Repository navigation
Conversation
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v5.211.18 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.211.18 release: v5.211.18
Signed-off-by: Andrew Anderson <andy@clubanderson.com>
β¦ (76b20eb) (hivecommons#17561) Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Signed-off-by: Andrew Anderson <andy@clubanderson.com>
β¦c citations - stagelabels: TestDecisionCommentNamesActionsPerLane hold/sentinel cases follow hivecommons#17327 (gate rides the 090 route as a pill; ReadyComment names the labels to lift). - automerge: TestManualMergeNamedWhileHoldLabelBlocks (090 held), TestManualMergeNotClaimedWithoutWorkflowFence and TestWorkflowFenceScopeStageNoteAgreesWithLane/unlisted (071, no 085 note) via a manualMergeSweepTo helper taking the expected stage. - policies: redteam-holdgated.md carries the owner-only park rule (TestLanePoliciesParkOwnerOnlyHandOffs). - docs: api-reference citations resynced (check-api-reference-citations.sh --fix), scale-envelope.md RankActionableIssues range. Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Signed-off-by: Andrew Anderson <andy@clubanderson.com>
β¦ivecommons#17565) (hivecommons#17568) Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
β¦ PR (hivecommons#17564) (hivecommons#17567) Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: clubanderson <407614+clubanderson@users.noreply.github.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.12.1 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v5.211.19 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.211.19 release: v5.211.19
The v6 'who finishes it' chip and the v5 human-gate pill both render on a gated row; the hold/needs-human fixtures now expect the gate pill instead of per-modifier chips. Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Signed-off-by: Andrew Anderson <andy@clubanderson.com> # Conflicts: # src/pkg/stageengine/derive_pr_classify.go
β¦ without writes (hivecommons#17571) * π automerge: observe-only pass on auto-merge-off repos records stages without writes Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> * π automerge: observe only repos switched off at L6, per the hivecommons#17552 ruling The maintainer's decision on hivecommons#17552 (A) stages only repos an owner switched off with auto_merge: false at L6; hives below L6 stay as they are. The pass keyed off !RepoAutoMergeEnabled, which is false for every repo below L6, so it would have staged every PR on a sub-L6 hive. Add the narrower config.RepoAutoMergeSwitchedOff predicate, gate the pass and missingStageLanes on it, and make the L5 test case assert nothing is observed, staged or written. Also reword the auto_merge: false entry in the labels reference, which still said the disabled repo is skipped. Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> * π automerge: observe-only test matches the transport's short repo display name The lanes hand RepoAutoMergeSwitchedOff the transport's display name, which for NewClient("token", "acme", []string{"widget"}) is the short "widget". The L6 case compared against "acme/widget" only, so the switched-off predicate never fired and the trusted lane recorded no observation. Accept both spellings, as the existing RepoAutoMergeEnabled tests do; production's config.RepoPolicyFor normalizes either form. Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> * π automerge: name integration-member reasons; refresh auto-merge-off docs Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> --------- Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
β¦R mutation report (hivecommons#17569) * π± mutation: verdict phase and misrouted/uncovered detection for the PR mutation report The Phase 2 decision on hivecommons#17349 asks the mutation job to (1) tell a run that produced no mutants because the diff was not matched from one whose added lines simply contain nothing gremlins can mutate, (2) fail on survivors and (3) warn on uncovered mutants with a label override. All of that logic lives in the script, which this change adds; the workflow step that calls the new verdict phase is left for a human to land (see the issue). Run 42842443653 (PR hivecommons#17557) is the reference case for (1): gremlins v0.6.0 has no mutator for unary ! or for return values, so 'if !looksLikeUUID(id) { return "" }' yields zero mutants and the 108 SKIPPED ones all sit on unchanged lines. Counting added lines alone would have failed that PR wrongly; a SKIPPED mutant inside an added range is the actual signature of a matching fault. Refs hivecommons#17349 Signed-off-by: quality <280983584+hivecommons-hive[bot]@users.noreply.github.com> * π± mutation: honour gremlins' hunk approximation and fail closed on a missing report Review fixes for hivecommons#17569: - misrouted was judged against the exact -U0 added ranges, but gremlins v0.6.0 honours only start=+pos+leading-context, end=start+added-1 per default-context hunk, so a hunk with non-contiguous added lines made the script fail legitimate PRs. Derive that range (gremlins.tsv) from the same default-context diff; SKIPPED inside it is misrouted, SKIPPED on an added line outside it is a non-blocking hunk-approximation note. - A gremlins non-zero exit or missing JSON produced verdict=pass with a false "nothing to mutate" explanation. Record each failed package in errors.tsv, write verdict=error, report it in the comment, and make the verdict phase exit 1 on error and on a missing verdict file (hivecommons#17349 item 1). - has_label: make arr local. - test-mutation-report.sh: stub gains GREMLINS_EXIT/GREMLINS_NO_OUTPUT, new split-hunk and failing-stub cases, missing-verdict expectation flipped. Hive-Fix-Attempt: 1/6 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> * π± mutation: re-pin TRUSTED_SCRIPTS_TREE to the mutation-report.sh tree The second commit edited src/scripts/mutation-report.sh after the first re-pin, so runtime-tests' integration-governance contract failed on the stale literal. Pin to git rev-parse HEAD:src/scripts. Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> --------- Signed-off-by: quality <280983584+hivecommons-hive[bot]@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: quality <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
β¦5-20261110 π± Top up v6 with v5 (durable hosted hive.yaml, held-gate pills, stale CI chevron re-list and 117 more)
β¦t grace (hivecommons#17576) Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.12.2 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.12.2 release: v6.12.2
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v5.211.20 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.211.20 release: v5.211.20
hivecommons#17578) PRs authored by a login in auto_merge.trusted_bot_authors (Settings bot toggles or a custom login) were evaluated by the trusted-bot lane but held on the review swarm's requires_human verdict and parked at "needs a person", while the contributor lane logged them as app-authored and the self-authored sweep's skip/hold lines carried no tier. - The swarm verdict is advisory on the trusted-bot lane by default (auto_merge.trusted_bot_skip_review, default true): recorded on the evidence bundle, never a hold or an 085 stage. - The contributor lane hands trusted bots off as trusted-bot-lane; every self-authored sweep skip/hold line carries tier=<lane>. - A config-named required check that never reported on the repo gates on the other reported checks (one warning per repo per hour) or holds with required-check-never-reported:<check>, shown on the Activity Monitor as the human-gate reason. Fixes hivecommons#17577 Signed-off-by: Andrew Anderson <andy@clubanderson.com> Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
β¦e lane (hivecommons#17581) * π Honor every configured trusted bot in the trusted-bot automerge lane PRs authored by a login in auto_merge.trusted_bot_authors (Settings bot toggles or a custom login) were evaluated by the trusted-bot lane but held on the review swarm's requires_human verdict and parked at "needs a person", while the contributor lane logged them as app-authored and the self-authored sweep's skip/hold lines carried no tier. - The swarm verdict is advisory on the trusted-bot lane by default (auto_merge.trusted_bot_skip_review, default true): recorded on the evidence bundle, never a hold or an 085 stage. - The contributor lane hands trusted bots off as trusted-bot-lane; every self-authored sweep skip/hold line carries tier=<lane>. - A config-named required check that never reported on the repo gates on the other reported checks (one warning per repo per hour) or holds with required-check-never-reported:<check>, shown on the Activity Monitor as the human-gate reason. Fixes hivecommons#17577 Signed-off-by: Andrew Anderson <andy@clubanderson.com> * π Make the v6 review-verdict hold advisory on the trusted-bot lane v6's reviewVerdictGate (hivecommons#12374) held trusted-bot PRs on a requires_human ledger entry even after the cherry-pick; route it through reviewVerdictGateForLane so the trusted-bot lane logs the hold as advisory and merges unless auto_merge.trusted_bot_skip_review is false. Part of hivecommons#17577 Signed-off-by: Andrew Anderson <andy@clubanderson.com> --------- Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.12.3 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.12.3 release: v6.12.3
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v5.211.21 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.211.21 release: v5.211.21
β¦abel provisioning stops retrying every sweep (hivecommons#17585) Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.12.4 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.28.0 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.28.0 release: v6.28.0
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.28.1 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦ keeps and shows the reason (hivecommons#18000) * β¨ feat(contribute): a relay that drops an issue says why, and the hub keeps and shows the reason A task that ended with no pull request and no no_work_needed / blocked / wrong_base verdict reached the hub as `verdict=idle verdict_reason=""`. The issue showed "Completion cooldown until ..." in the Withheld list, was offered again and was dropped again, and nobody could tell why without the relay's own pane. The agent's sentence after `HIVE_VERDICT: complete β` was parsed by the relay and thrown away. Relay: every issue task that completes with no PR attributed to it now sends a verdict_reason and still no verdict, so the hub books it idle as before. `complete` with no PR carries the agent's own sentence, marked as such; a chrome-idle or headless exit with no verdict says the agent exited without a conclusion. No-work-family verdicts, completions with a PR, and fix-pr / review tasks are sent unchanged. An older hub already receives this shape for a tooling-failure blocked verdict. Hub: a no-PR idle is never recorded without a reason (the completion signal decides the wording when the relay sent none). The latest reason per issue, with when and which contributor, is kept in idle-completions.json beside the no-PR streak ledger, one line, token redacted and capped; a verified PR or a no-work verdict removes it. The Withheld list ends a completion-cooldown row with "β last relay: <reason>", and the Activity monitor's issue row carries the same clause as last_idle (dashboard tooltip and stage_tracker_read). Cooldowns, the no-PR streak and the evidence-less rule (hivecommons#6723, hivecommons#7862) are unchanged: the reason is explanation only. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(contribute): an idle booked although a pull request was reported says so A completion can be booked idle with a pull request on the wire: the hub could not verify it (a GitHub lookup that failed while the budget was spent, another author). The relay sends no idle reason then, because it did attribute a pull request, and the hub filled in "agent reported complete without a pull request and gave no reason", which is false. The hub now records what it made of the reported pull request, with the verifier's own reason: "relay reported a pull request the hub could not verify: github lookup failed". Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(contribute): the idle reason says only what is known, and leaves the row with its cooldown Four corrections to the idle reason this branch adds, so that what is stored and shown is true. The Activity monitor note is returned only while the completion cooldown that idle booked is still in force (isTaskInCooldownKey, the gate the Withheld list's cooldown row is shown under). It used to stay on the row for the ledger's 14 days with no age. A task that later fails or is let go on the issue now removes the idle reason from the ledger, because a failure leaves the completion cooldown running. A blocked verdict that names nothing it waits on is demoted to idle by the hub. It is now recorded as "agent reported blocked without saying what it waits on" instead of as a complete with no reason. For a verdict signal with no reason the hub now says "agent reported complete without a pull request; the relay passed no reason along". It cannot know the agent gave none: an older relay discards the sentence. The newer relay's own "gave no reason" text is unchanged. TestIdleCompletionSurvivesRestart now writes the ledger file itself, so it fails if the expiry filter in loadIdleCompletions is removed. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π test(contribute): one comparison in the reasonless-blocked test, not two of the same go vet rejected `rec.Reason != const || rec.Reason != "literal"` as a suspect or. The test now pins the literal text once. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> --------- Signed-off-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
β¦has an open sub-issue (hivecommons#18046) * π fix(github): a merged PR does not mark done or close an issue that has open sub-issues Two paths wrote "done" on an issue because a PR naming it with a closing keyword merged, without looking at the issue's sub-issues: close-on-merge (the "Fixed by" comment and the close, or the awaiting-confirmation comment and hive/likely-done behind a confirmation gate) and the issue stage sweep's merged-PR close. A parent whose remaining work had moved into three open sub-issues was commented on and closed with all three open. Both now write nothing on an issue with an open sub-issue. Close-on-merge returns skipped/open_sub_issues, which settles the claim like its other skips; the stage sweep drops the merged claim and leaves the parent at its unclaimed stage, never stalled. The existing decomposed-parent sweep will close the parent once every sub-issue is closed. No new GitHub request: close-on-merge reads the sub_issues_summary block off the issue payload it already fetches (the single issue read, and the open-issue listing the sweeps work from), and the stage sweep uses the count on its enumeration row. An issue with no sub-issues, or with all of them closed, is handled exactly as before. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(github): a parent held for open sub-issues moves to 010 as progress, not a regression The stage sweep now leaves a parent with an open sub-issue open when the PR that claimed it with a closing keyword merges. Such a parent then falls from the PR's stage (040 or 030) to 010, and the store booked that as a regression; from 030 the sweep also posted its stage-regression note on the parent. Nothing was lost: the PR merged and the parent is waiting on its children. The sweep now tells Store.ObserveProgress that this fall is progress, the way a fall on an issue wearing hive/partially-addressed already is: no regression is booked, no note is posted, and a regression another writer booked for the same fall is cleared. An ordinary issue that falls from 030 to 010 without a merged claim is booked and told as before. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> --------- Signed-off-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
β¦ sub-issue open (hivecommons#18050) * π fix(github): three more automated paths leave an issue with an open sub-issue open An issue whose remaining work was moved into sub-issues is not done while one of them is open. Three automated paths still marked such an issue completed without looking at its sub-issues: - the already-landed close: the issues a PR closed as "already landed" names with a closing keyword were closed and given a done note; - question auto-close: an answered question was closed at the end of its window; - an agent's close request through the issue-request watcher. Each now leaves the issue open and writes nothing on it. The already-landed path skips the issue with one Info line. The question schedule ends with a new outcome, open_sub_issues, checked right where the close would happen, so everything that ended a schedule before still ends it the same way. The agent's request is refused before any write and parked as .denied with the reason in its result, the way a refused claim is; an override reason does not lift it. All three read GitHub's sub_issues_summary block off the issue payload. The first two take it from the issue read they already made (getIssueWithBlockedCount in place of Issues.Get), so they add no request and no new way to fail. The agent's request makes one extra issue read; a failure of it is handled like a failure of the close helper's own read. An issue with no sub-issues or with all of them closed, and a pull request, are handled as before. CloseIssue, close-on-merge and the issue stage sweep are not changed. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(github): say when a held parent is judged again, and test the refusal's edges No behaviour change. Words made exact, and tests added. Words: a question answer settled as open_sub_issues is not scheduled again only while its settled record exists (30 days). After the prune the next read of the issue schedules the same answer again with its deadline already past, and the tick after that settles it again while a sub-issue is open or ends the question as completed once they are all closed. The issue stage sweep takes over a held parent only with stage labels on (governor.labels.stage.enabled, the default); with them off the issue waits for a person. The agent-facing doc now says what to do about a parent that is itself a mistake or a duplicate. Tests: the question path after the settled record is pruned, both ways; an agent request for a person's bug awaiting reporter confirmation, a tracking issue and an already ended issue, each with an open sub-issue, is denied after one issue read with nothing written; an issue read that answers 404 quarantines the request at once. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> --------- Signed-off-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.29.0 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦.29.0 release: v6.29.0
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.29.1 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
β¦fixed instant (hivecommons#18055) TestReopenedClaimPersistsAcrossLoad saved a settled claim merged at a fixed date (2026-10-14T01:46:50Z) and read it back with LoadClaimLedger, which uses the wall clock. A settled claim is dropped at load once its 30-day retention has run out, so from 2026-11-13T01:46:50Z the load dropped the claim and the test failed on every run, on any branch. The reload now uses loadClaimLedgerAt with a clock fixed one hour after the fixture's reopen, the way the settled-claim tests already do. What the test checks is unchanged: a claim voided by a reopen is still stored, and still voided, after a reload. No production code changes. Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
β¦d, a refused note is logged, and a rate-limited park question is retried (hivecommons#18051) * π fix(park-recheck): the re-check note reaches a park the hive applied, a refused note is logged, and a rate-limited park question is retried A contributor relay parks an issue with a hub-minted token, so the park label's actor is the hive's App bot. On a person's issue without the review opt-in label the hive then said nothing: the park question was dropped by a GitHub rate limit, and the re-check note was refused by the write-back gate on every pass, silently at info level. - writeback: Item.HiveAppliedPark. For needs_decision_comment with that fact the gate skips the backfill guard and the opt-in label (reason hive-applied-park). A pause, the repo mode, project membership, own_only and the kind's minimum level still apply, and every answer that was an allow is unchanged. No other kind reads the fact and no other caller sets it. - park re-check: states the fact when the label event the park is dated from was made by one of the hive's own identities (the test the blocked sweep uses for a blocked label). A park a person applied is asked as before. The note still never un-parks. - park re-check: a refused note is logged once per issue and reason (info), repeats at debug, using the gate's hold memory; forgotten when the note is posted or the issue leaves the park. - contribute hub: a park question refused for a rate limit (primary, secondary, 429, or a pacer shed) is tried again after the reset or Retry-After, at most three more times, with an injectable clock and scheduler. A retry first checks the issue is still open and parked. Other errors are not retried. When the attempts run out the hub logs one warning. A pending retry is in memory and is lost on restart. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(park-recheck): a park is the hive's only when every label parking the issue is, and a note allowed for that reason lists only project changes Follow-up to the first commit on this branch, from review. - The hive's bot writes stage/045-needs-decision on any issue that carries a hard-suppress label, whoever applied it, and needs-spec is not a parking label, so a park a maintainer began with needs-spec was dated from the bot's 045 event and read as the hive's. HiveAppliedPark is now true only when the label event the park is dated from was made by a hive identity AND the newest labeled event of every hard-suppress label on the issue (built-in or configured) was too. A label with no labeled event in the timeline is not the hive's. This also covers a park label a person removed and re-added inside the relabel gap. - A note the gate allows only because of the fact (the same question without it is refused; Gate.Decide, which records no refusal) lists only the changes the park text names and the changes in the issue's own repository or another configured project repository. When nothing is left, nothing is written. A note that was already allowed (opt-in label, hive-filed issue, no gate) lists every change, as before. - The park question's retry counts an issue as still parked when it carries the configured decision label or stage/045-needs-decision, so a legacy label the stage reconciler replaced does not drop it. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> * π fix(park-recheck): a note allowed only because the hive applied the park claims no landing outside the project and quotes only a known closer Second follow-up from review. Both changes apply only to a note the gate allows only because of HiveAppliedPark; every other note is unchanged. - "The change this issue asks for has landed", the /fixed prompt and the landed marker are produced only for a named item in the issue's own repository or another configured project repository. A named item elsewhere is still listed as merged or closed; its patch is not read. The marker is read only by the "What to reply" notice to put /fixed first; the /fixed command itself does not read it. - The closing comment of the hive's adoption copy is quoted only when GitHub named who closed the copy. When the closer is not known the existing code quotes the latest comment by any person around the close; a fact-only note lists the copy as closed without author, quote or link, and drops a copy the park does not name. Tests added for both, and pins for: park text from a registered contributor naming an outside item, a hive comment in the park window repeating one, repo mode full with an issue older than the hive's write access, and a note posted before the opt-in label was removed. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com> --------- Signed-off-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Danathar <Danathar@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
Automated release commit. Compiles changelog.d/ fragments and moves the CHANGELOG.md Unreleased section into a dated v6.29.2 entry. See src/docs/releases.md. Signed-off-by: hive-release-bot <actions@github.com>
Brings tuna/v6 from upstream 6f6a5ca to the current upstream v6 head, on top of the earlier 46aa46a refresh. Upstream commits keep their original authors and sign-offs. Conflict resolution: - src/proxy/server.js, api_internal_header.test.js: take upstream. It now ships the fork's device-flow/hive_session allowance itself (exact-match PUBLIC_EXACT_POST_PATHS) plus the task/knowledge MCP paths. - src/pkg/ioscan/cli_client{,_test}.go: take upstream. It now ships the omp classifier backend, adding --no-context-files and a mandatory --model. - src/cmd/hive/boot.go, run_loop_deps.go: take upstream's sweep workers (per-sweep last-run fields removed). The fork's revert sheriff now runs in the red_main worker straight after the red-main pass (sweep_workers.go), which keeps its "after red-main in the same tick" order. - src/pkg/config/config.go: keep the fork's RevertSheriff next to upstream's new SharedCIIssues. Upstream now ships the Matrix / Code activity work (hivecommons#15945..hivecommons#16046, hivecommons#16239, hivecommons#16137 series) that #172 had cherry-picked, so those ports are superseded by this merge. Signed-off-by: James Reilly <jreilly1821@gmail.com>
Conflicts: src/pkg/github/open_pr_links{,_test}.go carry upstream 977d7ff
on both sides; keep tuna/v6's superset (977d7ff + #175 rate-limit backoff)
plus upstream's has-open-pr label wording. src/docs/operator-reference.md:
take upstream's split index and carry #175's rate-limit paragraph into
operator-reference/sweeps-and-operations.md where upstream moved it.
Signed-off-by: James Reilly <jreilly1821@gmail.com>
β¦iteback gate Upstream's writeback containment ratchet (pkg/writeback TestDirectWriteRatchet) refuses ungated direct writes. The fork's revert sheriff now consults the shared ci_rerun writeback policy before rerunning a failing job and again before opening a held revert, so operator pauses and per-repo write restrictions apply to it too. Ported from #172 (71ce0ef) without its Matrix kiosk test edits, which the upstream refresh supersedes. Signed-off-by: James Reilly <jreilly1821@gmail.com>
Upstream's notification write-site inventory (TestNotificationWriteSitesInventory) requires every notification-capable write to carry a class. The tuna-os revert sheriff's note on the culprit PR asks its author to look, so it is attention; listed in src/docs/notifications.md too. Signed-off-by: James Reilly <jreilly1821@gmail.com>
|
Decision (for @hanthor): keep open, unmerged. The code is ready; the one blocker is policy. Verified locally on head
Blocker: |
3e4b793
βΉοΈ Sentinel notice β informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @hanthor, head
Hive did not add the |
|
Superseded by #195, which merges current upstream |
Upstream refresh split out of #172 (part a). It is meant to land as a merge commit, never a squash, so every upstream commit keeps its original author and Signed-off-by.
Contents (first-parent)
46aa46a9f: hanthor's existing merge of upstream v6 through6f6a5ca63, unchanged from β¨ Refresh v6 pilot and enable upstream ambient code-rain testingΒ #172 (fork patches kept).4feb08923: merge of upstreamhivecommons/hivev6 throughdea69b350(v6.29.2). Conflict resolutions are listed in the commit message: proxy and ioscan take upstream, which now ships the fork's device-flow and omp-classifier patches itself; boot/run-loop take upstream's sweep workers, with the fork revert sheriff run in thered_mainworker; config keepsRevertSheriffnext toSharedCIIssues.3a66bdcfa: merge of currenttuna/v6(π fix(claims): open-PR guard backs off until rate-limit reset and refuses claims meanwhileΒ #175, π± ci: build stamped fork pilot binaries on tuna/v6Β #178). This keeps π fix(claims): open-PR guard backs off until rate-limit reset and refuses claims meanwhileΒ #175's rate-limit backoff on top of upstream 977d7ff and moves its doc paragraph into upstream's split operator reference.6a8c5c20a+3e4b79328: the only fork commits needed for this merge to stay green. Upstream's writeback ratchet rejects the fork sheriff's ungated writes, so the sheriff now goes through theci_rerunwriteback gate (ported from β¨ Refresh v6 pilot and enable upstream ambient code-rain testingΒ #17271ce0ef56), and its culprit note is classifiedattentionin upstream's notification write-site inventory.After this lands,
tuna/v6differs from upstream v6 by just 43 files: the fork patches (proposal gate, freeze windows, spoke proof tokens, revert sheriff, broker access), #175 and the pilot CI from #178. The Matrix / Code activity ports that #172 cherry-picked (hivecommons#15945βhivecommons#16046, hivecommons#16239) now come from upstream, with original attribution.Verification (local; GitHub Actions is blocked on tuna-os/hive GitHub-side, so no CI can report)
go build ./...andgo vet ./...: clean.gofmt -lon the hand-resolved Go files: clean.go testof./cmd/hive,./pkg/{config,dashboard,github,governor,hub,intent,ioscan,redmain,revertsheriff,writeback}/...,crux/agent/...: see the PR comment for the result on the final head.src/proxyandsrc/pkg/ioscanare byte-identical to upstream v6.Merge blocker: sign-off check has no fork-sync exemption
Sign-off survives the squash(check-squash-signoff-attribution.sh) exempts only:LANDED_REFS=origin/v4);merge-method/merge(the label is set here);This PR brings in about 1,450 non-merge upstream commits. About 717 of them are by human upstream contributors (andy@clubanderson.com, doug.baggett, Karibhou, thatguywob, β¦), each signed off as themselves. The checker reads those as "foreign sign-off" for a PR authored by hanthor, even though a merge commit lands them unchanged. Making the check pass would mean rewriting other people's sign-offs, which is wrong. So this PR stays open until a maintainer picks an exemption policy. One option: add the upstream protected line (
hivecommons/hivev6) toLANDED_REFSformerge-method/mergePRs into fork lines. Those commits already passed this gate upstream.Supersedes the upstream half of #172.