Skip to content

feat: ruvnet-brain release-pinned installs, single-owner updates, one drift story across status/statusline/dashboard#33

Merged
pacphi merged 5 commits into
mainfrom
feat/ruvnet-brain-release-pinning
Jul 18, 2026
Merged

feat: ruvnet-brain release-pinned installs, single-owner updates, one drift story across status/statusline/dashboard#33
pacphi merged 5 commits into
mainfrom
feat/ruvnet-brain-release-pinning

Conversation

@pacphi

@pacphi pacphi commented Jul 17, 2026

Copy link
Copy Markdown
Owner

Closes the gaps found in a live audit of the RuvNet Brain installer (cloned stuinfla/ruvnet-brain, tag v3.3.1, 2026-07-17 — installer byte-identical to ruvnet-brain@2.9.1 on npm). Every change below traces to that audit.

What's in it

📦 Release-pinned installs (consistency with every other managed tool)

  • INSTALL_SPEC is now the published installer ruvnet-brain@latest, never github: — the github: spec ran the unreleased default-branch HEAD (3.4.5-dev while releases/latest was v3.3.1), while ak's drift tracking measured against release tags. What's installed now matches what drift measures.
  • installRuvnetBrain() resolves the latest release tag first and pins with --version v<tag>, so the bundle on disk is exactly the release ak stamps in kit.json — closes the install-then-stamp race. Offline behavior unchanged (installer picks latest, stamp best-effort).

🔒 --yes no longer opts users into the installer's extras

The installer's --yes accepts every optional offer (verified in its ask(): if (FLAG_YES) return true). Under ak's non-interactive install that silently enabled a nightly self-update LaunchAgent (macOS, 03:47 — which at v3.3.1 applies downloads without signature verification) and wrote telemetry consent. INSTALL_ARGS now carries --no-nightly-prompt --no-telemetry alongside the existing --no-stack --no-enhance, with a regression-lock test on all four.

🩺 New ruvnet-brain-nightly subsystem

  • ak status warns when the installer's com.ruvnet.brain-update LaunchAgent is active (it bypasses ak-managed updates).
  • ak sync disables it via heal.disableRuvnetBrainNightly(): launchctl bootout then plist removal — both steps, mirroring the installer's own --disable-nightly. Runs under --no-upgrade too (it's a heal, not an upgrade). Own subsystem so a nightly-only plan never force-reinstalls the brain.
  • Deliberately untouched: the spend-watchdog agent (local-only, no code download) and any recorded telemetry consent (a user's answer, not ours to flip). Re-enrollment via npx ruvnet-brain --enable-nightly gets re-flagged; ruvnetBrain:false opts ak out entirely.

🧿 One version story across all three surfaces

Installed version now resolves disk-first: the bundle's own SOURCE.json → releaseTag (stamped by evergreen-era release bundles; validated, v-stripped) → ak's kit.json installedRelease → plugin semver. Shared by:

  • drift() → ak status rows
  • the statusline footer (stays correct even when the KB changes outside ak, e.g. a manual forge-update.mjs run)
  • the dashboard: subsystem cards inherit the rows, and the update banner now folds the brain into the npm drift array via a new pure foldBrainDrift() — previously the banner only covered driftReport() (npm tools), so an outdated brain never surfaced there. Pre-stamping installs show (unversioned), never a fabricated version.

📝 Docs

MAINTAINER.md brain section rewritten: the stale "--enable-nightly does not exist" note corrected, the --yes-accepts-everything gotcha documented, and the three-namespace note updated for the on-disk release stamp. The managed CLAUDE.md block now shows the real install command.

Testing

  • 10 new/updated test cases: SOURCE.json parsing edge cases (valid/missing/malformed/junk/oversized tag), the four-flag + npm-spec regression lock, LaunchAgent detection (darwin-gated, never throws), heal no-op (guarded so it can never mutate a real machine's plist), statusline resolution order, and foldBrainDrift shapes.
  • Full gate green: pnpm run check (typecheck, eslint, markdownlint, build, 103 kit tests + 5 standalone suites).

🤖 Generated with Claude Code

https://claude.ai/code/session_014ns5AAxHpc9GEyNmghanP9

Generated by Claude Code

claude added 5 commits July 17, 2026 23:09
@claude
feat(ruvnet-brain): release-pinned installs, single-owner updates, di…
4c39b01 
…sk-first version

Audited live against the v3.3.1 installer (2026-07-17); every lock below
traces to that audit.

- Install the PUBLISHED installer (`ruvnet-brain@latest`), never `github:` —
  the github: spec ran the unreleased default-branch HEAD (3.4.5-dev while
  releases/latest was v3.3.1), inconsistent with ak's release-versioned
  management of every other tool.
- Resolve the release tag BEFORE installing and pin with `--version v<tag>`,
  so the bundle on disk is exactly the release ak stamps (closes the
  install-then-stamp race).
- Add `--no-nightly-prompt --no-telemetry` to INSTALL_ARGS: the installer's
  `--yes` accepts EVERY optional offer, silently enabling the 03:47 nightly
  self-update LaunchAgent (macOS) — which at v3.3.1 applies downloads without
  signature verification — and writing telemetry consent.
- Installed version now resolves disk-first: SOURCE.json releaseTag (stamped
  by evergreen-era release bundles) → kit.json installedRelease → plugin
  semver. drift(), `ak status`, and the statusline footer share the order,
  so they can never disagree; the footer stays correct even when the KB
  changes outside ak.
- New `ruvnet-brain-nightly` subsystem: status warns when the installer's
  LaunchAgent is active; sync disables it (launchctl bootout + plist removal,
  mirroring the installer's own --disable-nightly; heal, so it runs under
  --no-upgrade too). Spend watchdog + recorded telemetry consent on affected
  machines are deliberately left alone.
- Docs: MAINTAINER.md brain section rewritten (stale "--enable-nightly does
  not exist" note corrected); managed CLAUDE.md block shows the real install
  command.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ns5AAxHpc9GEyNmghanP9
@claude
feat(dashboard): fold ruvnet-brain into the update banner — all manag…
f4de098 
…ed tools, one drift story

The dashboard's drift banner came from driftReport() (npm tools only), so the
brain — release-managed outside npm — never appeared, even when outdated.
collectData now folds it in via foldBrainDrift() (pure, exported, tested) from
the same TTL-cached drift() result `ak status` and the statusline read, so all
three surfaces resolve the version identically (SOURCE.json releaseTag →
kit.json stamp) and can never disagree. Fold happens only on the self-computed
path — an injected status payload that supplies its own drift owns the array
(keeps tests network-free). Absent brain stays off the banner; a pre-stamping
install shows "(unversioned)", never a fabricated version.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ns5AAxHpc9GEyNmghanP9
@claude
fix(test): dashboard suite stays network-free — stub drift in the bar…
ab0ee7b 
…e fixture

The bare-fixture stub omitted `drift`, sending collectData down the
self-computed path: driftReport's npm views plus the new brain fold-in's
GitHub fetch. Beyond violating the suite's own no-network contract (a 3.3s
stall in CI), the fetch's undici keep-alive handle trips libuv's
`!(handle->flags & UV_HANDLE_CLOSING)` assert (win/async.c:94) at process
exit on Windows runners — every test green, then exit code 1. Supplying
`drift: []` (as the first fixture already does) keeps the whole suite offline
and deterministic.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ns5AAxHpc9GEyNmghanP9
@claude
fix(dashboard): fold the kit's own selfDrift into the update banner too
26e305a 
Consistency audit follow-up: status checks self-drift and sync self-updates,
but the banner (driftReport + brain fold-in) never showed an outdated kit.
selfDrift already returns the banner's {pkg, installed, latest, outdated}
shape, so it folds in directly — same self-computed-path guard, same TTL
cache, same best-effort try/catch as the brain.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ns5AAxHpc9GEyNmghanP9
@claude
docs: MANAGED-TOOLS.md — the consistency contract for every tool ak m…
3e06699 
…anages

Dedicated doc for the audit that drove this branch: the four invariants
(disk-first installed versions; sync as the single update owner with honest
disowning; same-namespace drift comparisons; one drift story across
status/statusline/dashboard), the per-tool table mapping all six managed
tools onto them, where each piece lives in the source, and the checklist
that keeps a NEW tool consistent. Linked from README and MAINTAINER.md;
TROUBLESHOOTING's doctor command updated off the retired github: spec.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ns5AAxHpc9GEyNmghanP9
@pacphi
pacphi merged commit 90dd060 into main Jul 18, 2026
11 checks passed
@pacphi
pacphi deleted the feat/ruvnet-brain-release-pinning branch July 18, 2026 00:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@pacphi @claude