docs(pm-dispatch): review checklist names the renamed Lint & Repo Gates check run (#9420)

[os-zhuang]

Fixes#9420

Merge posture — still DRAFT, and still human-merge only

This PR stays DRAFT. Do not mark it ready, do not enable auto-merge, do not enqueue it. This surface is human-merge only (Prime Directive #14, skills PRs). Nothing in this PR was armed, queued, or flipped.

Ordering: satisfied. This section originally read "do not merge before #9421" — that was correct when this PR opened and is now stale, so it is corrected rather than left standing. #9421 landed as 81316864b, which is the commit immediately after this branch's base ad217b192. main now carries the rename:

$ git show origin/main:.github/workflows/lint.yml | grep -n "name: Lint & Repo Gates"
39: name: Lint & Repo Gates

So the change in this PR is now correct rather than premature — the "wrong in the other direction" window the card warns about has closed.

mergeable_state: blocked is expected and is not a defect. This branch's head predates the rename by exactly one commit, so its own lint.yml still names the job ESLint. The standing remedy is one branch update, the same as for every open PR whose head predates the rename. This seat deliberately did not perform it — the dispatch forbade chasing the state — so that decision is left to the merging human. No other PR was touched.

What changes

One line region in one file: .claude/skills/pm-dispatch/references/review-checklist.md.

The checklist tells the review seat, before flipping a PR ready / arming auto-merge / enqueuing it, to confirm the conclusion of two named jobs is success. One of those two is the job that carries the whole check:* gate family, and its check-run name moved to Lint & Repo Gates.

Before (lines 35-40):

- **CI 收敛读数只属于复核侧**(维护者 2026-08-10 裁定;dev 的契约是草稿 PR 时点交
报,报告里 gate `in_progress` 是诚实读数、预期内常态):翻 ready / 挂 auto-merge
/ 入队前亲核 ESLint 与 TypeScript Type Check 两个 job 的 `conclusion` 已为
`success`(门禁族跑在其内),⛔ 不因报告写了「本地绿」跳过。收敛期转红走补丁轮
(SendMessage 续派原 dev —— 那是这笔交换已付过的价钱,不是 REWORK 的理由;红着
合并才是)。重量级卡可在派发令显式写「本单等 CI」。

After (lines 35-40):

- **CI 收敛读数只属于复核侧**(维护者 2026-08-10 裁定;dev 的契约是草稿 PR 时点
交报,报告里 gate `in_progress` 是诚实读数、预期内常态):翻 ready / 挂 auto-merge /
入队前亲核 Lint & Repo Gates 与 TypeScript Type Check 两个 job 的 `conclusion` 已
为 `success`(必需检查认 check-run 名,改名前的 PR 仍列旧名 `ESLint`),⛔ 不因报告写
了「本地绿」跳过。收敛期转红走补丁轮(SendMessage 续派原 dev —— 那是这笔交换已付过
的价钱,不是 REWORK 的理由;红着合并才是)。重量级卡可在派发令显式写「本单等 CI」。

What is preserved, checked mechanically

A whitespace-insensitive diff of the bullet reports exactly three edits and nothing else:

delete OLD |ES| NEW ||
insert OLD || NEW |&RepoGates|
replace OLD |门禁族跑在其内|
NEW |必需检查认check-run名,改名前的PR仍列旧名`ESLint`|

Everything else is byte-identical modulo re-wrapping. Unchanged, deliberately:

• the verb — 亲核 ... 的 conclusion 已为 success, not "check", not "see green";
• the timing — 翻 ready / 挂 auto-merge / 入队前, all three, in that order;
• which jobs are covered — 两个 job, the gate-family one and TypeScript Type Check;
• the surrounding clauses (the 2026-08-10 ruling provenance, the "本地绿" prohibition, the patch-round rule, the 「本单等 CI」 escape).

Both job names are kept unbroken on a single line each, so a seat scanning for either literal finds it.

Why the phrasing is not the card's proposed phrasing

The card proposes mirroring **Lint & Repo Gates** (called ESLint until the #9325 rename). That exact sentence cannot land here: check:pm-skill-id-lint scans this file and forbids issue-ID citations in operative prose. Measured, not assumed — writing the card's parenthetical into this very line:

✗ check-skill-id-lint: .claude/skills/pm-dispatch/references/review-checklist.md: 1 issue-ID citation(s)
review-checklist.md:38: 为 `success`(必需检查认 check-run 名,旧名 `ESLint`(#9325 改名)),⛔ 不因报告写

exit 1. Same gate that went red on #9421 for the same reason. The sentence that actually landed in AGENTS.md is not the card's proposal either — it was rewritten there for this gate, and this wording mirrors what landed rather than what was proposed.

So the fact is distilled in place instead of pointed at: a required status check is matched by its check-run name, which is exactly why a PR opened before the rename still lists the old spelling. That is the operative content the seat needs at the moment it is reading a stale PR's checks list, and it is self-contained.

The one informational change, called out

The old gloss 「门禁族跑在其内」 is not carried over verbatim. It is folded into the new name, which now says it: the job is called Lint & Repo Gates precisely because the old name described one of its ~70 steps rather than the gate family it carries. AGENTS.md states the same gloss explicitly for the same required set.

This is not a stylistic preference — the file sits exactly at its ratchet ceiling: review-checklist.md is 82 lines against a ceiling of 82, headroom 0. The bullet had to absorb the longer job name plus the distilled fact inside its existing 6 lines. Wordings that kept the gloss too were measured and need a 7th line, which the ratchet rejects; raising a ceiling requires a maintainer ruling quoted in the raising PR, and there is none for this. Per the ratchet's own doctrine the change pays its way by compressing in place. The alternative considered and rejected was deleting rationale from the adjacent clauses of the same bullet — that would have cost instruction text rather than a gloss the new name now carries.

Line count after: 82. The ratchet does not move in either direction, and no ceiling is edited.

Grep sweep

$ grep -rn "ESLint" .claude/skills/
.claude/skills/pm-dispatch/references/review-checklist.md:37: ...亲核 ESLint 与 TypeScript Type Check...

One occurrence in the whole skills tree, and it is the one this PR changes. Nothing was left behind, and nothing meaning the ESLint tool was touched (there is no such occurrence under .claude/skills/). For contrast, lint.yml:81 on main is a step still legitimately named ESLint because it really is the ESLint tool — that one is the job's step, not the check-run name, and is correctly untouched by both PRs.

Adjacent references to the same job that do not spell the name were checked and correctly need no change — they already name the job by its role rather than its label:

• references/platform-readings.md:23 — 门禁放行判据 = 承载门禁族 job 的 conclusion
• references/landing-operations.md:32 — 承载门禁族的 job completed: success
• SKILL.md:124 — 承载门禁族的 job 须 completed: success

Those are rename-proof by construction and are the reason the blast radius is one line.

Gate union, derived not recalled

$ node scripts/pm/dispatch-gates.mjs .claude/skills/pm-dispatch/references/review-checklist.md

Six families derived from the changed path (the five the card names, plus check-adr-merge-approval via the .claude/skills gate source). Run at final commit 0ec6cfc64 (git rev-parse --short HEAD from that run), verbatim exit codes:

pnpm check:doc-authoring -> exit 0
pnpm check:pm-skill-id-lint -> exit 0
pnpm check:pm-skill-ratchet -> exit 0
pnpm check:skill-frame-sync -> exit 0
pnpm check:nul-bytes -> exit 0
pnpm --filter @objectstack/lint run check:doc-formula-expressions -> exit 0
node scripts/check-adr-merge-approval.mjs -> exit 1 (environmental, below)

check:nul-bytes is not path-derived; it is run on every edit by contract. 0ec6cfc64 is still the head — the only change since is this body correction, which moves no tree.

Selected output:

✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/review-checklist.md is 82 lines (ceiling 82; headroom 0).
✓ check-skill-id-lint: 9 file(s) clean (pattern /#[0-9]{3,}/g).
✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files
✓ doc authoring guard: 377 files clean — no bare metadata literals.
✓ check:doc-formula-expressions: 22 record-scoped formula example(s) across 396 files / 1411 TS blocks judged clean
check-nul-bytes: OK (scanned 6130 text file(s); no raw ASCII control bytes).

check-adr-merge-approval.mjs is red for a reason independent of this diff — it needs GitHub credentials this seat does not have, and this PR touches no ADR:

❌ GET https://api.github.com/repos/objectstack-ai/objectstack/commits/0ec6cfc6.../pulls answered HTTP 401:
{ "message": "Bad credentials", ... }
Missing input fails loud, never exit 0 (#4690, #4928).

CI runs it with a token. check:doc-formula-expressions also needed pnpm --filter '@objectstack/lint^...' build first (a fresh worktree has no built closure); it is green after that build, red before it for the same environmental reason.

Reverse verification

Predicted direction, stated before running: no gate notices this line's content at all. For an instruction-file change the honest reverse check is whether any gate reads what the line says, and the derived families check line counts, issue-ID patterns, TypeScript fenced blocks and escalation-frame anchors. None of them parses a check-run name.

Observed: exactly that. Replacing the job name in the line with Nonexistent Job Name and re-running the whole union:

pnpm check:doc-authoring -> exit 0
pnpm check:pm-skill-id-lint -> exit 0
pnpm check:pm-skill-ratchet -> exit 0
pnpm check:skill-frame-sync -> exit 0
pnpm check:nul-bytes -> exit 0
pnpm --filter @objectstack/lint run check:doc-formula-expressions -> exit 0

Green across the board. The tree was restored afterwards and git status --porcelain is empty.

This is a true reading about the surface rather than a manufactured green, and it is worth stating plainly: nothing mechanically binds this checklist line to the workflow's name: or to the pinned registry in scripts/check-required-contexts.mjs. That script's scan set is the workflow files; it does not read .claude/ at all. The rename is caught in lint.yml and in the registry, and it is caught in AGENTS.md only by a human reading it. This line was found by the dev seat that did the rename, not by a gate — which is precisely why the card exists.

The positive control above (the id-lint run) is the complement: the gates do read this file, they just do not read the job name. So the scan set is not the gap; the assertion is.

Filed as #9491, an observation-class card, rather than widened into this PR.

Changeset

skip-changeset per precedent: an agent instruction file publishes nothing. Same disposition as #9421 for its workflow and gate-script halves, and the same reasoning the checklist itself gives for tests/docs-only PRs in this repo (the label, not an empty changeset, which lingers in the release pipeline here).

---

Generated by Claude Code

[os-zhuang]

PM review — the compensating measure recorded on #9420, delivered.

This card is fable-mandatory under clause ① and ran at opus on a measured quota exhaustion. The exemption's stated price was that the PM seat reviews this diff against what the checklist actually instructs the review seat to do, not against the card's description of it. That review follows. Also answering the one red check.

The red ADR maintainer approval is correct — no action

One path hit governs the whole PR -- a mixed diff is never judged by proportion.
Drafting and pushing this PR was fine and stays fine -- only the MERGE is gated.

That is #9407's widening of GOVERNED_SURFACES to .claude/skills/, firing on a real skills PR for the first time since it landed tonight. The gate is the mechanical form of Prime Directive #14, and this PR is exactly what it exists to hold. It stays red until a maintainer approves; nothing here should try to make it green.

Worth recording that the surface-specific wording came through intact on a live diff — that clause was added on the assumption it would read correctly in production, and it does.

Instruction-preservation check — passed

Read against the current text of the line rather than the card's summary of it, the four things that could have silently moved:

	before	after
verb	亲核 … 的 conclusion 已为 success	unchanged	✅
timing	翻 ready / 挂 auto-merge / 入队前	unchanged, all three, same order	✅
coverage	两个 job — gate-family + TypeScript Type Check	unchanged	✅
surrounding clauses	2026-08-10 provenance, 「本地绿」 prohibition, patch-round rule, 「本单等 CI」 escape	unchanged	✅

The one substantive substitution is 「门禁族跑在其内」 → 「必需检查认 check-run 名,改名前的 PR 仍列旧名 ESLint」. I accept it, and the reasoning matters more than the verdict: those are two different facts, and dropping one to gain the other is only sound because the new job name carries the dropped one. Lint & Repo Gates says the job runs repo gates; that is what the old gloss said. Had the chosen name been Lint or ESLint 2, this substitution would have been a real loss of instruction and I would have sent it back.

The ratchet constraint (82 lines against a ceiling of 82, headroom 0) is a genuine forcing function and was handled in the right priority order — a gloss the name now carries was given up rather than instruction text from the adjacent clauses. Raising a ceiling needs a maintainer ruling quoted in the raising PR and there is none, so compressing in place was the only honest route. No ceiling was edited and the ratchet does not move.

The reverse verification is the most valuable thing in this PR

The finding, stated plainly by the dev seat rather than papered over:

nothing mechanically binds this checklist line to the workflow's name: or to the pinned registry in scripts/check-required-contexts.mjs. That script's scan set is the workflow files; it does not read .claude/ at all.

With the positive control (check:pm-skill-id-lint reads this file and is green) the diagnosis is sharp: the scan set is not the gap — the assertion is. The gates read this file; none of them reads a check-run name out of it.

That is why this card existed at all. It was found by the human-eye sweep of a dev seat doing the rename, and the next rename gets no such luck unless someone is again renaming by hand. I am treating the follow-up card as the durable output of the whole #9325 sitting, above this one-line fix.

Merge posture — unchanged and enforced twice over

Stays draft. #9421 has now landed (81316864), so the ordering constraint in the header is satisfied and this PR's base is already the post-rename main — but it is human-merge only regardless, and the ADR gate now enforces that mechanically rather than by instruction. No agent seat merges this.

---

Generated by Claude Code

[os-zhuang]

Correction round (takeover, adjudicated) — the diff is now a pure rename.

Maintainer ruling applied, verbatim and untranslated: 「临时性的问题不应该进入 skills 啊」. The transition-window parenthetical (必需检查认 check-run 名,改名前的 PR 仍列旧名 ESLint) was shelf-life knowledge and has been removed; the bullet is restored to origin/main's text byte-for-byte except the job name: ESLint → Lint & Repo Gates. The original gloss 「门禁族跑在其内」 is preserved. Authorization and rationale: the Adjudication comment on the card this PR fixes (2026-08-18 05:37 UTC).

Executed by the second seat under that adjudication, on this same branch (head 6ed8b4a58); the PR body above describes the superseded wording — this comment and the latest commit are current. Gate union re-run green at the new head (doc-authoring, pm-skill-id-lint, pm-skill-ratchet 82/82, skill-frame-sync, nul-bytes, doc-formula-expressions; adr-merge-approval red locally with HTTP 401 as designed, CI carries the real reading). PR stays draft, human-merge only.

---

Generated by Claude Code

---

Generated by Claude Code

[os-zhuang]

Correction to my review above — its central paragraph is superseded, and the maintainer's ruling is better than my reasoning was.

My review of head 0ec6cfc6 accepted the substitution of the durable gloss 「门禁族跑在其内」 by the transition parenthetical 「必需检查认 check-run 名,改名前的 PR 仍列旧名 ESLint」, on the grounds that the new job name carries the dropped gloss. The maintainer has since ruled, verbatim: 「临时性的问题不应该进入 skills 啊」 — recorded in the adjudication on #9420 at 05:37Z.

That ruling is right, and it exposes the actual miss in my review. I treated the ratchet as a forcing function and graded the trade it forced — durable gloss out, transitional fact in — as acceptable because the name absorbed the loss. What I did not ask is the prior question: does transition-window knowledge belong in permanent skill text at all? It does not. Its shelf life is days; the file's is indefinite. Asking that question dissolves the constraint instead of paying it — a pure rename needs no seventh line, so the ratchet never binds and the gloss never has to go.

A reviewer who validates the trade a constraint forces, without checking whether the constraint should have been entered, is doing the weaker half of the job. That is the correction worth recording, not the wording.

The new head verified — the compensating measure still applies, and is now trivial to satisfy

The compensating measure recorded on #9420 was mine to deliver regardless of which seat wrote the code, so I re-read the diff at 6ed8b4a5 rather than accepting the commit message's claim:

- / 入队前亲核 ESLint 与 TypeScript Type Check 两个 job 的 `conclusion` 已为
+ / 入队前亲核 Lint & Repo Gates 与 TypeScript Type Check 两个 job 的 `conclusion` 已为

One line, 1 addition / 1 deletion, down from 6/6. Everything else byte-identical — including (门禁族跑在其内), the verb 亲核 … conclusion 已为 success, the timing 翻 ready / 挂 auto-merge / 入队前, and both covered jobs. A pure rename preserves the instruction by construction, which is a stronger guarantee than the four-way comparison my earlier review had to perform by hand. Line count is unchanged at 82, so the ratchet is untouched in either direction.

The ADR maintainer approval check is red on this head for the same correct reason it was red on the last one — governed surface, no maintainer approval yet. Unchanged, and still not something to work around.

One process note, recorded once and not pursued

The 05:37Z adjudication finds its own 04:28:42Z claim void and mine valid and first, then executes on this branch anyway. Those two don't sit together: a void claim isn't made valid by carrying a ruling — the ruling would normally be handed to the claim holder to execute. I'm recording that and leaving it there. The maintainer's ruling is satisfied either way, the diff is smaller and more verifiable than what I approved, and re-doing correct work to settle whose branch it was would cost the repo something and gain it nothing.

No change to the merge posture: draft, human-merge only under Prime Directive #14, now also mechanically gated.

---

Generated by Claude Code