Repository navigation
🌱 chore: forward-merge v5 -> v6 (top-up through 7a506f71b, 43 commits, hand-resolved) - #14724
Conversation
…t step (#14595) * ✨ feat: mark lanes unauthorized when the App lacks rights for the next step A GitHub App API call the contributor lane needs to advance that fails with 403 Resource not accessible by integration now routes through markUnauthorized: label hive:unauthorized, one comment per (head, step) naming the missing permission and the human step, reason unauthorized:<step> parking the PR at 085, one INFO log per head, and clearUnauthorized in the tick the step is no longer needed. First call site: held-run approval without actions: write. Dashboard draws a muted unauthorized pill and reads Awaiting human: approve workflow runs. Fixes #14589 Signed-off-by: Andy Anderson <andy@clubanderson.com> * test: clear tick lets the lane post its own note Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
…e default > built-in (#14582) * ✨ feat(scheduler): house-style section every kick opens with (repo HIVE_STYLE.md > hive default > built-in) Every kick on every backend now begins with a HOUSE STYLE section: how changes are to be made, as distinct from what to work on. A managed repo's root HIVE_STYLE.md (read from project.checkouts_dir, like AGENTS.md) overrides project.style / project.style_file, which override the built-in text. Candidates are capped (project.style_max_bytes, 4096) and skipped whole when oversize; the repo file is untrusted and is refused whole on an ioscan block at the issue-body policy, with a house_style_skip audit line. The agent card and detail pane show which source applied, read back from the kick header (/api/status agents[].styleSource). Fixes #14575 Signed-off-by: Andrew Anderson <andy@clubanderson.com> * 🌱 settings parity: allowlist project.style_max_bytes until a Settings control exists Signed-off-by: Andrew Anderson <andy@clubanderson.com> * chore(changelog): move house-style fragment to repo-root changelog.d Signed-off-by: Andrew Anderson <andy@clubanderson.com> * docs(api-reference): refresh line citations after rebase Signed-off-by: Andrew Anderson <andy@clubanderson.com> * changelog: fragment for #14582 Generated by changelog-fragment-guard from the PR title (#5675). Edit the wording freely, or replace this file with the no-changelog label if the change is not user-visible. Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Signed-off-by: clubanderson <407614+clubanderson@users.noreply.github.com> * style(dashboard): use --line-width token for style-source chip border Signed-off-by: Andrew Anderson <andy@clubanderson.com> * fix(ci): allowlist pkg/housestyle for dashboard import ratchet; avoid scope-guard literal in comment pkg/dashboard → pkg/housestyle is a one-way read of the style source already recorded on the kick message (no new seam exists for kick metadata); the login-scope guard matches the quoted word in a struct comment, so reword. Signed-off-by: Andrew Anderson <andy@clubanderson.com> * chore(changelog): single bullet fragment for house style (drop duplicate) Signed-off-by: Andrew Anderson <andy@clubanderson.com> --------- Signed-off-by: Andrew Anderson <andy@clubanderson.com> Signed-off-by: clubanderson <407614+clubanderson@users.noreply.github.com> Co-authored-by: clubanderson <407614+clubanderson@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…ernor grid (#14597) fleetRunStripHtml landed in .oc-gov-strip, which is display:none, so the fleet figure from #14580 was never on screen. Render the same figure as a gov-stat in the Governor section and pin it with a webstatic test. Signed-off-by: Andrew Anderson <andy@clubanderson.com> Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
…rch (#14615) Add a gh-aw agentic workflow that researches projects adjacent to Hive every Monday at 13:00 UTC (and on dispatch), writes a 600-900 word plain-language report plus a SWOT for Hive, appends it to tracking issue #14613, and posts it to Discord through the DISCORD_RELATED_RESEARCH_WEBHOOK secret via a custom safe-output job. Compiled with the pinned gh-aw v0.89.21; no repository write permission; commits nothing. Fixes #14612 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…io hive domain (#14610) * 🐛 fix(contribute): stop pointing clankers at the retired kubestellar.io hive domain A contributor.env written by contribute-setup before the 2026-09-04 apex move still carries wss://hosted-<slug>.hive.kubestellar.io/contribute. The relay dialed it as-is, failed the TLS handshake (the old wildcard is no longer in the served certificate) and reconnected every 16 s forever with no hint that the hive had moved. - contributor-relay.js / contributor-agent.sh / Justfile: relocate a saved *.hive.kubestellar.io URL to *.hive.hivecommons.dev in memory, label intact, warning once and naming the contributor.env line to fix. - contributor-relay.js: on ERR_TLS_CERT_ALTNAME_INVALID park the hub (no reconnect timer), print one remediation line, and for a hosted hive ask the hub /api/registry for the hive's current dashboardUrl and retry there exactly once. relay-status.json reports readiness: hive-moved. - hub: guard test that deploy/k8s/error-pages.yaml never names the retired apex (the live ingress-nginx default backend is a stale pre-fix copy). - docs: troubleshooting entry for the altname-mismatch loop. Fixes #14609 Signed-off-by: Andy Anderson <andy@clubanderson.com> * fix(contribute): compose the retired hub host so repo-wide guards stay green TestHubStaticReferencesCurrentHost and TestContributeJustfileReachesOnlyTheCanonicalHubHost reject any literal spelling of the retired hub host in pkg/hub and Justfile code lines. Derive the host from the existing legacy_hive_hub sentinel in the Justfile and compose the apex in the error-pages guard test instead. Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
…14601) fetchPRsWithDetails built the held-PR record without BodyDigest, unlike the actionable record and the webhook patch. Held PRs, which the review swarm reviews, were therefore kicked without a "Review body digest:" line, dispatched with an empty digest, and skipped by the relay's dispatch-digest stamping, so each head's verdict landed in its own aggregate and the pending entries were never confirmed. Set the digest on the held record, and as defence in depth have the relay stamp a dispatched verdict with the dispatch record's digest even when that digest is empty, so all heads share one aggregate key. Closes #14594 Hive-Run: #14594 Hive-Plan: issue-14594 proposed fix steps 1-2 Hive-Spec: review-body-digest#held-pr-digest Claude-Session: https://claude.ai/code/session_01HRH6knAHzP9gAxFXPypLDi Signed-off-by: Karibhou <Karibhou@users.noreply.github.com> Co-authored-by: Karibhou <Karibhou@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…14642) The activity monitor drew the modifier chip and the since age in the title line's ghost strip (align-items: flex-start), so a wrapped title or an issue→PR row floated them to the top edge of the entry, where they read as the previous row's. The since text now closes the strip row, in line with the chevrons; the chip bottom-aligns in the title line so it sits directly over its own strip; entries get a hairline above and a whole-row hover tint. Fixes #14640 Signed-off-by: Andy Anderson <andy@clubanderson.com>
…udget (#14629) * 🐛 fix(ioscan): unavailable classifier must not consume the per-eval budget An ioscan.UnavailableClassifier never makes a model call, yet enforceTextFrom consumed the per-eval budget before calling Score. Under fail-closed the first 64 strings per governor eval were withheld and every later one passed on as budget_exhausted, so kick survival depended on render order: on the console hive the reviewer (rendered last) got one kick per pod lifetime for three days while earlier agents were withheld. Route the unavailable classifier straight to the fail-mode handler so fail-closed withholds every string, and log the fail-open error once per distinct error instead of once per string (64 WARNs per eval auto-filed selfreport #14620). Fixes #14627 Signed-off-by: Andy Anderson <andy@clubanderson.com> * changelog: fragment for #14629 Generated by changelog-fragment-guard from the PR title (#5675). Edit the wording freely, or replace this file with the no-changelog label if the change is not user-visible. Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Signed-off-by: clubanderson <407614+clubanderson@users.noreply.github.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com> Signed-off-by: clubanderson <407614+clubanderson@users.noreply.github.com> Co-authored-by: clubanderson <407614+clubanderson@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* ✨ review: route bot-authored PR reviews to a cheaper model PRs opened by Dependabot, Renovate, github-actions and other GitHub Apps are reviewed with review.bot_author_model when set, through the same per-kick model choice the review_models pool uses. The hive's own App login never matches review.bot_authors; a PR touching review.bot_author_full_review_paths (default CI workflows and Dockerfiles) or whose changed files could not be listed keeps the full-strength model. Fixes #14625 Signed-off-by: Andrew Anderson <andy@clubanderson.com> * test: allowlist review.bot_author_* in settings parity until a Settings control exists Signed-off-by: Andrew Anderson <andy@clubanderson.com> --------- Signed-off-by: Andrew Anderson <andy@clubanderson.com>
… before conflicts (#14624) 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: clubanderson <407614+clubanderson@users.noreply.github.com>
…, taken off hold (#14617) * ✨ RFC lifecycle notices to Discord: opened, closed, and taken off hold (#14616) Add pkg/rfcnotify, a transition detector over polled issue snapshots that posts one Discord line when an RFC issue (rfc label or RFC-prefixed title) is opened, closed (completed or not planned), or has its hold lifted or moves from held to stage/010-actionable. Per-issue records persist in /data/rfc-notify.json so restarts and the slack re-read never re-post. - notifications.discord.rfc_channel_id (env HIVE_DISCORD_RFC_CHANNEL_ID), default empty = off; posts with the existing bot token through a new discord.Bot.SendMessageTo without starting the command bot. - github.Client.ListIssueSnapshots lists issues (not PRs) across managed repos by state and since, the raw input the actionable enumeration cannot provide for closed or held issues. - Wired into the eval cycle's optional sweeps next to the RFC existing-coverage check; each post is audited as rfc-notify. Signed-off-by: Andrew Anderson <andy@clubanderson.com> * Add Settings control for notifications.discord.rfc_channel_id The Settings parity test requires every hive.yaml leaf to have a Settings dialog control. Expose the RFC channel id in the Discord bot subgroup, round-trip it through the notifications settings handler, and cover it in the handler test. Signed-off-by: Andrew Anderson <andy@clubanderson.com> --------- Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…lot-requests (#14653) The org-billed copilot-requests: write path returns 403 from the Copilot model catalog for hivecommons (Copilot Business, no seats), so the agent never starts. Drop the permission and recompile so the lock file validates and passes the COPILOT_GITHUB_TOKEN fine-grained PAT secret instead. Fixes #14652 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Avoid treating fresh downstream jobs in older workflows as runner-pool outages. Exercise the deployed watchdog against mocked GitHub job timestamps in CI. Hive-Run: #14567 Hive-Plan: queued-job-age Hive-Spec: ci-runner-canary#queue-stall Signed-off-by: thatguywob <thatguywob@users.noreply.github.com> Co-authored-by: thatguywob <thatguywob@users.noreply.github.com>
…ed (#14648) * 🐛 fix(stage): clear needs-triage once an issue is actionable or claimed A hive-filed self-report wears needs-triage from filing (fleetreport); 015 is a tracking state, not a gate, so the scanner claims it, and the stage reconciler never touched the non-owned label — leaving claimed and closed issues reading "Triage needed" on the activity monitor (#14586). - stagelabels: Desired.Clear names labels Plan removes alongside the stage; Stage.SettlesTriage() is true from 020 on (and PR-mirrored stages), false at 010/015/017/045 which may fall back to 015. - stageengine: IssueFacts.TriageAwaitingLabel; Decision.Clear set when the derived stage settles triage, so a direct 015→020 claim removes needs-triage in the same label write. - github: sweep reconciles open issues already at 020+ still wearing it. - dashboard: stage text comes from the stage label only; a stale needs-triage on 020+ is dropped from modifiers and reported as stale_labels with a muted tooltip line. Fixes #14644 Signed-off-by: Andy Anderson <andy@clubanderson.com> * test: compare Desired with DeepEqual; load stageTrackerStaleLabelsText into node fixtures Desired now carries a []string (Clear), so the struct is no longer comparable with !=; and the Go-driven node fixtures that evaluate stageTrackerTooltip must extract the new helper it calls. Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
…ead (#14622) A re-review of a PR the hive already judged at another head SHA now carries the previous verdict and findings and asks the reviewer to read only the compare between the two heads. An empty or trivial delta (under review.delta_trivial_lines, touching no previously flagged file) re-stamps the previous verdict onto the new head without a model call. Full review stays for first reviews, prior reject/incomplete, and stale changes_requested findings older than review.delta_max_age. Aggregates record delta.previous_head and delta.trivial so the cost dashboard can split delta reviews from full ones. The new review.delta_* keys are allowlisted in settings parity until a Settings control exists. Fixes #14621 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
) * ✨ feat(sandbox): scheduler_name for the Job runtime pod template Accelerator clusters commonly gate device-requesting pods behind a custom scheduler via a validating admission webhook. On the Spyre dev cluster the Job create is denied outright: admission webhook "spyre-job-validator.k8s.io" denied the request: A pod must use "schedulerName: spyre-scheduler". SandboxJobConfig had no way to set it, so every runtime: job kick that requests ibm.com/spyre_pf failed before a pod existed. Add agents.<name>.sandbox.job.scheduler_name, passed through verbatim as the pod's schedulerName (empty = cluster default), with launcher and config round-trip tests and a doc line. Signed-off-by: Andrew Anderson <andy@clubanderson.com> * feat(dashboard): expose agent_sandbox.job.scheduler_name in Settings Settings parity requires every hive.yaml leaf to be reachable from the Settings dialog; add the Job scheduler name field next to the other sandbox Job fields and allowlist the per-agent override like its siblings. Signed-off-by: Andrew Anderson <andy@clubanderson.com> --------- Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…14676) * 🐛 review-fix: dispatch fix kicks when reports carry no body digest A report without body_digest was read by bodyEditedVerdictRevisitable as "PR body edited since verdict", so the planner re-requested review instead of dispatching a fix, every cycle, forever. Missing digest now means unknown/unchanged; only a present-and-different digest triggers revisit. Writers now stamp body_digest: ModelStamps.apply fills it from dispatch records at Collect time, and the relay stamps advisory verdicts from the PR snapshot via review.PullRequestBodyDigest. Starvation guard: a changes_requested head suppressed by the revisit path for RevisitStarvationCycles (3) consecutive cycles is forced onto the fix path; each suppressed cycle records a FixExclusion and logs a WARN, and the planner summary line exposes fix_excluded / fix_excluded_reasons. Fixes #14665 Signed-off-by: Andy Anderson <andy@clubanderson.com> * 🐛 review-fix: qualify repo in starvation-guard test lookups PlanDispatch org-qualifies pr.Repo before recording state; the guard tests looked the counter up with the bare repo and read 0. Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
…action list (#14675) - bash allowlist uses cmd:* entries so curl | jq pipelines are permitted - replace DuckDuckGo (202 bot challenge) with Google News and Bing News RSS; HN Algolia query encodes > as %3E - prompt grounded in what Hive and Crux ship (runtime, engine launchers, ACMM governor, proxy, reviewer swarm, evidence cards, dashboard, hub, CLI) - Who does what table: Project | Who runs it | Category | Relationship to Hive | Licence (SPDX id from the GitHub API) - wrap URLs in <...> in the Discord step so no link cards expand - SWOT: each bullet must be something Hive controls or a dated external fact; filler such as release cadence is banned - new final section: Top 5 this week - stay ahead and stay integrated - 1,200-word cap; open_issues dispatch input (default off) to file Top 5 items as help wanted + competitor-watch issues Fixes #14673 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
* merge: owner approval releases the workflow-path fence A contributor-lane PR editing .github/workflows/** or .github/actions/** was never merged by the hive: an owner's approval only moved it to stage/090 with a "click Merge" caption (hive#14568, hive#14498, rationguard#401). The App has workflows: write, so the fence is now released by explicit owner approval on the current head: a trusted merger's APPROVED review on that commit, or a dashboard / typed `/hive decision approve` bound to it. The PR then goes through every other gate and the lane merges it, logging `workflow-path fence released by owner approval` and leaving one hive:workflow-approval marker per head. A merge refused for want of the workflows permission marks the PR hive:unauthorized (merge-workflow-path) and returns the workflow-path reason so the approved head parks at 090 as before, with no retry loop. Toggle auto_merge.contributor_lane.owner_approval_releases_workflow_fence (default on) wired through config, merge_eligibility, the dashboard auto-merge API and Settings UI. The 085 note for an unapproved workflow-path PR now reads "awaiting owner approval"; dashboard approval comments carry a head marker so later readers can bind them. Fixes #14664 Signed-off-by: Andy Anderson <andy@clubanderson.com> * merge: drop unused manualMergeRuleReason wrapper (lint) Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
* dashboard: say why a PR is waiting on a human merge Every stage/090 row's merge-gate word, tooltip, headline and nested sub-row now carry a compact reason code (prstage.HumanMergeReason*) derived from the lane reason already on the stage record — workflow files, tests removed, manual-merge policy, branch protection, merge method, auto-merge off, untrusted author, unauthorized, reviewer deferred, falling back to policy with the raw lane reason quoted. Fixes #14658 Signed-off-by: Andy Anderson <andy@clubanderson.com> * dashboard: 090 tooltip leads with the lane word, then the reason Signed-off-by: Andy Anderson <andy@clubanderson.com> * dashboard: nested and single-covering-PR rows carry the human-merge reason Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
Post each chunk with ?wait=true, log 'Posted chunk N/M id=<id>', write the ids to the step summary and the message_ids step output, and append them to the tracking-issue comment (post_to_discord now needs safe_outputs so the add_comment id is available; falls back to a short note on #14613). Refs #14673 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…14659) 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>
On spokes where HIVE_DASHBOARD_TOKEN (or HIVE_TTYD_CREDENTIAL) is set, deploy/entrypoint.sh starts ttyd with -c hive:<token>. The Go /terminal reverse proxy forwarded requests without an Authorization header, so after the dashboard's authenticate() admitted the browser, ttyd re-challenged it with a native Basic-auth dialog. Hosted spokes run ttyd uncredentialed and never saw it. The proxy now resolves ttyd's credential once at construction with the entrypoint's precedence (HIVE_TTYD_CREDENTIAL, else hive:<token>), sets it with SetBasicAuth on every forwarded request (assets, /token, websocket upgrade), and drops any browser-supplied Authorization instead of relaying it. The Node proxy's /terminal route had the same gap and is fixed the same way; its keepalive regression test is adapted to the injected credential since the proxy no longer relays a browser basic-auth dialog. Fixes #14693 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…14689) Within a column the Hive activity monitor now tiers rows before the existing keys: rows a person must act on now (a shown pill or a human-gated stage) first, plain rows next, hold / draft / closed-issue-anchored rows last; within a tier the longest on the step still comes first, then repo and number. Applies to the Furthest along and Longest waiting sorts and to the PR sub-rows under an issue; the direction arrow flips the age keys but never the tier. The server pre-sort mirrors the tier (stageTrackerAttentionTier in Go and JS) so chat /prs and the hub read the same order. Fixes #14686 Signed-off-by: Andy Anderson <andy@clubanderson.com>
…e fix (#14684) * fix(decision): move an answered PR out of needs-decision and dispatch the fix A trusted merger's /hive decision on a hive-authored PR at stage/085-needs-decision was acknowledged but never acted on: - parseUnparkCommand kept only the first line, so the recorded instruction (and the fix kick) read "approve" instead of the numbered instructions under it; - ClassifyStage answered the swarm's decision-requested verdict before the lane's human-changes-requested reason, so the owner's decision on the current head never outranked the question it answered and the PR stayed at 085; - a bare /hive decision approve was recorded as a change request too. Now the whole comment is the decision text; the recorded requester is noted on the stage detail and folded through ReviewFacts, and a decision on the current head derives fixing (owned rank, human-changes-requested) ahead of RequiresHumanStage. A bare approve is recorded with kind approve: HumanChangeRequestFor hides it from the planner and the lanes treat it as a trusted human approval that releases requires_human. The un-park path submits a PRDecided fact so 070 is written in the same pass as the reply, logs one 'decision: unpark pr=… head=… mode=… by=… → stage=…' line, tags the reply with the exact head, and on later sweeps rebuilds a missing or first-line-only record from the acknowledged comment thread for the live head only. Fixes #14679 Signed-off-by: Andy Anderson <andy@clubanderson.com> * fix(decision): put Mode on PullRequestDecisionEvent; route dashboard decisions through the same mode/stage path Signed-off-by: Andy Anderson <andy@clubanderson.com> --------- Signed-off-by: Andy Anderson <andy@clubanderson.com>
…14700) Every sandbox-opened PR was titled "[<agent>] sandbox changes" with the body "Sandboxed credential-free execution result ... Part of #2804" — a placeholder that also cross-referenced hive's own issue number in the target repo (torch-spyre/spyre-inference#1402 is a live example). The agent inside the Job cannot open the PR itself, so its description had no path to the PR. Resolve title/body in order: .hive/pr-title.txt (first line) and .hive/pr-body.md written by the agent, then the HEAD commit subject/body, then a neutral placeholder. Bounded to 200 / 60000 runes. Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…14699) /api/stage-tracker held stageTrackerCacheMu across the whole tracker build (one review-evidence lookup per open PR, each a Glob of the repo evidence directory on NFS: ~40s in the owner's hive). Every webhook delivery and every status publish invalidated that cache through the same mutex, so webhook handlers (~177/min), the governor's publish and every dashboard poll queued behind back-to-back rebuilds that never drained: blocked handler goroutines, connections, threads and heap grew without bound. - invalidation flips a stale flag under the cache mutex and never waits - at most one rebuild runs, outside the mutex; polls during it get the previous body as source=stale, max-age=0 - a waiter with no body honours its request context (503 on disconnect) - the evidence-directory sibling listing is memoised per repo dir (60s) instead of re-read per PR per build - regression tests: invalidation returns during a held build, N polls during a build cost one build and leave no goroutines behind, first waiters join the one build; sibling memo semantics Fixes #14681 Signed-off-by: Andy Anderson <andy@clubanderson.com>
…14697) Copilot CLI has a URL approval step separate from the shell tool allowlist and the AWF firewall. Run 40572955859 had shell(curl:*) and shell(jq:*) but every curl to a URL was still denied with 'no one to approve', so the report was partial (no HN, Google News, or licence lookups). Reproduced locally with copilot 1.0.103: --allow-url <host> fixes it. engine.args now passes --allow-url for every host in network.allowed; the firewall still enforces that list. Refs #14673 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
* ✨ review: wire review.pending_fix_ttl into the fix dispatcher DispatchOptions.PendingFixTTL was only settable in code, so operators always got the 2h default even though labels-and-control-signals.md presented PendingFixTTL as a tunable. Add review.pending_fix_ttl (Go duration; unset or 0 means 2h, negative disables expiry), validate it at config load, pass it to the planner, and document it next to review.pending_ttl. Closes #14643 Hive-Run: #14643 Hive-Plan: review-backlog/pending-fix-ttl-config Hive-Spec: review-dispatch#pending-fix-ttl Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HRH6kmKQdyGqpPWbCsKpUX Signed-off-by: Karibhou <Karibhou@users.noreply.github.com> * fix(dashboard): allowlist review.pending_fix_ttl until the Settings dialog gains a control The settings-parity test fails closed on any new hive.yaml leaf the dialog does not reference; record the gap against #14643 like the sibling review keys. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> --------- Signed-off-by: Karibhou <Karibhou@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Karibhou <Karibhou@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…osed (#14657) A PR whose every closing link names a closed issue is still live work. Its row is now the PR's own: closed_issue on the item carries the demoted issue (with GitHub's stateReason so a not-planned close can say so) and the tracker draws the PR's number and title in the lead cell with the closed issue as a muted chip. The PR's stage, modifiers and timestamps are untouched — the issue closing is not a regression. Fixes #14655 Signed-off-by: Andy Anderson <andy@clubanderson.com>
|
The failing required check 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 |
cf026b7
ℹ️ Sentinel notice — informational; author is trusted (hive agent), merge is not blockedHive flagged this PR (author @hivecommons-hive[bot], head
Hive did not add the |
…ter forward-merge Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
|
hive/awaiting-ci: pushed fixup for AdoptionPRBody test call sites and the src/docs/README.md anchor; govulncheck is shared incident #14726. CI pending — sweep will merge when green. 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 |
38d0a8c
ℹ️ Sentinel notice — informational; author is trusted (hive agent), merge is not blockedHive flagged this PR (author @hivecommons-hive[bot], head
Hive did not add the |
There was a problem hiding this comment.
This top-up is already four v5 commits stale, and one of them is almost certainly the reason govulncheck (known vulnerabilities) is red here.
Verified at head 38d0a8c
git log pr-head..origin/v5lists four commits merged to v5 after the snapshot this PR took (7a506f71b):648ca07cafix(deps): bump golang.org/x/net to v0.62.0 to clear GO-2026-6860 (#14731),f5cc08b48(#14706),0c1061d9a(#14618),4d8bd7312(#14737).src/go.mod:78on this head still pinsgolang.org/x/net v0.61.0;origin/v5src/go.mod:70hasv0.62.0. The only red check is govulncheck, whose job log I could not retrieve, so I am stating the link as the likely cause rather than a confirmed one.- Merge completeness otherwise looks right:
git diff origin/v5 pr-headshows only v6-only paths (src/test/runs_e2e_v6_test.goadded,systemd/hive-discord.serviceremoved), i.e. nothing from the v5 side was dropped in the 25-file conflict resolution.
Suggestion: re-top-up sync/v5-to-v6 through 4d8bd7312 (which carries the x/net bump) rather than patching go.mod by hand on this branch; otherwise the next forward-merge conflicts on go.mod/go.sum again and this PR merges a v0.61.0 pin that v5 has already moved off.
I have not reviewed the 43 constituent commits line by line; they landed on v5 through their own reviews. Not approving or requesting changes.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
I did not review the 292-file merge body itself; a 25-file hand-resolution of 43 commits needs a human pass, and the body says no local build or tests were run. What I did verify, so whoever is fixing this does not chase the wrong red check:
govulncheckis red on thev6base head too, not because of this merge. Base9da5696(job 122512605239) reports the sameGO-2026-6860ingolang.org/x/net@v0.61.0(fixed inv0.62.0). The fix belongs in a separatego.modbump onv6, not in this PR.- The v2-ci coverage gate passed on
38d0a8c, so the per-package floor risk the body calls out did not materialise this time. - Probot
DCOshowsaction_requiredwithdco-signoff: yesset and the prowdcocheck green — the usual non-blocking state for App-authored PRs. - "Sign-off survives the squash" is
cancelled, not failed; it was superseded by a later run rather than rejecting the sign-off.
@clubanderson: the conflict resolutions in config.go (v5 fields re-homed into the v6 split files) and main.go (both dashboard deps sets kept) are the places where "keep both sides" can silently drop a wiring line; those two files are where a human read is worth the time.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
….62.0 Resolve api-reference.md citation conflict via check-api-reference-citations.sh --fix and upgrade golang.org/x/net to clear govulncheck GO-2026-6860. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
|
hive/awaiting-ci: pushed the regenerated NOTICE (x/net v0.62.0 entry, taken verbatim from the notice-generated CI artifact). CI pending — sweep will merge when green. 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 |
056db68
ℹ️ Sentinel notice — informational; author is trusted (hive agent), merge is not blockedHive flagged this PR (author @hivecommons-hive[bot], head
Hive did not add the |
There was a problem hiding this comment.
Looks correct to me. I checked the merge itself rather than re-reading 295 files; CI (compile, all test shards, coverage gate, golangci-lint) is green at 056db68.
What I verified at head 056db68:
- Ancestry:
compare/056db68...7a506f71b→ahead_by: 0— every v5 commit through the stated top-up point is in this branch.compare/v6...056db68→behind_by: 0, so it fast-forwards ontov6. (v5has since gained 15 more commits — expected; the next top-up picks those up.) - No conflict markers:
git grep -E '^(<<<<<<<|=======|>>>>>>>)$'across the tree is empty. - Hand-resolved ports exist where the body says, with one small correction:
src/pkg/config/notifications_config.go:100,115—DiscordRFCChannelIDEnvVar,RFCChannelID; env override atsrc/pkg/config/load.go:421-425.src/pkg/config/project_config.go:65—StyleMaxBytes;src/pkg/config/review_config.go:272,724—PendingFixTTL,OwnerApprovalReleasesWorkflowFence.ExpandAgentReplicasis insrc/pkg/config/reviewer_autoscale.go(notconfig.goas the table says) and is called fromsrc/cmd/hive/main.go:4293andsrc/cmd/hive/reviewer_autoscale.go:188. Just a table nit, the code is right.src/pkg/discord/bot.go:172—Bot.SendMessageTodelegates todiscordBackend.sendToat:179, which keeps v6'slogscrub+ emptyallowed_mentionspayload as described.src/pkg/review/dispatch.go:1455,1496,1510— both v6PlacedByVerdictand v5pendingFixExpiredare present and wired.src/cmd/hive/main.go:2065,8003-8004—OpenLedgerScoped,planReviewDispatchWith+emitReviewHumanEscalationsas described.src/proxy/package.json:10— test script runs bothttyd_credential_inject.test.jsandoutput_log.test.js.
No findings.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
Looks correct to me as a forward-merge; what I verified at head 056db68:
compare/7a506f71b...056db68→behind_by: 0, so the head contains every v5 commit through the stated top-up point.compare/v6...056db68→ 47 commits ahead, 0 behind (43 v5 commits + the merge/resolution commits).- The config split the body describes is real:
ExpandAgentReplicasis insrc/pkg/config/config.go:272and called fromload.go:112,243;StyleMaxBytesatproject_config.go:65;RFCChannelID/DiscordRFCChannelIDEnvVaratnotifications_config.go:100,115with the env override atload.go:421-425;PendingFixTTL,BotAuthors,DeltaRereview,DeltaTrivialLinesatreview_config.go:272-312. Nothing from the v5 side was dropped in the files I sampled. - All test shards, coverage, lint, gosec and govulncheck are green. The only red checks are
DCO/dco, which fail on every forward-merge that carries other people's commits (same as #7199 and later top-ups) and are non-blocking for an admin merge.
One note for whoever merges: v5 has moved 15 commits past 7a506f71b since this branch was cut, so a follow-up top-up will be needed after this lands. That does not affect this PR.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Thank you for your contribution! Your PR has been merged. We'd love to hear how your experience was: share feedback |
Why
The v6 Top-up workflow on the push to
v5(run 40591617022) could not auto-merge:origin/v6..origin/v5(43 commits, through 7a506f7) conflicts in 25 files, so it needs a person to resolve it (precedent: #7199). This PR is that merge, made withgit checkout -B sync/v5-to-v6 origin/v6 && git merge --no-ff --signoff origin/v5. Every conflict is resolved to keep both sides: all v6-only changes and all v5 changes.Hand-resolved files
src/pkg/config/config.goconfig.gois kept. v5's config.go changes are ported into the split files:ExpandAgentReplicasautoscaled replicas (config.go),Style/StyleFile/StyleMaxBytes(project_config.go),RFCChannelID+DiscordRFCChannelIDEnvVar(notifications_config.go, env override inload.go),PendingFixTTL,BotAuthors*,Delta*,Reviewers(review_config.goReviewConfig),OwnerApprovalReleasesWorkflowFence+Effective…(review_config.go).src/cmd/hive/main.goReviewerAutoscaleStatus). The mutation ledger uses v6'smutationStateDir()together with v5'sOpenLedgerScoped(... MutationMaxWritersPerTarget(), TargetWriterScope). The eval cycle uses v5'splanReviewDispatchWith+ reviewer autoscaler together with v6'semitReviewHumanEscalations.src/pkg/convergence/mutation/boundary.gosrc/pkg/discord/bot.godiscordBackendrefactor. v5'sBot.SendMessageTo(channel, content)is added on top of a sharedsendTothat keeps v6's scrub/no-mentions payload.src/pkg/github/automerge/automerge_stage_human.godecisionActionswithstagelabels.DecisionActions, so the v6 route is kept. v5's workflow-fence logic (workflowFenceManual,manualMergeRuleReasonFor,decisionCauseFor) merged cleanly and feedsManualMergeRule/CauseReason.src/pkg/github/automerge/pr_approve_command_test.goTestDecisionCommentNamesLaneActionalong withdecisionActions, so v5's one added case there has no target.src/pkg/github/automerge/automerge_sweep.gohgithub.ClassifyBranchUpdateErrordelegation. v5's TODO(#14589) comment moved intobranch_update_refusal.go.src/pkg/review/dispatch.goaddHuman/PlacedByVerdict/humanReason+ v5pendingFixExpired.src/pkg/dashboard/stage_tracker.goHumanApprover/ManualMerge/HumanMergeReason(Text)and v6'sFinisher*fields and helpers.src/pkg/dashboard/deps.go,pr_unpark_decision.go,main.goimportssrc/pkg/stageengine/derive_pr_classify.gounauthorized:→ needs-decision branch with v6's linked-issue hold/needs-decision branch.src/pkg/stageengine/prstage/reasons.go,reasons_test.goLinkedIssueReason/ParseLinkedIssueReason+ v5HumanMergeReason. Both tests are kept in one file.src/pkg/fleetreport/runtime.goagent_github_api_cap_reached,transcript_scan_slow) and v5's "failed open" ioscan keyword.src/pkg/github/client.gopr_batchstruct keeps both v6DependsOnand v5BodyDigest.src/pkg/dashboard/static/index.htmlsign-based direction also applies tofinisher, so reversing it still puts needs-a-person rows first. Linked-PR mapping carries both field sets. v6 had moved the Discord bot to the Channels tab, so v5's RFC channel id field went into the Notifications tab's Discord section, which posts through the mergeddiscordRfcChannelIdhandler. Both CSS additions are kept.src/proxy/package.jsonttyd_credential_inject.test.js(v5) andoutput_log.test.js(v6).bin/contributor-relay.test.jssrc/docs/api-reference.mdsrc/scripts/check-api-reference-citations.sh --fix(16 rewritten, 0 unresolved).src/docs/scale-envelope.mdRankActionableIssuesto its merged location (client.go:4813-4823).src/docs/dashboard-sections.md,labels-and-control-signals.md,notifications.md,fleet-report.mdVerification
No tests, builds or linters were run locally; CI is the verdict. The conflicts were resolved by reading the code.
git grepshows no conflict markers left.Known risk in these forward-merges: the v2-ci per-package coverage gate (e.g.
tuifloor 90,githubfloor 89). If that gate fails, the follow-up is a test top-up, not a revert.— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 requested_by=@clubanderson