[thatguywob]

Fixes #14816

Owner approvals with rationale now stay approvals instead of scheduling a cosmetic fix that invalidates the approved head. The full rationale remains in the durable decision record. Commands beginning with approve, approved, lgtm, or yes approve; change requests start with the instruction.

• Read /hive decision at 070 as well as 085, and let a later trusted APPROVED review on the same head supersede the stored request so metadata-only fixes need no new commit. Approval selection uses each reviewer's latest opinion across all review pages.
• Preserve non-closing references when a Hive decision reply is recorded or the PR body has been edited. Treat edited Refs conservatively as intentional, including a downgrade from Closes, rather than restoring automatic issue closure. Pasted decision markers from unrelated commenters do not count.
• Read trusted head-scoped contributor change requests before admission/verdict early returns, so requires_human yields to 072 and the existing author nudge path.

Added regression coverage for rationale parsing and retention, same-head 070 release, stale/untrusted/wrong-head approvals, superseded reviews, owner-directed references, and contributor routing. Updated maintainer command documentation and added a changelog fragment.

Validation: gofmt -l on all changed Go files returned no output. Tests and lint are reserved for CI by AGENTS.md and were not run locally. Merge gates and trusted-merger requirements remain in force.

— hive: backend=codex

---

🐝 Hive Agent: contributor | SHA: 6c8d05c7e

[hivecommons-hive]

@thatguywob this PR is back in your court (stage/072-returned-to-contributor). The hive will not merge it until the head changes.

Failing checks:

• test (rest 1/4)

Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path.

Queued in your relay fix queue (1 waiting): fixes are handed out before new work as soon as you have capacity.

[hivecommons-hive]

Read the diff against v5 plus the callers (automerge_sweep.go:1646, automerge_review_gate.go, review/human_change_request.go Record/For, pr_request_watcher.go:501). The four changes do what #14816 asks and I found no defect. Looks correct to me.

What I checked:

• Approval parsing (src/pkg/github/issue_unpark_command.go:102-115): first-word match on DecisionText() is the semantics #14816 §1 proposed; the only caller is prDecisionModeFor (pr_unpark_decision.go:139), so issue-side /hive decision is unaffected.
• Same-head approval releasing 070 (automerge_human_changes_requested.go:101-116): Record keeps the standing changes-kind entry (different SourceID, same head) and For returns the newest RecordedAt, so the approve-kind record outranks it on the next pass exactly like the dashboard direction: approve path already does. The After(standing.RecordedAt) guard is sound now that review-sourced requests stamp RecordedAt from SubmittedAt (line 133).
• Shared latestTrustedHumanReview: latestTrustedHumanApproval now uses latest-opinion-per-login, so an approval later superseded by the same reviewer's CHANGES_REQUESTED no longer counts. That is a tightening of the review gate, and a correct one.
• lastEditedAt == "" guard (pr_issue_link.go:331): the attribution trailer is appended at creation (pr_request_watcher.go:501), not by a post-open edit, so freshly opened hive PRs still get the Refs→Closes repair; only edited bodies are left alone, as the body states.

One low-severity note, not a blocker:

• issue_unpark_command.go:110 keeps yes in the prefix set. Under the old exact-match BareApproval that was harmless; as a prefix it turns /hive decision yes, but also change X into mode=approve and silently drops the instruction. #14816 §1 proposed approve/approved/lgtm only. Suggest dropping yes from the prefix list (or keeping it bare-only) so a hedged answer still reaches the fix lane.

Open question I could not verify from the tree: #14816 §4 (crux#22 staying at 085) is addressed by the early return at automerge_contributor_lane.go:1222, but the issue also floated crux being outside the trusted-merger config. If it is, this change will not move that PR — worth a look at the crux lane config after merge rather than a change here.

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

[hivecommons-hive]

Read the tree at 09d61ee (not just the diff): automerge_human_changes_requested.go, automerge_contributor_lane.go:1210-1265, pkg/review/human_change_request.go (Record/For), issue_unpark_command.go, pr_issue_link.go. Per #14816 item 1 the first-word approval rule is the owner's stated choice, so I judged against that, not against the old bare-only behaviour.

What checks out

• latestTrustedHumanReview keeps the newest of APPROVED|CHANGES_REQUESTED|DISMISSED per login before filtering, so an older approval can no longer shadow a later change request on the same head — the right fix for item 4.
• The approval-clears-standing-request path is sound: HumanChangeRequests.For returns the newest record by RecordedAt (human_change_request.go:219), and the approve record is stamped with the review's SubmittedAt, which the guard already required to be after standing.RecordedAt, so the next pass sees IsApproval() and stops scheduling a fix.
• No textual conflict with #14821 even though both touch pr_unpark_decision.go (git merge-tree of the two heads is clean).

Two things worth fixing before this lands

1. yes as a first word now approves and discards the rest — src/pkg/github/issue_unpark_command.go, Approval() case list ("approve", "approved", "lgtm", "yes"). /hive decision yes, but rename the flag to --foo first was changes_requested with the text kicked to the author; it is now approve, and the instruction becomes inert rationale. #14816 proposes only approve/approved/lgtm for the prefix form; yes is far more often the opener of a conditional answer. Suggest keeping yes bare-only. (medium)
2. A trailing comma defeats the new rule — same function, strings.TrimRight(words[0], ".:"). /hive decision approve, the body change is applied yields words[0] == "approve," → not matched → changes_requested → fix kick, which is exactly the loop item 1 describes. Add ,; (and the em dash the tests use after a space) to the trim set, or split on strings.FieldsFunc. (low)

Open question, not a defect: pr_issue_link.go now skips the Refs→Closes repair whenever lastEditedAt != "". Any body edit — including the hive's own at pr_contributor_link.go:109 and pr_supersession_resolve.go:303, or an agent editing its PR description — permanently disables that repair for the PR. It fails toward leaving the issue open, which is the safe side, so fine if intended; flagging so it is a choice rather than a surprise.

Minor: the contributor lane now lists reviews three times per PR per tick (latestTrustedHumanApproval, then the new early latestTrustedHumanChangesRequested, then hasOutstandingChangesRequested). Not blocking; could reuse one read.

I have not approved, merged, or changed anything.

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

[github-actions]

Open PR overlap check

Base branch: v5

This PR conflicts with v5 itself. Rebase it first: until it merges cleanly into the base, conflicts with other open PRs cannot be told apart from base drift, so only same-file overlaps are listed.

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
#15233	🐛 fix: preserve owner intent through PR decision paths (adopted)	same files	src/docs/maintainer-commands.md src/pkg/github/automerge/automerge_contributor_lane.go src/pkg/github/automerge/automerge_contributor_lane_test.go src/pkg/github/automerge/automerge_human_changes_requested.go src/pkg/github/automerge/automerge_human_changes_requested_test.go src/pkg/github/automerge/automerge_review_gate.go src/pkg/github/issue_unpark_command.go src/pkg/github/issue_unpark_command_test.go src/pkg/github/pr_issue_link.go src/pkg/github/pr_issue_link_test.go src/pkg/github/pr_unpark_decision.go src/pkg/github/pr_unpark_decision_test.go
#15186	🐛 Activity Monitor: one coherent post-decision state for Let hive decide rows; 060 reviewed in CI / Review	same files	src/pkg/github/issue_unpark_command.go
#15159	🌱 chore: remove unreachable MITM cancellation guard	same files	.github/workflows/v2-ci.yml
#15104	🐛 Activity Monitor: Rebase, then merge as maintainer; fixer owns parked conflicting PRs; dismiss self-inflicted changes_requested	same files	src/pkg/github/issue_unpark_command.go src/pkg/github/pr_unpark_decision.go src/pkg/github/pr_unpark_decision_test.go
#15216	🐛 Sentinel: a dashboard Approve releases the head for good; skip Prow-rejected /lgtm + /approve	same files	src/pkg/github/automerge/automerge_contributor_lane_test.go
#15226	✨ feature: enqueue eligible PRs into GitHub native merge queues	same files	src/pkg/github/automerge/automerge_contributor_lane.go src/pkg/github/automerge/automerge_contributor_lane_test.go

[hivecommons-hive]

@thatguywob this PR is back in your court (stage/072-returned-to-contributor). The hive will not merge it until the head changes.

Failing checks:

• overlayfs-exec-guard

Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path.

Queued in your relay fix queue (1 waiting): fixes are handed out before new work as soon as you have capacity.

[hanthor]

P2: Reconcile persisted review-backed approvals after dismissal (src/pkg/github/automerge/automerge_human_changes_requested.go:104-110). A later same-head APPROVED review is now persisted as an approval-kind durable decision. If that review is dismissed before merge, live review selection rejects it, but humanChangesRequestedForHead returns early for standing.IsApproval(), recordedHumanApprovalForHead accepts the persisted record without revalidating its source review, and stage folding continues treating requires_human as answered. The revoked approval can therefore suppress needs-decision and prevent the prior instruction reaching the fixer. Reconcile review-backed records against current review state while preserving durable explicit comment/dashboard decisions; cover approval -> dismissal on the unchanged head.

Owner/workflow intent is already recorded for this head; this is not a request to override it. Before merge, fix the above, rebase onto current v5, obtain a successful overlayfs-exec-guard verdict (the current-head run exceeded its new 45-minute budget), and renew the head-scoped workflow approval after new commits. Preserve existing delegated-decision guards; do not waive the workflow credential fence or close this as superseded without an actual replacement.

[hivecommons-hive]

Hive decided close this PR and open a fresh hive-authored PR with the same commits at 2026-10-28T13:01:29Z, but it hasn't taken effect.

What should have happened: The hive closes this PR and opens a fresh hive-authored PR with the same commits and sign-off, then finishes that one.
Why it hasn't: The branch has merge conflicts with the base branch; it needs a rebase before anything can merge.
What you can do: Merge as maintainer (on GitHub) · Re-run decision · Close as superseded · Send back for a new commit.

Posted by the hive as the owner's delegate; the rule was: the author's hive hasn't been connected since the nudge (contributor-offline).

[clubanderson]

Adopted into #15233 per operator decision (adopt-supersede); your commits are preserved with authorship. Thanks @thatguywob!

[clubanderson]

Superseded by #15233 (adopted per operator decision; your commits are preserved with authorship). Closing so the tracker follows the live PR. Thanks @thatguywob!