merge-gate refuses every real PASS: review-gate only reads a bare Ran: line, not the reviewer's Ran heading and fenced $ commands #667

Environment

main at 79f4742 (after #658, c74e120). merge-gate on merge_group, which reuses .github/scripts/review-gate.mjs decide/parseReport.

Actual

Every queue entry since #658 refuses: REFUSE — Birfy's PASS lists no command it ran — runs 39707184514 (#660), 39707395505 (#661), 39707549765 (#662), 39707707974 (#663). Each PR has a real PASS at its head, e.g. #662 (comment):

Ran at this tip (tarball of the tip, nothing patched by hand):

```text
$ cd plugins/social && TZ=UTC DSH_HARNESS=/opt/flybot node --import ./test/resolve.mjs --test "test/*.test.js"
# tests 318 / pass 318 / fail 0
```

Expected

A Ran …: heading over a fenced block of $ commands counts as the Ran evidence, as - bullets do. Policy unchanged: at least one real command; empty Ran: / Ran: - refuse; a quoted block's Ran: cannot replace the conclusion's own (#640).

Where it sits

• .github/scripts/review-gate.mjs:32 strips every backtick, so a fence is invisible.
• :65 /^Ran\s*:\s*(.*)$/ matches only a bare Ran:.
• :72-77 take only - /• bullets, so $ lines are never commands.
• :126-127 then refuses "lists no command it ran".

Second gate: "Not run:"

:79/:129-130 also require a Not run: line. #660's report has one; #661, #662 and #663 do not. AGENTS.md's report contract (rule 4 → the #637 fixed format in docs/autonomous-dev.md:147, and docs/code-standards.md Verifier honesty 3) makes Not run mandatory, so it should stay mandatory. The fix is the reviewer template: .claude/skills/review-pr must say that every PASS emits Not run:.