Files
ruvnet__ruflo/.github/CODEOWNERS
ruv 0181fb85c2 ci: #2046 supply-chain hardening — five-layer audit + dependency review + CODEOWNERS
After ADR-122 + ADR-123 expanded ruflo's published-package surface
(@claude-flow/browser, ruflo-graph-intelligence) the supply-chain attack
surface grew alongside it. This commit puts five layers of defence on
every PR + push that touches package.json, lockfiles, or the audit
itself.

New files:

  .github/supply-chain/allowed-deps.json
    Whitelist of direct deps per audited package (ruflo-graph-intelligence,
    @claude-flow/browser). New top-level deps require an explicit edit here.
    Also lists policies.rejectInRegistry (typosquat block-list — ruflo-graph-
    intellligence, sublinear-time-solvor, agent-browse, etc.) and
    publisherTrust.criticalUpstreamPackages (upstream identities snapshotted
    in CI logs so maintainer takeovers are visible).

  .github/supply-chain/accepted-findings.json
    Triaged findings the audit currently tolerates. Each entry cites a
    tracking issue or ADR and carries an expiresAt date so the audit
    re-fails after triage window closes. CODEOWNER review required to
    edit.

  .github/supply-chain/README.md
    Operator guide: how to run the audit locally, how to triage new
    findings, how to add a new package to coverage.

  .github/CODEOWNERS
    Workflows / lockfiles / plugin manifests / supply-chain config now
    require explicit maintainer review. Pairs with the automated audit.

  scripts/audit-supply-chain.mjs
    Five-layer audit:
      [1/5] CVE — npm audit --audit-level=high, fail on HIGH/CRITICAL in
            DIRECT deps unless accepted
      [2/5] Lockfile integrity — every downloaded entry must carry
            SHA-512 in `integrity` (skip pnpm-shadow + workspace-link
            entries by construction)
      [3/5] Top-level allowlist — package.json deps must match
            allowed-deps.json
      [4/5] Typosquat reject — names matching policies.rejectInRegistry
            fail regardless of upstream status
      [5/5] Publisher trust — npm view maintainers, log-only (informational)

  scripts/__tests__/audit-supply-chain.test.mjs
    Smoke test for the audit itself (full-run / --json / each --scope).

Workflow additions in .github/workflows/v3-ci.yml:

  supply-chain-audit job — runs the audit + the smoke test on every PR
    and push that touches package.json, lockfile, allowed-deps,
    accepted-findings, or the audit script.

  dependency-review job — actions/dependency-review-action@v4 (PR-only)
    flags any new vulnerable dep added by the PR vs base, fails on
    `high` severity, denies known-bad licenses (AGPL/GPL/SSPL) and
    blocks specific compromised versions (event-stream, flatmap-stream,
    ua-parser-js@0.7.29 / 0.8.0 / 1.0.0, colors@1.4.1, faker@6.6.6).

  Path filter expanded to include **/package.json, **/package-lock.json,
    **/pnpm-lock.yaml, .github/supply-chain/**, scripts/audit-supply-chain.mjs

Dependabot additions in .github/dependabot.yml:

  Per-plugin npm stanzas for ruflo-graph-intelligence and
  @claude-flow/browser. Major-version bumps are blocked from auto-PR
  (they require allowlist re-approval); patch + minor auto-PR'd.

One real-world finding the audit caught on first run + how it was
triaged (recorded in accepted-findings.json):

  - HIGH CVE on @claude-flow/browser → agentic-flow (via @xenova/
    transformers, agentdb). Already tracked in ADR-121 Phase 4
    (Xenova retirement migration). Marked accepted with expiresAt
    2026-08-19 so the audit re-fails if the migration hasn't
    landed by then.

Audit run on this branch: PASS (0 unaccepted CVE / 0 lockfile issue
/ 0 allowlist violation / 0 typosquat / 5 publisher-trust entries
logged). Audit smoke test: 5/5 passing.

Co-Authored-By: RuFlo <ruv@ruv.net>
2026-05-19 08:23:30 -04:00

39 lines
1.6 KiB
Plaintext

# CODEOWNERS — security-sensitive paths require explicit review.
#
# A PR that touches any of these files cannot be merged without sign-off
# from the listed owners. This is the human-review layer of the supply-chain
# defence (the CI guard is the automated layer).
#
# Format: <path-glob> <reviewer> ...
# See https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/customizing-your-repository/about-code-owners
# CI workflow files — a malicious workflow change can compromise everything
# that runs in CI. Always require explicit review.
/.github/workflows/ @ruvnet
/.github/CODEOWNERS @ruvnet
/.github/supply-chain/ @ruvnet
/.github/dependabot.yml @ruvnet
# Lockfiles — silent dep substitution via lockfile editing is a known supply-
# chain attack vector. Maintainer review required.
/package-lock.json @ruvnet
/v3/pnpm-lock.yaml @ruvnet
**/package-lock.json @ruvnet
**/pnpm-lock.yaml @ruvnet
# Plugin manifests — listing in the marketplace, publishing identity.
**/.claude-plugin/plugin.json @ruvnet
# Witness manifests + provenance — the ADR-103 root of trust.
/verification/ @ruvnet
/v3/docs/adr/ADR-103-witness-temporal-history.md @ruvnet
# Supply chain scripts themselves — the auditor's audit.
/scripts/audit-supply-chain.mjs @ruvnet
/scripts/audit-package-dep-overlap.mjs @ruvnet
/scripts/audit-plugin-packages.mjs @ruvnet
/scripts/audit-hook-commands.mjs @ruvnet
# Default fallback — every other PR gets standard review.
* @ruvnet