Repository navigation
Description
Symptom
hivecommons/docs#500 (bot PR by github-actions[bot] via peter-evans/create-pull-request, same-repo head bot/hive-docs-refresh, labels role/trusted, needs-rebase, stage/070-fixing) received an owner-delegate decision:
/hive decision supersede now… Hive decided: Close this PR and open a fresh hive-authored PR with the same commits … expected within 30 min.
The decision comment (<!-- hive:delegated-decision by=hosted-available-oke-11-placeholder-r05x policy=owner-delegate -->, rule contributor-offline) was posted 2026-10-28T23:59:55Z and the adoption/supersede label applied at 2026-10-29T00:00:00Z. No fresh PR was opened, #500 was not closed, the label stayed on. (The PR had been needs-rebase since 09:19Z, hence the ~15h wait before the owner intervened and re-ran the refresh workflow.)
Mirrors #14666 / #14787: a decision is recorded but no executor runs it.
Evidence
- Logs (hosted hive, pod
hive-787656c98f-mp9fr, started before the decision,--since=24h, 16k lines): zeroadopt|supersede|decisionlines for docs#500. Onlyautomerge skip(trusted-bot-lane,contributor-lane:app-authored,label:needs-rebase) andstage labelslines, every one with"lane":"self-authored". - State (
/data/stage-labels.json, recordhivecommons/docs#500):author: github-actions[bot],adopted_at/adopt_checked_at/adopted_head_atall zero,activity_digestcontainsadoption/supersede→ the durable request is present, no attempt was ever recorded. - Control: hive#15105 (contributor lane) was adopted + superseded fine in the same hour by the same pod.
Root cause — (a) executor filters out bot/App-authored heads, silently
automerge_stage_labels.go runs the executor every sweep while the label is present (adoptContributorPR(...), label removed only once an attempt is recorded). But adoptContributorPR opens with
if !ok || obs.Lane == stageLaneSelf || obs.Author == "" || obs.BaseRef == "" {
return 0, nil
}sweepLaneForAuthor classifies github-actions as trusted-bot, and the trusted-author tier hands trusted-bot PRs to the self-authored sweep, whose recordStage stamps stageLaneSelf. So a github-actions PR is observed under lane self-authored → the executor returns silently on every sweep: no log line, no attempt recorded, label never cleared, decision never executed.
Ruled out: (b) replay never ran; (c) decision is durable (label + record on disk, pod predates the decision); (d) executor is in the engine sweep, not an agent prompt.
Secondary gaps found on the same path:
- under
maintainer_policy: never,authorMaintainsReporeads collaborator permission forgithub-actions[bot], which 404s → deferred forever; supersedeAdoptedPRnever replays a conflicting head onto the base: the "fresh PR" would open conflicting, and a failure to do so has no comment or stage reason.
Fix
- Run the executor for trusted-bot PRs observed on the self-authored lane; refuse (once, with a note) only when the author is the App itself.
- Skip the collaborator-permission read for
[bot]authors. - Before opening the copy for a conflicting head, merge the base into the hive branch; on 409 post ONE comment with the concrete failure, consume the request (label cleared, no live adoption), delete the hive branch and keep the PR at its fixing stage (070/071/072) with reason
supersede-conflict. - Log every early return for a requested adoption.
Activity
github-actions commentedon Oct 29, 2026
@clubanderson this issue was closed by @hivecommons-hive[bot] as completed.
If this was closed by a merged PR, the fix only takes effect once it ships in a tagged release and your deployment updates to it — check your running version against the fix before concluding the fix failed.
If you are still seeing the problem, comment
/reopenand it will be reopened for you — you do not need write access. Please add what you are still observing (running version, logs, steps) so it is not closed the same way again.