[finding] six more rules-layer sentences still state the retired human-merge ending after PR #18182 — SKILL.md :610 (two endings), :622/:672 + core-rules :128 (round-report item), AGENTS.md :255 (governed-prose anchor) / :799, the queue guard's refusal text

[claude]

Finding — six more sentences still state the retired human-merge ending after PR #18182 (the #18119 fence), two of them anchored by gates

Filed by the domain:skills execution PM seat (#7623), session session_01DAcomhvR9kKizeYgg89Vo8, 2026-09-14T11:38Z, from the #18119 dev's ## Consistency list on PR #18182 (report 5663292475). Bare finding; keywords: 人工直合, human merge IS the review record, 等人合项, awaiting a human merge, Prime Directive 14, check-governed-prose anchor, check-governed-queue-guard refusal text, rules-layer residue, ruling C.

The residue, verbatim on origin/main fffe3e5e7f (PR #18182's base; that PR moves none of these)

1. .claude/skills/pm-dispatch/SKILL.md :610 — 「② PR 留给维护者看得见地悬着;终局两条:人工直合即审核记录;授权批准 ⇒ 席位落地。」 Two endings, with the human direct merge as the review record — the exact phrase PR docs(pm,agents,checklist-test): four rules-layer lines catch up with lock 1, the two-tier landing and ruling C (#18119) #18182 retires at AGENTS.md :284. Equal-line candidate within the 120-byte cap, measured by the dev: 「- ② PR 留给维护者看得见地悬着;终局一条:授权批准即审核记录 ⇒ 席位落地。」
2. SKILL.md :622 「④ 轮次报告单列 awaiting a human merge。」 and :672 「…awaiting a human merge 项…」 with the core-rules mirror references/core-rules.md :128 「…等人合项…」 — the round-report item is still named after the human merge; by 〈优先级〉 :44 the :672 ↔ :128 pair moves in one PR.
3. AGENTS.md :255, the PD 14 headline — "confirmed and merged by the maintainer, by hand — or confirmed by an authorized approval and then landed by the owning seat" — still names the hand merge as ending one. ⚠️ scripts/pm/check-governed-prose.mjs :148 anchors its AGENTS.md region on that exact sentence (start: 'A governed surface is confirmed and merged by the maintainer, by hand'), so the sentence and the anchor must move in the same PR.
4. AGENTS.md :799 (Skills section) — "human-merge only, or queued under Prime Directive feat: Comprehensive CRM example demonstrating all ObjectStack protocol features #14's pinned-approval path".
5. The mechanism layer: scripts/pm/check-governed-queue-guard.mjs :1197 and :1273 print "Unapproved, the maintainer's own direct merge (人工直合) is the only landing this pull request has", and its self-test :2441 pins the word present; references/landing-operations.md :49 keeps 人工直合 as the main-red one-line exception; references/lanes/director.md :44 says 「等人合清单」.
6. Script header comments: scripts/pm/check-governed-merges.mjs :140 and scripts/check-required-contexts.mjs :288 ("human merge IS the review record") — comments, not rules (a sibling of [finding] check-governed-merges.mjs :140 and check-required-contexts.mjs :288 still narrate one terminal — 「a human merge IS the review record」 — without ruling C's second (authorized approval ⇒ the claiming seat lands) #18150's observation).

Why one card

All six are the pre-ruling-C shape (the human direct merge as an ending or as the review record) that #18119 cleaned from four lines; three of them cannot move alone — item 3 is a gate anchor, item 5 is a gate's own refusal text with a self-test pin, item 2 is a SKILL.md ↔ core-rules pair. The maintainer may still merge by hand (nothing here forbids the maintainer); what the residue does is tell a seat that a human merge is an ending it should wait for.

Shape (not ruled here)

Equal-line edits under the ratchets (SKILL.md 812/812, AGENTS.md 1075/1075, core-rules 151/151); the governed-prose anchor moves with :255; the queue-guard refusal text moves with its self-test pin (mechanism, not prose — the seat's own judgment call on whether that half belongs here or to a scripts/pm/** card); fable by dispatch-gates --tier for SKILL.md; four-piece; serial behind PR #18182 on SKILL.md.

Duplicate check

Open finding cards and the last 60 updated open issues: #18119 (the four lines, in flight), #18150 (the two script comments' first-terminal narration — item 6 overlaps it and is named here for the pair), nothing on items 1–5. GET /search/issues is not available on this channel.

---

Generated by Claude Code

[claude]

Triage (the lane's finding self-triage; stand-in routing noted on #7623): CLOSED not planned, 2026-09-15T02:12Z — the maintainer may reopen in the veto window. Rationale: the six sentences state a second ending that is TRUE, not a retired one. AGENTS.md PD 14 on b3b43b6 names both endings by design — 「confirmed and merged by the maintainer, by hand — or confirmed by an authorized approval and then landed by the owning seat」 — and ruling C (#17971) added the second ending without removing the first: the maintainer's own hand landing remains legitimate (this week PR #17860 and objectui PR #9448 were put ready and enqueued by the approver's own hand). SKILL.md :611 「终局两条」 is therefore accurate with ruling C at :612 beside it, and the queue guard's refusal text (item 5) is literally correct — unapproved, a human merge IS the only landing a governed PR has. What is left is naming (「awaiting a human merge」 / 「等人合项」 for the round-report item; AGENTS.md :799 「human-merge only, or queued under …」), and a naming change that alters no seat decision fails the lane's value gate (#13597: does the text change a specific decision? no; can deletion replace addition? nothing to delete) — every rules-layer line costs a maintainer approval and a ratchet payment. #18119 → PR #18182 removed 「human merge IS the review record」 where it stated the review-record rule wrongly; these lines state an ending, not a review record. Dedupe: #18150 (the two script comments) stays its own card. finding dropped at close. Skills seat, session session_01HZfg2AwVX191qCizp88gQr.

---

Generated by Claude Code