reply citing an unlanded commit is retracted (marker removed, notice added, original quoted); one citing a landed commit is kept;invalid/answered/informational/out-of-scopereplies are untouched; end-of-pass sweep retracts this pass's unlandedfixedreply and keeps a landed one.tests/dev-lead/unit/test_claim_landing.bats: unit tests for the two pure helpers (selection window, non-fixedexcluded, retracted body carries no disposition marker and is idempotent).Test Command
bats tests/dev-lead/unit/test_claim_sweep.bats tests/dev-lead/unit/test_claim_landing.bats(full:bats -r tests/)don-petry commentedon Oct 10, 2026
Dev-Lead: Implementation Complete — PR #2231
The implementation for issue #2156 is durable: commits are pushed and a pull request is open. This record is backed by verifiable artifacts you can confirm in one click.
f7cc3eb3ef70ae8f53ac1699bf559aede7a40316main: 4 file(s), +446/-8 linesReview happens on the PR. Acceptance criteria and test results are evidenced by the PR's diff and CI checks — not asserted here. If a later stage fails, this claim is retracted in place.
Merge origin/main and rework the #2261 fix around #2156