Found by the control tower's merge digest of PR #634 (shared babysit classifier, #534).
count_classified in plugins/source-control/skills/babysit-prs/scripts/babysit_classify.py counts self-authored classification pipe-rows in ANY comment, including PR-level review-summary comments that are never thread-resolved. A stale classification posted outside a review thread keeps counting after its finding is discounted by the thread-state guard, so classified can be inflated past a fresh open finding — a fail-open path for READINESS_OK.
Requires a specific posting pattern (classification rows in non-thread comments), so lower likelihood than the truncation defect fixed on #634, but same gate surface and same failure direction. Candidate fix: only count pipe-rows that live inside review-thread comments, or apply thread-state discounting to the comment container before counting.
Related: #534, PR #634 (digest thread), #465 (open-state discount).
Found by the control tower's merge digest of PR #634 (shared babysit classifier, #534).
count_classifiedinplugins/source-control/skills/babysit-prs/scripts/babysit_classify.pycounts self-authored classification pipe-rows in ANY comment, including PR-level review-summary comments that are never thread-resolved. A stale classification posted outside a review thread keeps counting after its finding is discounted by the thread-state guard, soclassifiedcan be inflated past a fresh open finding — a fail-open path forREADINESS_OK.Requires a specific posting pattern (classification rows in non-thread comments), so lower likelihood than the truncation defect fixed on #634, but same gate surface and same failure direction. Candidate fix: only count pipe-rows that live inside review-thread comments, or apply thread-state discounting to the comment container before counting.
Related: #534, PR #634 (digest thread), #465 (open-state discount).