Skip to content

🌱 top-up: forward-merge v5 into v6 (v6 release line cut, v5.187–v5.189) - #17118

Merged
clubanderson merged 36 commits into
v6from
topup/v5-to-v6-release-line
Nov 7, 2026
Merged

clubanderson merged 36 commits into
v6from
topup/v5-to-v6-release-line

Conversation

@clubanderson

Copy link
Copy Markdown
Member

Manual forward-merge of v5 into v6 (plain merge commit β€” merge with --merge, not squash, so v6 keeps v5's history and the native-queue-policy DCO path for v5-reachable commits applies). The automated v6-topup.yml runs were failing on conflicts.

What this carries

  • v6 is now an official release line (.github/release-lines.yml: release_lines: [v4, v5, v6]), from v5 PR "✨ release: make v6 an official release line (v6.x.y tags)" (merged as bde4411c9):
    • tagged-release.yml parameterised on RELEASE_LINE=github.ref_name, with the relay job that dispatches the workflow on each non-default declared line after a successful docker.yml run there, and the --is-release-line guard replacing the hard-coded v5 check.
    • promote-stable.yml keeps stable/candidate/:latest on v5 only.
    • derive-release-version.sh / compile-changelog.sh take RELEASE_LINE; on v6 with no v6.* tag the first run derives v6.0.0 (bump=first), consuming the ~2.1k fragments carried on v6 (verified locally on this merge: release=true version=6.0.0 bump=first).
    • Docs: releases.md, release-channels.md, release-line-guard.md checklist, v5-ga.md/release-rollback.md anchors.
  • Tag the built commit instead of deferring (v5 PR merged as dc4f20c13): tagged-release.yml state machine (precheck/backstop/push_line), per-run docker.yml concurrency groups β€” this is what unblocked v5.187.0–v5.189.0.
  • Every other v5 commit since the last top-up (contributor nudge dedupe, review backlog min-severity hold and per-PR cap, cli_memory fleet config, knowledge adviser, settings parity, comment-marker ratchet, …).

Image channel mapping after this lands

Channel / tag Branch Published by
ghcr.io/hivecommons/{hive,hive-contributor,hive-hub}:stable, :candidate, :latest v5 promote-stable.yml (unchanged)
:v5.x.y v5 tagged-release.yml on v5
:edge, :v6-latest v6 docker.yml rows v6 false edge (unchanged)
:v6.x.y (first: v6.0.0) v6 tagged-release.yml dispatched on v6 by the v5 relay job

stable does not follow v6.

Conflict resolutions (v5 behaviour carried into v6 structure)

  • .github/release-lines.yml: v5 manifest + v6-only pinned entries re-added as [-v4, -v5] (hive-action-tests, hive-chat-phone-sweep, prometheusrule-check, kustomize-render).
  • tagged-release.yml, releases.md, release-line-guard.md, test-release-push-retry.sh: v5 versions (already parameterised).
  • pkg/config (v6 split into files): ported v5's cli_memory decode/encode, AdvisorMaxPerKick default, NudgeRepeatAfter; the v6 marshal early-return now also checks AgentsCLIMemory.
  • crux/api/host_setters.go: SetKnowledgeAdviser row. issue_request_watcher.go: origin-note branch in v6's comment chain. review_backlog.go: v5 collapse rewrite + v6 findingNeedsNoAction and SameFilePath. stagelabels/store.go: WithdrawNudge (v6) and MarkNudgedReason/NudgeReasonSlug (v5) both kept; WithdrawNudge now also forgets the reason class. automerge_contributor_nudge.go: withdrawNudge matches v5's reason=<slug> head=<sha> marker.
  • v6-only tests adjusted to v5 semantics: shared-CI withdrawal test expects the rewritten one-comment nudge; same-file-path test sets min_severity: P3; comment-marker ratchet baseline 46 on v6 (v6-only emitters).
  • Docs: packs-and-acmm.md (v5 table + v6 blocker/handoff row, citations re-resolved), stage-labels.md, scale-envelope.md citation.
  • changelog.d/fixed-14225.md: v6-only fragment was missing its leading - (would have failed the fragment guard).
  • TRUSTED_SCRIPTS_TREE re-pinned to 23eed03bebbb82deb0128f396b551939326729f2 (= src/scripts tree of this merge).

Validation

go build ./..., go vet, go test on pkg/..., internal/..., crux/{api,config,checkout,cli,kubejob,serve} (only pre-existing local releasesentinel git-env failures, identical on pristine v5/v6); check-release-lines.sh, test-release-lines-guard.sh, test-release-push-retry.sh PASS; actionlint clean; docs link/citation checks clean; compile-changelog.sh + RELEASE_LINE=v6 derive-release-version.sh β†’ v6.0.0.

After merge

docker.yml push run on v6 β†’ v5's tagged-release.yml relay dispatches tagged-release.yml --ref v6 β†’ tag v6.0.0, GitHub Release with hive-v6.0.0-sbom.spdx.json, hive-contributor-v6.0.0-sbom.spdx.json, hive-hub-v6.0.0-sbom.spdx.json, NOTICE; images ghcr.io/hivecommons/{hive,hive-contributor,hive-hub}:v6.0.0.

clubanderson and others added 30 commits November 7, 2026 17:50
…17014)

* ✨ crux: contained run service with saved run state (S2 + S7 minimum)

Refs hivecommons/crux#114

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* πŸ“ changelog: crux contained run service

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 crux/runstate: cover write obstacles, unreadable records and list ordering

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
…policy verdict (#16988)

* fix(ci): re-validate policy checks before replaying a native-queue-policy verdict

Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>

* fix(ci): fall through to the full wait when a replayed policy provider is pending

replayLastVerdict re-validates POLICY_CHECKS before replaying a prior
success, but assertEvidence throws PendingEvidence for a missing or
still-running provider and the workflow's replay step only tolerates
transient API errors. A cosmetic label flip during a provider re-run
therefore failed native-queue-policy outright with no retry. Catch
PendingEvidence and return null so the caller runs the normal polling
wait; add queued/in-progress/missing provider tests.

Hive-Fix-Attempt: 1/6
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: ci-maintainer <280983584+hivecommons-hive[bot]@users.noreply.github.com>

* fix(ci): fall through on DCO UNKNOWN during replay; refresh replay docs

Address review findings on #16988:
- replayLastVerdict now returns null for ProviderUnavailable too, so the
  full wait re-raises it once and renders the runbook job summary instead
  of a bare setFailed from the label-flip step.
- Re-flow the replayLastVerdict doc comment to ~80 columns.
- native-merge-queue.md: replay re-validates the five policy checks on
  the head; red fails, pending/missing/UNKNOWN falls through.
- Test: UNKNOWN DCO on replay falls through.

Hive-Fix-Attempt: 2/6
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Signed-off-by: ci-maintainer <280983584+hivecommons-hive[bot]@users.noreply.github.com>

---------

Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Signed-off-by: ci-maintainer <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>
…er guess exit code 0 (#17023)

* fix(kubejob): drop SA token mount, harden securityContext, keep unknown exit code honest

Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>

* fix(kubejob): drop runAsNonRoot β€” image USER is non-numeric

Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>

* fix(kubejob): refresh ctx per collection step; let verify pass on unknown exit code

A cancel that lands during pod log collection left podExitCode on a dead
context, so a readable Complete pod reported -1 and
TestRunSuccessDeletesJobEvenIfContextExpiresDuringLogCollection failed.
Every collection/cleanup request now re-checks ctx and falls back to a
fresh request-bounded context. The empty exitCodeUnknown branch (SA9003)
is folded into a comment.

SandboxVerifyResult.Passed now accepts kubejob.ExitCodeUnknown when the
launcher returned no error and did not time out, so a Complete verify
Job whose pod is gone no longer withholds the PR β€” matching the publish
gate and the PR description. Changelog notes the -1 transcript/dashboard
rendering.

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>

---------

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>
Since #16785 the proxy's destination guard covers plain-HTTP forwards,
uninspected CONNECT tunnels and transparent-TLS passthrough, and
HIVE_PROXY_ALLOW_PRIVATE_FORWARD=true lifts it for all three legs.
env-vars.md and security-model.md still described the guard and its
override as plain-HTTP-only; bring both in line with the code.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
* 🌱 build(deps): bump ws from 8.21.3 to 8.22.1 in /src/proxy

Bumps [ws](https://github.com/websockets/ws) from 8.21.3 to 8.22.1.
- [Release notes](https://github.com/websockets/ws/releases)
- [Commits](websockets/ws@8.21.3...8.22.1)

---
updated-dependencies:
- dependency-name: ws
  dependency-version: 8.22.1
  dependency-type: direct:development
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

* changelog: fragment for #16567

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: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

---------

Signed-off-by: dependabot[bot] <support@github.com>
Signed-off-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Answers the 2026-11-05 community-meeting question nobody could answer:
which context reaches an agent and from where. Separates the three layers
(Hive knowledge primer, CLI conversation, CLI persistent memory), traces
the kick-primer path (Primer.PrimeScoped -> primeKnowledge -> # Relevant
Knowledge) with its scoping rules and size caps, explains why repo
AGENTS.md is injected by the scheduler rather than read natively, lists
which per-agent home directories are fleet-shared, and tabulates what
survives a kick, /clear, pod restart or a new agent. In-flight memory
isolation work (#17031, #17032, #17033, PR #17039) is linked as known
gaps rather than described as current behaviour.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…o placeholders (#17046)

handleApproveProvision only ever looked in the pool mapped from the
request's auth_method, so a public request 409'd when hive-oke had no
available placeholder even though lke648397 had 30. Auto-pick now tries
the preferred pool first and, for the soft public/default preference,
falls back to any other cluster registered in clusters.json that has an
available placeholder (most available first, tie-break by cluster id),
logging one INFO line. Private requests stay pinned to the GPU pool, the
GPU pool is never used for public requests, and an admin's explicit
placeholder choice remains strict.

Fixes #17044

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…b's cap; GitHub failures no longer read as "hive unavailable" (#17041)

* πŸ› fix(dashboard): Hold works on a hive whose hold label exceeds GitHub's cap and GitHub failures are not reported as "hive unavailable"

Clicking Hold on an Activity Monitor row of a hosted hive answered
"hive unavailable" while the hive was up. The hive's hold label
hive-pause/hosted-available-oke-11-placeholder-r05x is 51 chars, over
GitHub's 50-char label-name cap, so CreateLabel answered 422;
EnsureIssueLabel swallowed every 422 as "already exists",
AddLabelsToIssue then failed 422, the handler mapped it to 502, and the
hosted Ingress (custom-http-errors 502,503) replaced the body with its
generic outage page.

- CanonicalHiveHoldLabel always fits the cap: over it, the id is
  truncated and suffixed with an FNV hash of the full id; the dashboard
  band classifier now derives the label through the same function.
- EnsureIssueLabel/ensureLabel only treat an already_exists 422 as
  success and return any other 422 with GitHub's reason.
- handleRepoItemHold answers GitHub failures (and a missing client) with
  424 and the step + reason, logged on the spoke, never 502/503.
- A row action that still receives the gateway's placeholder body on a
  502-504 names the gateway and the status instead of repeating
  "hive unavailable" as the hive's word.

Fixes #17037

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* docs: refresh scale-envelope citation after client.go line shift

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
… scoped knowledge entries (#17051)

AGENTS.md goes from 3736 to 1980 bytes. Every removed fact lives in exactly
one src/docs/knowledge/<topic>.md entry with frontmatter and a scope: block
(paths, labels, lanes), traceable via source: AGENTS.md. The rules themselves
(CI is the verdict, never run the suite in an agent pane, fragment required,
citation --fix first) stay.

Proposed by @gregoryhunt at the 2026-11-05 community meeting.

Fixes #17047
Part of #17043

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…nd entries skipped for the cap (#17064)

primeKnowledge now returns its numbers (bytes, entries, skipped-for-cap,
cap) alongside the rendered section, and every primed kick logs one line
"knowledge: injected N bytes from M entries (skipped K for cap)" with the
AGENTS.md and skills block sizes that share the kick. The old
"injecting facts into kick" line is folded into it; the debounced
over-cap detail line stays. Silent when the primer did not run, so a hive
without a knowledge store does not log a zero per kick.

Proposed by @gregoryhunt at the 2026-11-05 community meeting.

Fixes #17049
Part of #17043

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
….gemini/.bob state to the shared home (#17039)

* πŸ› agents: give each agent its own CLI memory β€” stop bridging .claude/.gemini/.bob state to the shared home

interactiveHomeBridgeDirs symlinked ~/.claude, ~/.gemini and ~/.bob whole
into /data/home, so every Claude Code agent on a hive read and rewrote
the same projects/<cwd>/memory/MEMORY.md (and shared transcripts/todos),
Gemini agents shared GEMINI.md and bob agents shared chat recordings.

Make those three REAL per-agent directories following the .copilot
pattern (#9444): only the login/settings entries are bridged
(.credentials.json + settings.json for claude; antigravity-cli/,
settings.json, oauth_creds.json, google_accounts.json, installation_id
for gemini; settings.json + settings/ for bob). Legacy whole-dir bridges
are retired on launch; nothing under /data/home is deleted β€” an existing
shared memory store is logged once for operator review.

Dashboard token scanners (claude_sessions_dir, bob_sessions_dir) now also
walk /data/home/agents/*/.claude/projects and /.bob, attributing each
per-agent tree to its owner, mirroring CopilotSessionTrees.

Fixes #17031

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* πŸ“– agents: correct the shared-memory wording β€” MEMORY.md is per-agent by cwd, the tree is what was shared

Claude keys auto-memory by cwd (/data/agents/<name>), so MEMORY.md was never
one physical file; the user-level CLAUDE.md, history.jsonl and todos/ were.
The bridged tree being readable and writable by every agent is the bug.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 agents: keep manager.go inside its file-size ratchet baseline

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* πŸ“– agents: bring agent-knowledge-and-memory.md in line with per-agent CLI homes

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
…tead of reposting per head (#17071)

* 🌱 automerge: upsert signoff-attribution and unauthorized comments instead of reposting per head

hive:signoff-attribution becomes one comment per PR under a stable marker,
rewritten in place on a new head with an 'updated …' history capped at five
lines. hive:unauthorized becomes one comment per PR and blocked step; a new
head hitting the same block edits it with an attempt count and the heads it
covers. Legacy per-head comments are adopted and upgraded in place.
workflowFenceReleased reads head coverage from the upserted note.

Fixes #17066

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 automerge: fix capped-history test heads; move nudge raw-write baseline line to the signoff file

The sign-off attribution CreateComment moved files unchanged (2β†’1 in the
nudge file, 1 in automerge_signoff_attribution.go); the test heads used
%040d, whose 12-char short SHA was identical for every head.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…s work enters a new scope (#17068)

* ✨ knowledge advisor: surface scoped knowledge mid-task when an agent's work enters a new scope

Phase 1 of the mid-task knowledge advisor (#17052). The
kick primer runs once; this watches the agent's pane on every poll tick,
and the first time a scope (directory prefix or gh command family) shows
up that was not on screen at kick time, asks the same knowledge Primer
under the same agent scope and types one 'πŸ“š Knowledge:' line per hit
through the existing live-session primitive used by the CI-poll nudge.

- crux: KnowledgeAdviser host seam (in-process-only, like SetSpanStarter),
  crux/agent/knowledge_advisor.go: baseline at first post-kick poll,
  per-kick cap, cooldown, dedupe against kick-primed slugs, never during
  a tool call or after the turn ends, prompt-history trigger
  knowledge-advisor for measurement.
- hive: cmd/hive/knowledge_adviser.go adapts Primer.PrimeScoped; the
  primer now records which slugs survived the byte cap (RenderedSlugs,
  Scheduler.KickPrimedSlugs) so those are never re-sent.
- config: knowledge.advisor (default off), advisor_max_per_kick (3),
  advisor_exclude_agents.
- docs: knowledge-curator.md section, operator-reference row, crux.md
  HostSetters row.

Phase 2 (LLM-ranked, OMP-style) is #17054.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 knowledge advisor: reuse oneLine, keep runtime.go and recorder.go at ratchet baselines

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 knowledge advisor: rune-cap the typed summary, golden + settings-parity fixtures, keep manager.go at baseline

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…#17083)

Tagged Release deferred whenever v5's tip had moved past the commit
docker.yml built, expecting the successor run to release. With merges
landing faster than a multi-arch build no run ever saw an unmoved tip,
every successor deferred again, and the hourly backstop applied the same
tip-only test β€” v5 went from v5.186.4 (2026-10-18) to ~1,700 untagged
commits.

- precheck: release the built SHA when it is an ancestor of v5's tip and
  newer than the last v5.* tag; defer only when it is off-branch or a
  release already landed after it (version collision).
- push_v5: the version tag names the release commit (built SHA + changelog),
  never the merge commit; 409 base-moved is retried, head-moved and a
  CONFLICTING PR hard-fail; the post-merge docker dispatch passes
  release_version only when the merge adds nothing but the changelog.
- backstop: pick the newest successfully built ancestor of the tip instead
  of requiring the tip itself.
- docker.yml: push builds get a per-run concurrency group β€” GitHub's queue
  holds one pending run per group and cancels the rest, which is what was
  killing most v5 builds.
- test-release-push-retry.sh covers the new state machine; releases.md and
  the RACES/BACKSTOP headers describe the new rule; TRUSTED_SCRIPTS_TREE
  re-pinned.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…Primer (#17060)

* ✨ knowledge: scope block on entries and a kick-context filter in the Primer

A knowledge entry can now say where it applies with a frontmatter scope:
block (paths, labels, lanes). FileStore parses it into Fact.Scope;
Primer.PrimeForKick takes the kick's lane, issue labels and known file
paths and keeps a scoped fact only when one dimension matches (OR), both
in the ranked set and in graph expansion. Unscoped facts are unchanged.
The scheduler passes the base lane name and the primed issues' labels;
paths stay empty on the kick path. The count excluded is reported as
PrimedKnowledge.ScopedOut and in the "knowledge primed" log line.

Proposed by @gregoryhunt at the 2026-11-05 community meeting.

Fixes #17045
Part of #17043

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 test: Agents lives in the embedded AgentRuntimeConfig

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…ame-family findings (#17072)

* review backlog: min_severity floor, per-PR lifetime cap, collapse same-family findings

Findings below review.backlog.min_severity (default P2) are no longer
filed as backlog issues; they are appended to one upserted
hive:review-followups comment on the PR. The issue cap is now
review.backlog.max_per_pr (default 3) over the PR's lifetime, persisted
in the review-backlog ledger, instead of per UTC day
(max_per_pr_per_day is read as a retired alias). New findings from one
pass that share a perspective and scope are filed as a single
checklist item.

Operators who want every finding filed as before set min_severity: P3.

Fixes #17065

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* review backlog: tests follow collapsed items and uncapped still-present notes

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* review backlog: expose min_severity and max_per_pr in Settings

TestSettingsDialogReferencesEveryHiveYAMLLeaf requires every hive.yaml
leaf to have a Settings control. Add a min_severity select and a
lifetime max_per_pr input next to the retired per-day cap, accept both
on PUT /api/config/review, and cover them in the 11088 settings test.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…/v5-to-v6 forward-merges (#17050)

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 v5.187.0 entry.
See src/docs/releases.md.

Signed-off-by: hive-release-bot <actions@github.com>
…ocal checkout (#17093)

Hosted spokes keep no clone of their primary repo, so both roots
agentsRepoRoot knows (project.checkouts_dir, policies.local_dir) are empty
and primeAgentsMd was a silent no-op: hive's own hive showed zero agentsmd:
log lines. Add a third source: when no local root exists and a GitHub
client is wired, fetch AGENTS.md from the repo's default branch, cache it
for 15 minutes (the conditional-GET transport makes refreshes free), keep
the last-known-good copy across API failures, and materialise it under
/data/agentsmd-cache/<owner>/<repo>/ so Parse/Render/primeSkills and the
injection cap run unchanged.

Fixes #17090

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
…per-agent memory before each kick (#17089)

* ✨ agents: CLI auto-memory off by default; clear_memory_on_kick wipes per-agent memory before each kick

Adds agents.cli_memory {auto_memory, clear_on_kick} with per-agent override,
seeds the vendor off-switches (Claude env var, Codex [features] memories,
Copilot settings.json memory) and removes the agent's own memory stores next
to the /clear kick when clear_on_kick is set.

Fixes #17032

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 agents: keep manager_homes.go inside its ratchet baseline; document cli_memory in agent-knowledge-and-memory.md

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 agents: cli_memory test imports crux/config (outbound ratchet); settingsparity tool fixture gains the two cli_memory leaves

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.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: clubanderson <407614+clubanderson@users.noreply.github.com>
…17078)

Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
…ates lists draft (#17079)

The kick primer admits approved facts only and the agent scope can only
narrow, so knowledge.agent_scopes.<agent>.include_states: [draft] never
reached a kick. Thread the agent's identities (name, base agent, owner
logins) into TOCScope.DraftAuthors when the scope lists draft, admit drafts
whose author: frontmatter matches, and render them with a (draft) marker.
Approved-only stays the default for every other agent.

Fixes #17076

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 v5.188.0 entry.
See src/docs/releases.md.

Signed-off-by: hive-release-bot <actions@github.com>
…ces design status (#17035)

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 v5.189.0 entry.
See src/docs/releases.md.

Signed-off-by: hive-release-bot <actions@github.com>
…e link check on v5 (#17096)

Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
hivecommons-hive Bot and others added 5 commits November 7, 2026 21:25
Add docs/rfc-17054-llm-ranked-knowledge-advisor.md arguing the open
questions on #17054: rank-only (slug must be a matcher candidate, model
reason never typed), hive-paid per-agent/fleet call caps, in-process
behind phase 1's KnowledgeAdviser seam with Advise moved off the poll
goroutine, pane tail + kick summary as signal, and shadow/holdout
measurement through the retro lane's existing counters.

Hive-Run: #17054
Hive-Plan: knowledge advisor phase 2 (RFC)
Hive-Spec: knowledge-advisor#17054

Signed-off-by: Karibhou <Karibhou@users.noreply.github.com>
Co-authored-by: Karibhou <Karibhou@users.noreply.github.com>
Co-authored-by: clubanderson <407614+clubanderson@users.noreply.github.com>
Co-authored-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
…; marker ratchet (#17073)

* 🌱 contributor-nudge once per PR+reason; fold origin notes into status; marker ratchet

Contributor nudges carry a hive:contributor-nudge reason=<slug> marker and
are edited in place for a new head or changed reason, with one re-nudge after
auto_merge.nudge_repeat_after (default 72h). Agent free-text relay comments
whose only marker is the hive:origin footer fold into the hive:status card
in quiet mode or one hive:origin-notes comment per item otherwise. A
grep-based ratchet in internal/testutil caps unmarked raw comment emitters
under pkg/github at the current 45.

Fixes #17067

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* contributor nudge: expose auto_merge.nudge_repeat_after in Settings

TestSettingsDialogReferencesEveryHiveYAMLLeaf requires every hive.yaml
leaf to have a Settings control. Add a Re-nudge-after field to the
Labels tab next to the legacy renudge_interval it supersedes, read and
write it via /api/config/governor/labels with the same positive-duration
validation, and cover the round trip.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* stagelabels: cover MarkNudgedReason, NudgeReasonSlug and NudgeStart

The per-reason nudge ledger added for #17067 had no
direct tests, dropping pkg/stagelabels under its 90% coverage floor.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Co-authored-by: hivecommons-hive[bot] <280983584+hivecommons-hive[bot]@users.noreply.github.com>
…and refuse non-canonical paths (#17104)

Allowed() compared the caller's URL string with strings.HasPrefix and
parseRepo never normalized the path, so dot segments (which libcurl
squashes and local git follows literally), percent-encoded segments and
mid-segment prefixes all cloned repositories outside the allowlist.

Signed-off-by: sec-check <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Co-authored-by: sec-check <280983584+hivecommons-hive[bot]@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* ✨ release: make v6 an official release line (v6.x.y tags)

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* πŸ“– docs: full release-line cut checklist, ROADMAP line table for v6

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* πŸ“– docs: repoint releases.md anchor after section rename

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

* 🌱 ci: carry v6's test-release-push-retry.sh (push_line step id) into v5

The parameterised tagged-release.yml renames push_v5 to push_line; the
contract test must follow. Re-pins TRUSTED_SCRIPTS_TREE.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

---------

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
Manual v5 -> v6 top-up carrying #17074 (v6 under release_lines,
parameterised tagged-release.yml with relay job, promote-stable guard,
release docs) and #17083 (tag the built ancestor; per-run docker
concurrency), plus every v5 commit since the last top-up.

Conflict resolutions (v5 behaviour carried into v6 structure):
- release-lines.yml: v5 manifest + v6-only pinned entries as [-v4, -v5].
- tagged-release.yml, releases.md, release-line-guard.md,
  test-release-push-retry.sh: v5 (already parameterised on RELEASE_LINE).
- pkg/config: v5's cli_memory decode/encode, AdvisorMaxPerKick default,
  NudgeRepeatAfter ported into v6's split files.
- crux/api: SetKnowledgeAdviser row in host_setters.go.
- issue_request_watcher.go: origin-note branch in v6's comment chain.
- review_backlog.go: v5 collapse rewrite + v6 findingNeedsNoAction and
  SameFilePath argument.
- stagelabels/store.go: WithdrawNudge (v6) and MarkNudgedReason/
  NudgeReasonSlug (v5) both kept.
- held_runs_test.go: both sides' new tests.
- docs: packs-and-acmm table (v5 + v6 blocker/handoff row, citations
  re-resolved), stage-labels (v5 per-PR nudge semantics + v6 prose),
  scale-envelope citation.
- changelog.d/fixed-14225.md: v6-only fragment given its missing '- '.
- TRUSTED_SCRIPTS_TREE re-pinned to 23eed03.
- v5's reason-keyed nudge marker and one-comment-per-PR rewrite applied
  to v6's shared-CI nudge withdrawal: withdrawNudge matches the head
  field behind any reason slug, WithdrawNudge forgets the reason class,
  and the withdrawn comment is rewritten into the fresh first nudge.
- marshalConfigYAMLWithAgentCap early-return keeps encoding
  agents.cli_memory (v6 split's guard did not know the new field).
- review_backlog_same_file_path_test: min_severity P3 so the low nit is
  filed under v5's below-min-severity hold.
- comment-marker ratchet baseline 46 on v6 (its own emitters outnumber
  the v5-only ones it lacks).

Signed-off-by: Andrew Anderson <andy@clubanderson.com>
@kubestellar-prow kubestellar-prow Bot added dco-signoff: yes Indicates the PR's author has signed the DCO. size/XXL Denotes a PR that changes 1000+ lines, ignoring generated files. labels Nov 7, 2026
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

Docs map: this PR adds page(s) under src/docs/ that
src/docs/documentation-map.md does not link:

  • knowledge-entries.md

A page missing from the map is one readers do not find. Please add a
line in the section a reader would actually look in, matching the
format of its neighbours.

If the page is indexed from a subdirectory README, or is deliberately
unlisted for now, ignore this. It is a reminder, not a gate; it never
blocks a merge.

@kubestellar-prow

Copy link
Copy Markdown
Contributor

[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 kellyaa 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:

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

@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

Open PR overlap check

Base branch: v6

Conflicts are checked with every PR merged onto the current tip of v6, so a PR that only needs a rebase is not reported as conflicting with its siblings.

PR Title Relation Files
#17025 🌱 Forward-merge v5 into v6 (1 commits) same files .github/scripts/native-queue-policy.cjs
.github/scripts/native-queue-policy.test.cjs
src/crux/agent/sandbox_executor.go
src/crux/agent/sandbox_kick_log.go
src/crux/agent/sandbox_kick_log_test.go
src/crux/api/runs.go
src/crux/api/runs_client.go
src/crux/api/runs_test.go
src/crux/cli/cli.go
src/crux/cli/remote.go
src/crux/cli/runs.go
src/crux/cli/runs_test.go
src/crux/cli/serve.go
src/crux/cli/verbs_test.go
src/crux/kubejob/kubejob.go
src/crux/kubejob/kubejob_test.go
src/crux/runs/backend.go
src/crux/runs/config.go
src/crux/runs/config_test.go
src/crux/runs/provision.go
src/crux/runs/service.go
src/crux/runs/service_test.go
src/crux/runstate/runstate.go
src/crux/runstate/runstate_test.go
src/crux/serve/runs.go
src/crux/serve/runs_test.go
src/crux/serve/serve.go
src/docs/agent-knowledge-and-memory.md
src/docs/agents-md.md
src/docs/documentation-map.md
src/docs/env-vars.md
src/docs/knowledge-curator.md
src/docs/manual-provisioning.md
src/docs/native-merge-queue.md
src/docs/operator-reference/README.md
src/docs/security-model.md
src/pkg/hub/autopick_pool_fallback_test.go
src/pkg/hub/saas_provisioning.go
src/proxy/package-lock.json
src/proxy/package.json

@hivecommons-hive

hivecommons-hive Bot commented Nov 7, 2026 •

Copy link
Copy Markdown
Contributor

Hive status β€” updated in place; edits do not notify. Full history is in the Activity Monitor.

@clubanderson this PR is held at stage/071-needs-author and the hive will not merge it yet.

Failing checks:

  • check-failure

Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path.

Review findings from reviewer-2 (folded here because this head already carries its hive review):

Read at e2d3260. The merge is structurally sound; I found no code defect in the conflict resolutions I could check. One thing the body does not say, which a human should consciously accept before this merges and auto-publishes v6.0.0:

The merge commit restores ~1,373 changelog fragments that v5 deleted. 6888c77 is dd6b372 (v6 tip) β¨― bde4411 (v5). Between the merge-base 7284b56 and bde4411, v5 deleted 1,373 changelog.d/* files (consumed by the v5.187.0–v5.189.0 releases) and added 3. A mechanical git merge-tree dd6b372 bde4411 carries those deletions and leaves 770 fragments on v6; the merge commit keeps 2,143 (= v6's 2,140 + v5's 3). Earlier top-ups (#17024, #16997, #16688, #16683) never faced this because no v5 release had consumed fragments since v6 was cut. The body says the first v6 run "consum[es] the ~2.1k fragments carried on v6" and derives v6.0.0, so retaining them looks deliberate β€” but the Conflict resolutions section does not mention it, and the consequence is that v6.0.0's compiled changelog repeats everything already shipped in v5.187–v5.189. If that is intended, a one-line note in the body would save the next person the same archaeology; if not, the fix is a follow-up commit deleting the 1,373 paths v5 removed.

What I verified and found correct:

  • .github/scripts/integration-governance.cjs:12 β€” TRUSTED_SCRIPTS_TREE = 23eed03b… equals git rev-parse e2d3260:src/scripts, so the :102 check passes on this head (and the follow-up commit e2d3260 only touches .github/scripts/, which is not what the pin covers).
  • .github/workflows/docker.yml β€” the only diff against v5's copy is the three v6 false edge / RELEASE_BRANCH: v6 / CHANNELS: edge rows, exactly the per-branch divergence src/docs/release-line-guard.md item 7 requires; tagged-release.yml, promote-stable.yml, derive-release-version.sh, compile-changelog.sh are byte-identical to v5 at bde4411.
  • .github/release-lines.yml β€” v5 manifest plus the four v6-only [-v4, -v5] entries, as the body states.
  • e2d3260 updates .github/scripts/test-native-queue-workflows.py for the pin now reading [-v4]; the pre-merge v6 literal was [v6, -v4], so the follow-up is needed, not optional.

Also for whoever merges:

This touches release workflows and triggers a public release on merge, so I am not signing off on it β€” flagging for clubanderson as repository owner. I have not changed anything.

@hivecommons-hive

Copy link
Copy Markdown
Contributor
6888c77

ℹ️ Sentinel notice β€” informational; author is trusted (owner), merge is not blocked

Hive flagged this PR (author @clubanderson, head 6888c7720c83) because it matches behaviors that can override security controls, escalate privileges or damage the codebase. This is a heuristic, not an accusation β€” a maintainer should confirm the change is intended before it merges.

  • sensitive_path β€” changes 14 sensitive path(s) (Touches a sensitive path (OWNERS, workflows, policies, hive config, security docs, …))
    • .github/release-lines.yml
    • .github/workflows/dco-post-merge.yml
    • .github/workflows/dco-push-delta.yml
    • .github/workflows/docker.yml
    • .github/workflows/podman-arm64-lane.yml
    • .github/workflows/podman-contract.yml
    • .github/workflows/podman-rootful-lane.yml
    • .github/workflows/podman-rootless-lane.yml
    • .github/workflows/quadlet-gate.yml
    • .github/workflows/scorecard.yml
    • .github/workflows/suid-contract.yml
    • .github/workflows/tagged-release.yml
    • src/proxy/package-lock.json
    • src/proxy/package.json
  • permission_escalation β€” widens workflow permissions or uses pull_request_target (Widens GitHub Actions permissions or switches to pull_request_target)
    • .github/workflows/tagged-release.yml
  • ci_gate_weakening β€” CI configuration is deleted, softened or has gate checks removed (Disables or softens CI gates (continue-on-error, || true, removed checks))
    • .github/workflows/tagged-release.yml
  • remote_code_execution β€” adds a remote-fetch-and-execute, decoded payload, reverse shell or embedded credential (Adds curl|sh, eval of encoded payloads, reverse shells or embedded private keys)
    • src/crux/runs/service_test.go

Hive did not add the sentinel-alert label because this author is trusted. The finding remains recorded for audit and the dashboard Security tab; set sentinel.trusted_authors_block: true to require the blocking label for trusted authors too.

v6 is a release_lines entry now, so native-queue-policy.yml's pin reads
[-v4] (v5 and v6 covered) rather than [v6, -v4]; the test asserted the
old literal.

Signed-off-by: Andrew Anderson <andy@clubanderson.com>

@hivecommons-hive hivecommons-hive Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Decision needed: @clubanderson should decide whether to admin-merge #17118 past the red probot DCO / native-queue-policy checks, or re-shape it (push the tip as a two-parent merge of origin/v5 onto v6 on sync/v5-to-v6 via v6-topup.yml) so the v5-reachable DCO substitution in native-queue-policy.cjs actually applies.

Routing to @clubanderson β€” this is a 214-file forward-merge that touches .github/workflows (tagged-release.yml, release-lines.yml), re-pins TRUSTED_SCRIPTS_TREE, and currently cannot clear native-queue-policy. I did not review the 214-file content; what follows is the one thing I could verify end-to-end.

The DCO path the body relies on does not apply to this PR as shaped.

The body says the plain merge commit is kept so "the native-queue-policy DCO path for v5-reachable commits applies". The substitution is forwardMergeDCOPath in .github/scripts/native-queue-policy.cjs:171-180, and it returns null (probot DCO stays required) unless all of:

  • pr.head.ref === 'sync/v5-to-v6' (FORWARD_MERGE.head, line 30) β€” this PR's head is topup/v5-to-v6-release-line;
  • pr.user is github-actions[bot] id 41898282 (line 31) β€” this PR is authored by a human account;
  • the tip commit has exactly two parents, one reachable from v6 and one from v5 (lines 186-194) β€” the tip e2d3260 ("🌱 ci: native-queue contract test resolves the pin against release_lines") has a single parent 6888c77, because the fix-up was committed on top of the merge.

Consequence: probot DCO is the evidence (DCO: action_required), and assertEvidence throws at native-queue-policy.cjs:156 β€” that is the Unhandled error: Error: DCO: action_required in job 128551301218. Of the 36 commits in the PR, the three without Signed-off-by are GitHub-generated merge commits already on v5 (ec3d6fe8e #17094, 9f04cc744 #17100, 30468b7e9 #17108). Whether those are what probot objects to is an open question I could not confirm from the check output; I only verified that the v5-reachable bypass is not reached.

The comment in the policy file says "Any doubt keeps probot DCO required" by design, so this is a decision for the owner, not a code fix I can recommend: either merge it as-is with admin rights (accepting the red required check), or re-shape the branch so the narrow path matches.

Not touched: the Go/config conflict resolutions and the release-line workflow changes. Someone should read tagged-release.yml and promote-stable.yml in the merged tree before this lands, since they decide where :stable and v6.0.0 get published.

I have not approved, closed, or labelled anything.

β€” hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88

@clubanderson
clubanderson merged commit 87bfaf3 into v6 Nov 7, 2026
96 of 99 checks passed
@kubestellar-prow
kubestellar-prow Bot deleted the topup/v5-to-v6-release-line branch November 7, 2026 22:52
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

Thank you for your contribution! Your PR has been merged.

We'd love to hear how your experience was: share feedback

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

dco-signoff: yes Indicates the PR's author has signed the DCO. size/XXL Denotes a PR that changes 1000+ lines, ignoring generated files.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants