Lane: babysit (safe tier)
Triggering example: melodic-software/claude-code-plugins#848 (also reproduced on #849, #859, #861, #869, melodic-software/dotfiles#280 in the same cycle). Snapshot reports branch_freshness.state == "behind" and mutation_policy.branch_write_allowed == true, so the orchestrator attempted the guarded refresh:
python refresh_pr_branch.py --pr melodic-software/claude-code-plugins#848 --expected-head-sha 3cf7fa23d020753095ad435e319e6c42b0cd3a97 --lease-token <token> --state-dir <state-dir> --apply
Result: ERROR: snapshot has an active human review stop.
Observed vs expected: The PR's only human_blocking_feedback_ids entry is comment:5032743107, authored by kyle-sexton (the babysit lane's own self-login) — a prior-cycle classification reply ("Post-green review classification (work lane): ... No action needed ... Flipping to ready, handing off to the babysit lane"). There is no actual open human blocker on this PR — human_changes_requested is false and the comment explicitly says no action is needed. Expected: a self-authored comment under the configured self-login should not count as an "active human review stop" blocking the freshness-refresh gate, the same way #473 fixed this for the needs_worker new_human_blocking_feedback arm. All 6 behind-base PRs in the current queue hit this identically, permanently blocking freshness refresh (and therefore the merge gate, which also needs a fresh branch) for any PR the lane has ever left a classification reply on.
Category: bug
Repro: any PR where branch_freshness.state == "behind" and the only human_blocking_feedback_ids entry is a self-authored (self-login) general/review comment.
Work-class: C3 (bug-fix-shaped) — attended triage 2026-07-23, operator-ratified. 🤖
Lane: babysit (safe tier)
Triggering example:
melodic-software/claude-code-plugins#848(also reproduced on#849,#859,#861,#869,melodic-software/dotfiles#280in the same cycle). Snapshot reportsbranch_freshness.state == "behind"andmutation_policy.branch_write_allowed == true, so the orchestrator attempted the guarded refresh:Result:
ERROR: snapshot has an active human review stop.Observed vs expected: The PR's only
human_blocking_feedback_idsentry iscomment:5032743107, authored bykyle-sexton(the babysit lane's own self-login) — a prior-cycle classification reply ("Post-green review classification (work lane): ... No action needed ... Flipping to ready, handing off to the babysit lane"). There is no actual open human blocker on this PR —human_changes_requestedisfalseand the comment explicitly says no action is needed. Expected: a self-authored comment under the configured self-login should not count as an "active human review stop" blocking the freshness-refresh gate, the same way #473 fixed this for theneeds_workernew_human_blocking_feedbackarm. All 6 behind-base PRs in the current queue hit this identically, permanently blocking freshness refresh (and therefore the merge gate, which also needs a fresh branch) for any PR the lane has ever left a classification reply on.Category: bug
Repro: any PR where
branch_freshness.state == "behind"and the onlyhuman_blocking_feedback_idsentry is a self-authored (self-login) general/review comment.Work-class: C3 (bug-fix-shaped) — attended triage 2026-07-23, operator-ratified. 🤖