Slice A of #644: an author-login Verdict counts only with a tip attestation #1158

Owner: Dev

Slice A of #644. The umbrella stays open: this is the first slice of option B only — the gate parsing an attestation line — and signed markers (#1110) are not in it. No Fixes for this issue may say #644.

The gap this closes

Today decide() accepts a Verdict: written by the pull request author's own GitHub login, because selfReview lists the shared Birfy account. That is how a PASS appeared on #1148 at 10:08:41Z signed Written-by: Reviewer while its Ran numbers (427 pass) did not match the tree (422 tests / 420 pass / 2 skipped). The shared login is not going away in this slice; what changes is that the self-review path stops being free.

Scope

decide() — shared by review-gate, enqueue-guard and merge-gate — accepts a conclusion from the author's login only when the comment header carries, before the first blank line:

Attestation: <full 40-character tip sha> role=<Bot name>

and the sha is the whole tip, the role equals Written-by, and the role is not an Owner: of the pull request. A conclusion from a different GitHub login omits the line and behaves as it does today. Fail-closed in every mismatch: malformed, short sha, wrong role, owner-as-role, missing line.

Acceptance

1. A Verdict: from the author's login without the line is not a conclusion (the case that must be red on the parent of this work).
2. It is a conclusion with a well-formed line naming the whole tip, and stops being one when the sha is a prefix, when role is not Written-by, or when role is an Owner: of the PR.
3. review-gate, enqueue-guard and merge-gate all go through the same decide() — one implementation, tests at that seam.
4. .github/scripts/*.test.mjs green, the four #644: cases pass, and changelog.d/ carries a changed fragment (it is a change to the merge gate's contract, so a fragment is required).
5. docs/autonomous-dev.md, .claude/skills/review-pr/SKILL.md and AGENTS.md tell a reviewer to write the line — a rule the docs do not state is one a reviewer cannot follow.

Explicitly out of this slice

• Ed25519 signatures (#1110) — this line is still self-reported text, so the shared-login hole is narrowed, not closed. The issue title says slice A for that reason.
• Any change to who may merge, or to reviewers/selfReview membership.
• Fixing attestationInHeader's message, which reports "over 8 lines" as "no Attestation line" — same fail-closed direction, different wording; a non-blocking follow-up if anyone wants it exact.

Related #644 (umbrella, stays open), Related #1110 (signatures), Related #1148 (the pull request that implements this slice).