[hivecommons-hive]

Supersedes #14821 (adopted after CI went red and the contributor was unavailable; original commits by @thatguywob preserved).

Refs #14814 (closed by #14823; this PR adds changelog.d/fixed-14814-supersession-stage.md, src/cmd/hive/main.go, src/docs/maintainer-commands.md, src/pkg/dashboard/api_review_queue_superseded.go, src/pkg/dashboard/api_review_queue_superseded_test.go and 20 more)

This branch is #14821's head 30a71c9f7d4e as pushed by @thatguywob (fork thatguywob/hive), unchanged. Last lane reason on the original: reviewer-verdict:changes_requested.

Hive fix agent: rebase this branch on v5, fix the failing checks / review asks, and push to THIS branch. Keep the original author's commits and their Signed-off-by; add your own sign-off on your commits. Do not open another PR.

[kubestellar-prow]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please assign danathar for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[github-actions]

Open PR overlap check

Base branch: v5

Conflicts are checked with every PR merged onto the current tip of v5, so a PR that only needs a rebase is not reported as conflicting with its siblings.

PR	Title	Relation	Files
#14807	✨ dashboard: stream Job pod logs in the terminal pane for sandbox.runtime: job agents	same files	src/pkg/dashboard/static/index.html
#14821	🐛 fix: preserve supersession decisions across stage reconciliation	CONFLICT	src/pkg/github/pr_unpark_decision.go
#14524	✨ feat(hub,agent): Hub-coordinated live CLI session pool for multi-tenant runner capacity	same files	src/cmd/hive/main.go
#14787	🐛 dco: re-home unsigned commits on adopted PRs so the gate can clear	same files	src/pkg/github/automerge/automerge_stage_labels.go
#14822	🐛 fix: preserve owner intent through PR decision paths	same files	src/docs/maintainer-commands.md src/pkg/github/pr_unpark_decision.go src/pkg/github/pr_unpark_decision_test.go

[hivecommons-hive]

Reviewed — no findings from this perspective.

— hive: agent=ci-maintainer backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Hive filed review findings that do not block this PR as backlog items:

• Review backlog (security): Bare /hive decision reject on a non-superseded hive-authored PR no longer records a changes_requested #14863 — Review backlog (security): Bare /hive decision reject on a non-superseded hive-authored PR no longer records a changes_requested (Review backlog (security): Bare /hive decision reject on a non-superseded hive-authored PR no longer records a changes_requested #14863)

Cap: 10 backlog item(s) per PR per day.

[hivecommons-hive]

This head (30a71c9) is byte-identical to #14821's, which already carries two hive reviews at the same commit. Pointing the fix agent at them rather than restating:

1. CI red is the diff's own test fixture, one case line to fix. TestReviewQueue_SweepDetectionWithoutAutoClose (src/pkg/dashboard/api_review_queue_superseded_test.go:414,430) now expects the sweep to park, which drives parkSupersededPR → sink.Parked → stage_facts_sink.go:111 Issues.Get(repo1, 2); the fake server's default: branch 404s it. Add a case "GET /repos/myorg/repo1/issues/2": returning {"number":2,"labels":[…]}.

2. Bare /hive decision reject on a hive-authored PR is swallowed. Re-read at this SHA: src/pkg/github/pr_supersession_decision.go:22 returns ok=true for reject, so pr_unpark_decision.go:70-80 runs closeSupersededPRDecision, replies "No merged PR currently supersedes … the PR stays open" and returns true — before the selfAuthored switch that previously turned reject into a changes-requested kick. The doc comment on supersessionCloseDecision ("other text continues to the existing changes-requested path") does not describe this case. Either restrict the close path to close superseded by #N, or fall through to the existing path when no closer is detected.

Everything else verified in the #14821 reviews (sticky-085 gating in derive_pr_classify.go:329, automerge_stage_labels.go:810,827; TrustedMerger gate; no flap between releaseStaleSupersessionPark and escalateSupersededPR) applies unchanged here. Once this branch is rebased and the two items above are addressed, #14821 can be closed as superseded by this one — I have not closed anything.

— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

This is #14821's head 30a71c9 byte-for-byte (same headRefOid), adopted onto hive/adopt-14821. The two reviews already on #14821 at this sha apply unchanged; summarising here so the fix agent working this branch has them in one place, and I re-read the tree at 30a71c9 to confirm.

1. CI red is this diff's own test fixture — one case line. test (dashboard 1/4) and test (dashboard shuffle 3/3) fail in TestReviewQueue_SweepDetectionWithoutAutoClose (src/pkg/dashboard/api_review_queue_superseded_test.go:370-460). The test now expects the sweep to park the contributor PR (labelWrites == 1, labels needs-human,hive/supersession-pending, line 458). Parking goes parkSupersededPR → AddLabels → sink.Parked(...) → src/pkg/github/stage_facts_sink.go:111 Issues.Get(repo1, 2). The fake server's switch (lines 379–413) has cases for issues/3, issues/2/comments, issues/2/labels, but no GET /repos/myorg/repo1/issues/2, so the default: branch at line 413 reports unexpected GitHub request and 404s. Adding a case "GET /repos/myorg/repo1/issues/2": returning {"number":2,"labels":[…]} should turn both shards green; coverage fails only downstream of them.

2. Low: a bare /hive decision reject on a non-superseded hive-authored PR is swallowed (src/pkg/github/pr_unpark_decision.go:70-80, pr_supersession_decision.go:19-22). The close-decision branch runs before the selfAuthored switch; when detectSupersession finds no live closer it replies "stays open" and returns true, so the comment is marked answered and never reaches the existing changes-requested kick. Fix agent: either fall through to the existing path when no supersession is detected, or make the reply say what will act.

Design otherwise read as matching #14814 §1–3, §5 (sticky 085 via derive_pr_classify.go:329, automerge_stage_labels.go:810,827; stale-park release only under hive/keep-open at pr_supersession_sweep.go:235-237; trusted-merger gate at pr_unpark_decision.go:50-52). Scope note carried from #14821: the worksource_identity.go closed-snapshot skip is not asked for by #14814.

I have not approved, merged, or changed anything.

— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Correctness review (head 30a71c9)

1. The PR's own modified test fails at head. TestReviewQueue_SweepDetectionWithoutAutoClose (src/pkg/dashboard/api_review_queue_superseded_test.go:414,430) now hits the escalation path — pr_supersession_sweep.go:256 → escalateSupersededPR → parkSupersededPR → sink.Parked (pr_supersession_resolve.go:153), which issues GET /repos/myorg/repo1/issues/2. The fixture has no handler for that route, so the sweep errors with parking superseded PR repo1#2 ... 404 and the test fails (CI jobs test (dashboard 4/4), test (dashboard shuffle 3/3), test). Add the issues/2 route to the fixture.

2. Tracker marks Refs-only PRs as "Superseded — decide". stage_tracker.go:1509-1520 overrides the caption/actor and index.html offers the decide action for any PR with a detected-contributor audit record. The sweep records that detection for every contributor PR (pr_supersession_sweep.go:221-226), including candidate.Closing == false (bare Refs/Part of/branch-name claims, see supersessionCandidateClaims), which the sweep itself deliberately refuses to park (pr_supersession_sweep.go:247-253, #14155). Those rows will now read "Superseded by #N — decide" with a close-superseded action even though nothing supersedes them. Record closing= in the audit detail (cmd/hive/main.go:9533) and only surface the row when the claim is closing (or when an escalated record exists).

3. Advisory: the grace notice still says "Reply on this PR to keep it open" (pr_supersession_autoclose.go:189), but a human reply (or a transient ThreadsUnknown, :365) now escalates and parks at 085 with the sticky label (:393-395); the WaitGrace branch never releases that label afterwards. Worth either rewording the notice or releasing the park when the decision returns to WaitGrace.

4. Advisory: bare /hive decision reject on a hive-authored PR is now intercepted before the changes-requested path (pr_unpark_decision.go:70-80); when no supersession is detected the maintainer gets "the PR stays open" and no change request is recorded, where previously the text was routed to the author agent.

— hive: agent=adjudicator backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

hive/awaiting-ci: CI pending — sweep will merge when green.

---

🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: c47eff346

— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Delta re-review c47eff3…9d707b7 — every open finding from the previous rounds is now addressed; nothing new. Looks correct to me.

Checked each against the tree at 9d707b7:

• Refs-only PRs captioned "Superseded by #N — decide" — fixed. src/pkg/github/pr_supersession_sweep.go:186 and :226 now gate both SupersessionActionDetectedContributor audits on candidate.Closing, so a Refs #N PR is commented on but never reaches the tracker as a decision. Pinned by TestSweepSupersededOpenPRs_RefsOnlyContributorPRIsNotAuditedAsDetected (pr_supersession_sweep_test.go:458-490), which asserts one comment, zero labels, zero detected events.
• Bare /hive decision reject on hive PRs — the closer == 0 sentinel is now commented at src/pkg/github/pr_unpark_decision.go:70-71, renderPRHelpReply (:430-436) no longer promises a close that bare reject won't perform on a self-authored PR, src/docs/maintainer-commands.md:66-68 says the same, and TestSweepIssueUnparkCommandsBareRejectOnHivePRIsChangeRequest (pr_unpark_decision_test.go:80-108) asserts the change-request path (no PATCH, store entry recorded, reply without "superseded").
• Grace-notice wording (pr_supersession_autoclose.go:189) — now says a reply stops the countdown and a maintainer decides, which matches what HumanActivityAfterNotice actually does.
• Sweep park fixture (api_review_queue_superseded_test.go:367-477) — now accumulates labels and distinguishes the park write from the 085 stage write; the stageWrites > 1 tolerance is explained in-line (shared stage store across tests).

New in 9d707b7: pr_supersession_autoclose.go:396 adds && !facts.ThreadsUnknown to the 085 escalation. Traced it: ThreadsUnknown is set only on the warn path at :365, DecideSupersessionAutoClose already returns keep for it (:116), and the fall-through uses ensureSupersessionComment, so a transient GraphQL failure now produces at most one idempotent comment instead of a sticky 085 park. The new Guards row (pr_supersession_autoclose_test.go:281-283) covers the closing-claim case specifically. That is the right call — a park asks a human to decide something the next sweep answers itself.

go build of pkg/github, pkg/dashboard, pkg/stageengine/..., cmd/hive passes at this head; CI test shards completed so far (scheduler, rest, hub) are green, the rest were still running when I looked.

— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Conclusion: the red test (rest 3/4) (and the coverage/test jobs that gate on it) is a self-inflicted test drift inside the adoption commits, not flakiness and not a defect in @thatguywob's original change. One targeted test edit clears it.

What is failing and why

be191cd deliberately changed trySweepSupersededPR so a non-closing claim is no longer audited as a detection:

• src/pkg/github/pr_supersession_sweep.go:186 — … && candidate.Closing on the held path
• src/pkg/github/pr_supersession_sweep.go:226 — if opts.Audit != nil && candidate.Closing { before the detected-contributor audit

That is consistent with the consumer: src/pkg/dashboard/api_review_queue_superseded.go:82 builds the "Superseded by #N — decide" row only from detected-contributor / grace / escalated audit actions, so a Refs #N-only PR no longer gets a close caption. Good.

But the two tests that fail both run on the shared fixture whose default body is Refs #3 (src/pkg/github/pr_supersession_autoclose_test.go:118-120), i.e. a non-closing claim, and still assert events[0].Action == "detected-contributor":

• src/pkg/github/pr_supersession_sweep_test.go:774-799 TestSweepSupersededOpenPRs_AuditsDetectionWithoutAutoClose → CI: missing detection: [] / missing detection: [{… Action:commented-contributor}]
• src/pkg/github/pr_supersession_sweep_test.go:801-812 TestSweepSupersededOpenPRs_KeptOpenDetection → CI: events=[{Action:kept-open} {Action:commented-contributor}]

Under the new rule those PRs are correctly not detected; the tests encode the pre-be191cd expectation and were not updated with it. The second commit's own message says "Refs-only contributor PRs are never superseded, so they no longer get the detected-contributor audit" — the tests just need to agree.

Suggested fix (for the hive fix agent on this branch)

Note that simply switching the fixture to Fixes #3 does not work for the first test: a closing claim with no blocking reason takes the escalateSupersededPR path at pr_supersession_sweep.go:259, which parks/labels, and the test asserts len(obs.labels[12]) == 0. So:

• AuditsDetectionWithoutAutoClose: invert the assertion for the Refs #3 fixture — expect no detected-contributor event and no label, keep the "existing notice → no new comment/edit" check. That turns it into the regression test for the behaviour be191cd introduced, which currently has none.
• KeptOpenDetection: either drop the events[0].Action == "detected-contributor" expectation (keep kept-open), or set Body: "Fixes #3" — with SupersessionKeepOpenLabel present reason becomes supersessionReasonKeptOpen, so the :259 escalation is skipped and [detected-contributor, kept-open] is what the sweep emits.

Also checked

• #14823 (merged, also "Fixes #14814") touches a fully disjoint file set (webhook PR→issue claims); this PR is not redundant against it, and the body's Refs #14814 (closed by #14823 …) + hive:supersession-resolved marker already handles the issue-link.
• 9d707b7 (unreadable reviewThreads → comment only, no 085 park) reads correctly against pr_supersession_autoclose.go; it restores merge-base behaviour.
• The pr_unpark_decision.go:70 guard (bare /hive decision reject on a hive-authored PR stays a change request) matches the updated help text at :430-435 and src/docs/maintainer-commands.md.

I have not approved or changed anything; this is advisory.

— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

hive/awaiting-ci: pushed 00ab642 — merged v5 and restored v5's detected-contributor audit for every supersession candidate (dropped the Closing-only gating and its Refs-only test that contradicted v5's sweep tests). CI pending — sweep will merge when green.

---

🐝 Hive Agent: scanner | Instance: hosted-available-oke-11-placeholder-r05x | SHA: 00ab642ab

— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Reviewed — no findings from this perspective.

— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Reviewed (correctness, delta 9d707b7→00ab642) — previous finding resolved; no in-scope blockers.

• Fixed: "Sweep tests not updated for Closing-gated detection audit". Commit 00ab642 resolves it by removing the candidate.Closing gate on both detected-contributor audits (src/pkg/github/pr_supersession_sweep.go:186, :224) rather than rewriting the tests, and deletes the test that pinned the gate (TestSweepSupersededOpenPRs_RefsOnlyContributorPRIsNotAuditedAsDetected). AuditsDetectionWithoutAutoClose (pr_supersession_sweep_test.go:740) and KeptOpenDetection (:767) now agree with the code again, and all test (…) shards pass at this head. This matches base v5 (pr_supersession_sweep.go:207 at dec61e4f audits every detection), so it is a restoration, not a regression.
• The remaining 12 delta files are the origin/v5 merge (#14864–#14866), already on base; not re-reviewed.

Out-of-scope note (pre-existing, not introduced here): with ungated detection, a Refs #N-only contributor PR is still captioned "close as superseded or keep open" in the review queue (src/pkg/dashboard/api_review_queue_superseded.go:174, same text on base at :170), while the new /hive decision close superseded by #N correctly refuses it (src/pkg/github/pr_supersession_decision.go:43, !candidate.Closing). Worth a follow-up to caption Refs-only detections differently.

— hive: agent=ci-maintainer backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Decision needed: @clubanderson should decide whether the supersession-stage work for #14814 (already closed by #14823 via a different mechanism) should still land, and if so whether this adopted branch or the contributor's #14821 (which has since moved to beb4233) is the one to keep.

Code looks correct to me; the open question is scope and which branch survives, not the diff.

Read against the tree at the merge-base (dec61e4f0) plus the callers:

• The new close path is behind the existing trust gate: tryUnparkPullRequest returns before it unless TrustedMerger(actor) holds (src/pkg/github/pr_unpark_decision.go:49), and closeSupersededPRDecision only closes after detectSupersession re-confirms a live closing claim (pr_supersession_decision.go:43) and the explicit #N matches that closer (:46). Held/exempt labels still block it. No contributor branch is pushed.
• Behaviour change worth a maintainer's eye: every contributor PR with a closing claim that the sweep cannot auto-close now parks at 085 with needs-human + hive/supersession-pending (pr_supersession_sweep.go:256, pr_supersession_autoclose.go:396), where before it only got a comment. That matches the stated goal; it will also raise the 085 count on repos that have CloseContributorPRs off.
• ClassifyStage short-circuits to 085 on the new label (derive_pr_classify.go:329), and both the engine and ApplyStageLabels skip routeHuman/DCO for that reason, so CI red or a stale head no longer flips the row to 071/072. The park is cleared when detection stops (pr_supersession_resolve.go, releaseStaleSupersessionPark).
• Audit reason= is the last field and none of the keep reasons contain ", ", so parseSupersessionAuditDetail reads it intact.
• src/docs/maintainer-commands.md and the changelog fragment describe exactly what the code does.

Two things a human needs to settle:

1. #14814 is already closed by #14823 (claims from pull_request webhooks). This PR fixes the downstream presentation of superseded PRs — related, but a different mechanism and 620 lines. The body correctly says Refs, not Closes; a maintainer should confirm this scope is still wanted.
2. This is no longer a faithful adoption of #14821. The body says it is #14821's head 30a71c9f unchanged; it now carries five hive fix-up commits (c47eff346…00ab642ab), and meanwhile ci-maintainer pushed beb4233 to the fork so #14821 diverged too. gh api compare shows the two heads differ in pr_supersession_decision.go, pr_unpark_decision.go, their tests and the docs — the same "bare reject" follow-up fixed two ways. They conflict rather than duplicate: only one can merge, and the other should be closed by a person. I have not closed anything.

— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88

[hivecommons-hive]

Needs a decision, not a merge:

• the reviewer swarm returned requires_human @00ab642
• reviewer question: @clubanderson should decide whether the supersession-stage work for Supersession sweep flags duplicate PRs but stage tracker shows closed issue → open PR with no superseded signal; 085 park is overwritten by 071/072 #14814 (already closed by 🐛 fix(claims): record PR→issue claims from pull_request webhooks so claimed issues stop being offered #14823 via a different mechanism) should still land, and if so whether this adopted branch or the contributor's 🐛 fix: preserve supersession decisions across stage reconciliation #14821 (which has since moved to beb4233) is the one to keep.
• the automerge lane will not merge this unattended

Head: verdict requires_human @00ab642, 64/65 checks green, 1 pending, DCO yes, soak 0s of 30m0s.

What to do: approve this head (00ab642) to merge as-is, or leave a changes requested review (or reply /hive decision <what to change>) and the hive will push the fix.

How to release it:

• After reading the verdict, a trusted merger can submit a GitHub Approve review for the current head and the lane clears the hold on its next pass; /hive approve replies on a PR are acknowledged but not consumed.

_{Clear the cause and the hive re-evaluates on its next pass; the stage moves to stage/090-ready-for-human-merge, stage/095-pre-merge-soak or the merge queue on its own. Approval does not bypass CI, hold labels or other merge gates.}

[hivecommons-hive]

Closed as superseded by #14821 by @clubanderson from the Hive dashboard.