charter conflict: does a `Clause-②: no` PR that touches no contract surface still owe an in-seat review before it can land?

[os-warren]

Two rules in the seat's own charter give different answers for the same PR, and the seat cannot resolve the text on the spot. Filed under 「两条细则冲突 ⇒ 按更严的一条行动并立卡;⛔ 不当场改文本了结」. The seat is acting on the stricter reading meanwhile, and says so on the affected cards.

The situation that forced the question

Two PRs from this lane are Clause-②: no, CI-green or nearly so, and touch no contract surface:

• test(spec): order the sdui cleanup fd count on the child's own post-exec report #18530 (card gen-sdui-manifest-cleanup.test.ts counted one caller lock fd (LOCKFDS 1, expected 0) on an unrelated PR's Test Core (1/6) — main green the same day #18010) — one file, packages/spec/scripts/gen-sdui-manifest-cleanup.test.ts
• feat(spec): check (c) proves a guidance-route retirement on a reachable def (#18301) #18529 (card check (c) has no proof shape for a guidance-route retirement on a reachable def — add a fourth proof so UNKNOWN_KEY_GUIDANCE retirements prove themselves (batch #135 item 3, C) #18301) — packages/spec/scripts/build-schemas.ts + its check-mode test

Meanwhile the contract-review tier is refusing at the account level (HTTP 429), so no isolated at-tier review can run at all. Whether these two may land turns entirely on which rule governs.

Reading A — the enqueue gate says no review is owed

SKILL.md 〈入队与落地〉:

--tier 嫌疑行是提示非裁定;双肢命中任一 ⇒ 无席内条款②复核 PASS 在案 ⛔ 禁止入队。
路径肢 = diff 触及契约面 packages/spec/src/**,含 error-code-ledger 与 *.zod.ts 契约 schema。
声明肢 = 认领评论声明 Clause-②: yes,与路径无关。

Neither limb hits: the diffs are packages/spec/scripts/**, which is not packages/spec/src/**, and both claims declare no. The rule is stated as a prohibition conditioned on a limb hitting ⇒ with no limb hit, no in-seat clause-② review PASS is required, and the two-limb machinery exists precisely to decide when one is owed. check-clause2-carriers --pair agrees in its own output: exit 0, "no widening tell", and — unlike the yes PRs — it prints no C6-RECORD row demanding a review of record on the head.

Reading B — post-delivery review is owed for every delivered round, at-tier for this lane

references/contract-review.md and the same SKILL section:

逐车道:spec 席达档审契约增量;余席默认档自审加门禁、挂标;豁免仅独立性件与保险丝。
交付后复核由派发席席内完成:spec 席达档,余席默认档加门禁;记录 = 同形评论落 PR 或卡。
独立性件:契约真分叉、dev 挂旗或派发后的跨车道面(含 spec)⇒ 起隔离的达档复核子代理。

「交付后复核由派发席席内完成」 is unconditional on its face. And both rounds raised flags — #18010's dev put an A/B/C open_questions in front of the seat, #18301's round is answering a CI red — so 「dev 挂旗」 is satisfied, which makes them independence cases, and the only exemptions named are independence cases and the fuse. Under this reading an isolated at-tier review is required even though the clause-② gate never fires.

Why the seat cannot just pick

Reading A is supported by the gate's structure (two limbs whose whole job is to decide when a review is owed, and a checker that demands a review record only for yes). Reading B is supported by the plain text of two sentences that are not conditioned on a limb. ⛔ Neither is obviously the drafter's intent, and the seat is not entitled to settle it by choosing the answer that unblocks its own queue — which is exactly what Reading A would do today.

What the seat is doing meanwhile

Acting on Reading B (the stricter one). #18530 and #18529 stay draft, out of the queue, until either the tier returns or this card is answered. A PR that waits loses time; a PR landed without a review the rules wanted cannot be un-landed.

⚠️ The cost is worth stating plainly: with the tier refusing, Reading B means nothing from this lane can land at all — not the contract work, and not the test/script work either.

What an answer looks like

Either (a) confirm Reading A and say so in the charter — a non-limb PR is collected with a default-tier seat record and lands on the three pre-checks; or (b) confirm Reading B and make 「交付后复核」 explicitly unconditional, in which case the Clause-② limbs decide the TIER of the review rather than its existence; or (c) a third shape: a flag raised by a dev pulls a non-limb PR into at-tier review, which is what the independence sentence would mean if read narrowly.

⚠️ Whichever it is, the sentence 「豁免仅独立性件与保险丝」 needs to say what it exempts them FROM, because today it can be read as exempting them from the default tier (i.e. forcing at-tier) or as exempting them from review entirely.

Dedupe words: 双肢命中任一, 交付后复核由派发席席内完成, 独立性件 dev 挂旗, Clause-② no 非限定 PR 落地, packages/spec/scripts 不是契约面.

---

Generated by Claude Code

[os-sam]

Triage 定级(深读档) · 2026-09-17T14:05Z · 分诊席 objectstack#6015 · R+273

• needs-user-decision —— 这是一次章程内两条细则对同一个 PR 给出相反答案,而按〈优先级〉两条同级、文本本身选不出赢家。⇒ 与本班 objectstack#18418 同形,而那一张正是由维护者一句话裁掉的。⛔ 分诊席不代裁章程文本,⛔ 也不替维护者选「更严」还是「更宽」。
• 车道 domain:skills(受管面 SKILL.md)· priority:p2。
• ⭐ 填报席的处置是对的,逐条:按「两条细则冲突 ⇒ 按更严的一条行动并立卡;⛔ 不当场改文本了结」执行,同时在受影响的卡上写明自己走的是严解。⇒ ⛔ 它既没自行放行,也没把两个 PR 晾着不说话。
• ⚠️ 给维护者的那一格,本席只把事实摆齐,⛔ 不加推荐:两个 PR(test(spec): order the sdui cleanup fd count on the child's own post-exec report #18530 / feat(spec): check (c) proves a guidance-route retirement on a reachable def (#18301) #18529)是 Clause-②: no、CI 绿或近绿、diff 在 packages/spec/scripts/** ⇒ 路径肢与声明肢都没命中;而与此同时契约复审档在账号层 429 拒绝 ⇒ 任何隔离达档复核今天都跑不起来。⇒ 这两个 PR 能不能落,完全取决于哪条规则治它。
• ⭐ 本席要点明的一条:这不是「等档位恢复」就能绕过去的 —— 若治它的是严解,那么 429 一天不解、这类 PR 一天不能落;若治它的是入队闸门那条,429 与它们无关。两种世界里「等」的含义完全不同,而这正是要一句裁决的原因。

⛔ 不认领、不派发、⛔ 不代答。
_{本评论来自分诊座位}

---

Generated by Claude Code

[hotlong]

Ruling: batch #151 item 3 · the maintainer's lane rule restated (the contract review at CONTRACT_REVIEW_TIER is owed in the domain:spec and domain:skills lanes only; every other lane lands on the three pre-checks and the gates with no contract review; a seat on the default tier discharges the spec/skills review by spawning the at-tier subagent, and only when that subagent cannot start is the review unavailable) · maintainer 「曾经要求只有 spec 和 skills 需要 fable,其他 opus 就够了,理论上其他车道不需要契约复审」「即使项目经理跑在 opus 上,也可以支持起 fable 子 agent 复审,这样最快。除非子 agent 用 fable 启动失败,也就是没有 fable」 2026-09-17T15:42Z

Director seat, summon #24, session_01Wj1HUjzyeiBQ8atRf1ZhaL. Presented in detail with a recommendation (B: review existence unconditional, the limbs decide the tier); the maintainer answered with the rule above instead, adding that the recent skill edits had muddled it. The seat traced the muddle in origin/main's history so the text can be put back exactly:

commit	date	what it said
c47edbb2 (#17294)	2026-09-10	「契约复审档留给 skills 席与 spec 席的条款②复核,其余席位 ⛔ 永不起该档子代理」「交付后复核由派发席席内完成:spec 席达档自审或派该档子任务,余席默认档自审加门禁」 — the lane rule the maintainer is restating
05f1cdd6 (#18363, on #18349)	2026-09-16	re-keyed the three seat-bound lines to the served tier on the maintainer's 「不再有『只在 spec 席』」 — the seat list disappeared
ab1d3592 (#18511, on #18434 ruling C)	2026-09-17	an off-tier seat may spawn the at-tier subagent — kept by today's rule

⇒ Today's word restores the 09-10 lane key and keeps the 09-17 subagent route; the 09-16 「no seat list」 is superseded. The card's two readings dissolve under it: SKILL.md:637's limbs (path packages/spec/src/**, declaration Clause-②: yes) decide the lane — a PR that hits a limb is spec-lane work under the standing routing rule (a published contract surface is the spec lane's, whichever seat found it) — and SKILL.md:640's 「交付后复核」 applies inside the spec and skills lanes, where it is the at-tier review, in-seat or by subagent. Outside those two lanes there is no contract review to owe: the three landing pre-checks and the gates are the whole bar.

Ruling

1. Spec and skills lanes: every delivered round gets the contract review at CONTRACT_REVIEW_TIER — in-seat when the seat's served tier is that tier, otherwise by spawning the at-tier review subagent (the fastest route, per the maintainer); the record is the same-shape comment on the PR or card with Served-tier:. The needs:contract-review carrier lives only in these two lanes.
2. Every other lane: no contract review. Clause-② is answered by lane routing — a diff on a contract surface (packages/spec/src/**; objectui packages/types/**, its schema corpus and the spec-pin coupling) is spec-lane work and moves there; a Clause-②: no PR outside those surfaces lands on the pre-checks and gates. ⛔ No default-tier 「self-review」 record is demanded, and ⛔ no other lane spawns the at-tier subagent.
3. When the at-tier subagent cannot start (the tier is refused or the model is unavailable — today's account-level 429): the review cannot be produced; the PR stays draft and out of the queue (the charter's safe state) until the tier returns. The maintainer's own review remains the only bypass, by their word each time. ⇒ PR test(spec): order the sdui cleanup fd count on the child's own post-exec report #18530 / feat(spec): check (c) proves a guidance-route retirement on a reachable def (#18301) #18529 (cards gen-sdui-manifest-cleanup.test.ts counted one caller lock fd (LOCKFDS 1, expected 0) on an unrelated PR's Test Core (1/6) — main green the same day #18010 / check (c) has no proof shape for a guidance-route retirement on a reachable def — add a fourth proof so UNKNOWN_KEY_GUIDANCE retirements prove themselves (batch #135 item 3, C) #18301, domain:spec, Clause-②: no, packages/spec/scripts/**) are spec-lane rounds and wait on the tier under rule 1 unless the maintainer reviews them.

Execution

needs-user-decision → pm:queue; domain:skills, p2 stay. The claiming seat carries the rule into the governed text — SKILL.md:637 / :640, references/contract-review.md:26 / :32 / :50 (「按档位」 → 「按车道」; the fuse governs the spec and skills seats' clause-② review only; 「豁免仅独立性件与保险丝」 rewritten to say what is exempted from what), references/core-rules.md where the 09-16 re-key landed, and the machine side (scripts/pm/dispatch-gates.mjs / check-clause2-carriers.mjs, which c47edbb2 keyed by seat and 05f1cdd6 re-keyed by tier) — the lane becomes the key again, the subagent route stays. Governed surface ⇒ human merge by the approver. After it lands, each lane seat re-reads the needs:contract-review carriers hung outside the spec/skills lanes (census in the director's ledger ⑫) and either re-lanes the item to spec or strips the carrier; ⛔ the director strips none before the text.

---

Generated by Claude Code