[green-dalii]

main's own Gate 1 went red between 2026-10-23T15:53 and the next run, and every open PR reads as a failure with it. CI runs pnpm audit before the five checks, so a new advisory hides lint, types, build, tests and css-lint behind one line that has nothing to do with the change:

package	vulnerable	patched	advisory
zod	<4.6.6	4.6.7	GHSA-mrhx-c7xc-m3w5
undici	<8.12.2	8.12.3	GHSA-xjj6-f33q-929c

Both are HIGH. Both bumps are inside the same minor line, so neither is a breaking change. pnpm audit --audit-level high --ignore-registry-errors now exits 0; 8 vulnerabilities remain, 2 low and 6 moderate, none high.

The failure mode worth recording: the overrides live in two files, and only one of them is the one you would think of first. package.json carries them, and pnpm-workspace.yaml carries the same list with the reasoning attached. pnpm 10 resolves from the workspace file, so editing package.json alone updated zod — a direct dependency, so it resolved either way — and left undici pinned at 8.10.2. pnpm install reported success in both cases, and the lockfile's own overrides: block was the only place that showed the stale value. The workspace comment already said "keep this list in step with package.json's overrides"; a comment is not a mechanism, so the next version bump will hit this again.

Both lockfiles regenerated — pnpm-lock.yaml, and package-lock.json via npm install --legacy-peer-deps --package-lock-only, per AGENTS.md's note that pnpm overrides are flat and npm cannot honour them.

THIRD-PARTY-NOTICES.md regenerated, because a version bump changes the attribution it records. The test that caught this was third-party-notices.test.ts — which is the check working: it failed with "zod@4.6.7 must have its own section" before the regeneration and passes after.

Gate 1: lint 0, tsc 0, build clean, 349 test files / 4872 tests, css-lint 0.

[green-dalii]

Maintainer merge audit — GitHub does not let the author approve the author's own PR, so this comment is the audit trail for the merge that follows. The maintainer authorised the bypass for this PR explicitly.

Gates checked before merging 16960d0d:

• Gate 1 — lint 0, tsc 0, build clean, 349 test files / 4872 tests, css-lint 0, and pnpm audit --audit-level high --ignore-registry-errors exits 0. The GitHub check Gate 1 / Five-Gate is SUCCESS on this head, which is the point of the PR: it is the first green run on this tree since the advisories landed.
• Closing keywords — none. This PR closes no issue by keyword; the advisory record lives in pnpm-workspace.yaml next to the override.
• AI trailers — none.
• Side effects — two lockfiles and one generated file. undici 8.10.2 → 8.12.3 and zod 4.6.5 → 4.6.7, both inside the same minor line. THIRD-PARTY-NOTICES.md regenerated by the repo's own generator, and third-party-notices.test.ts is the check that required it.
• Breaking changes — none at the API level. The plugin's runtime dependencies are bundled by esbuild, so a minor bump inside zod's 4.x and undici's 8.x does not change the Obsidian-facing surface. Full suite green is the evidence.
• Performance — no change. Both packages are dev-side or bundler-resolved; undici arrives through jsdom in the test tree.
• Why bypass — main's CI is red, and every open PR inherits that red. Merging this is what restores the five-gate signal for feat(core): two-stage Rocchio query expansion — the corpus supplies the bridge #855, fix(ingest): the automatic path now honours Obsidian's excluded-files list #879, feat(tools): the answer context's cost, measured through the production path #881 and feat(tools): the follow-up rate, measured on the user's own conversation #882. The alternative is to wait for a reviewer on a dependency patch while five PRs sit blocked on a failure none of them caused.