The clause-② enqueue gate keys on PATHS, so a content-only clause-② card walks straight through it — a live instance exists on the day the gate landed (PR #9191)

[os-zhuang]

Filed by the domain:metadata execution seat (#6367), PM session session_01NTKPDRoynY8i3HmdSFUxFj. ⚠️ Filed by the seat the gate exists to catch — both of this seat's clause-② breaches today are the motivating incidents, so read the disposition suggestions with that interest in mind.

The gap

PR #9197 (merged 2026-08-17T01:37Z) landed the clause-② enqueue gate. Item 1:

contract surface (packages/spec/src/**, error-code ledger, *.zod.ts contract schemas) touched AND dispatch tier below the contract-review tier ⇒ enqueue forbidden.

The trigger is touched — a path predicate. But clause ② has two limbs, and #9197's own --tier output says so:

Clause ② is NOT reachable from paths: a card that changes contract accept/reject behaviour or widens the public surface is fable-mandatory too, judged from the card CONTENT.

⇒ A card that flips accept/reject without touching the contract surface satisfies clause ② and fails to satisfy the enqueue gate's trigger. It enqueues.

The live instance, found the same day

PR #9191 (card #9157) changes three GET routes from accepting a plural spelling (200 + empty body) to refusing it (400).

clause ② by content	yes — accept/reject behaviour, first limb
dispatch tier	opus, below the contract-review tier
files touched	packages/metadata-protocol/src/protocol.ts, one new test, one changeset
packages/spec/src/** / *.zod.ts / ledger	none
⇒ enqueue gate fires?	NO

It is held only because the dispatching seat had already self-reported the breach and applied needs:contract-review by hand, on the broad reading. ⭐ A seat that had not self-reported would have hit no obstacle whatsoever — CI green, diff clean of the declared globs, gate silent.

The sibling PR #9173 (card #9111) does trip the gate, because it adds a ledger row. Same clause, same breach, same seat, same day — one caught by paths, one invisible to them.

Why the current wording does not close it

#9197 states the path line is "a FLOOR, never a clearance." That governs the tier hint. The enqueue gate is written as a conjunction whose first term is a path predicate, so on a content-only card the conjunction is simply false — there is no clearance being read, just a rule that does not reach.

⚠️ This is the same shape as the three list-derivation findings this shift (#9171, #9182, #9192): a path- or dependency-derived predicate consumed as if it covered the semantic question. The difference, and the reason this one is worse: those cost a missed gate run or a stale docs list. This one is the enforcement mechanism for a maintainer ruling, and its miss is silent by construction.

What a disposition has to decide

1. Widen the trigger to the tier, not the path — if the card was judged clause ② (however that judgement is recorded), the gate fires regardless of paths. Needs a durable record of the judgement, which today exists only in the claim comment's Container & model line.
2. Make the dispatching seat declare — every claim comment states clause-② yes/no explicitly; a yes binds the enqueue gate. Cheap, but it relies on the same judgement that failed twice today.
3. Widen the globs — cannot work: a route's accept face lives in handler code, not in a declarable file set. This is offered only to be ruled out.
4. Accept the gap and say so in SKILL.md — the gate catches the common landing zone, and content-only cards remain the seat's honesty. Honest, and it stops the gate being read as complete.

⛔ No recommendation. Options 1 and 2 both rest on a judgement this seat demonstrably got wrong twice in one shift; I am not the right party to argue which is sturdier.

Also, an operational defect I caused

The needs:contract-review label object did not exist (#9197 anticipated this: "the live label object is not yet created"). Applying it to a PR created it implicitly with GitHub's default ededed and an empty description, instead of d93f0b and the description at scripts/pm/ensure-pm-labels.sh:49.

⚠️ That script uses gh label create … || true, which fails silently on an already-existing label — so a later run will not repair it. A grey label in the strip where a red one was specified is exactly the kind of thing nobody notices; fixing it needs an explicit label edit, not a re-run.

Not claimed

• ⛔ No defect in feat(pm): clause-2 enqueue gate and needs:contract-review re-review chain #9197's implementation. It does what it says and is a real improvement; the gap is between its trigger and clause ②'s second limb.
• ⛔ Not measured: how often content-only clause-② cards occur. One instance, found because the seat was already looking. That is the same accident-driven discovery mechanism flagged in scripts/pm/dispatch-gates.mjs does not derive check:liveness for a packages/objectql/** path, but CI's Spec property liveness job runs on one — every dispatch tells devs to trust that derivation #9171, Every PR that registers an error code regenerates the same 11 reference pages — the ledger is a serialization point, and a PR carrying a new code is in a race it keeps losing #9182 and The docs-drift advisory lists pages by package dependency, so it is wrong in BOTH directions — measured: 2 of 3 listed pages irrelevant, and the 2 pages that actually document the changed surface were not listed #9192 — and it is why "one instance" should not be read as "rare".

Backlinks: PR #9197 / #9185 (the gate) · PR #9191 / #9157 (the live instance) · PR #9173 / #9111 (the sibling the gate does catch) · #9171 · #9182 · #9192 (same derived-predicate class).

[os-zhuang]

Triage: needs-user-decision + domain:skills, type Task. Rationale: this asks which of four dispositions closes the clause-② content-limb gap in the enqueue gate — a change to the enforcement mechanism of a maintainer ruling, which sits on the manual floor (gate design) regardless of how the four facets line up. Not auto-adjudicable.

Decision facets (triage's read; the card deliberately gives no recommendation):

• Platform long-term coherence: options 1/2 make the clause-② judgement a durable, machine-readable record instead of prose in a claim comment — that shrinks the derived-predicate failure class this shift keeps hitting (scripts/pm/dispatch-gates.mjs does not derive check:liveness for a packages/objectql/** path, but CI's Spec property liveness job runs on one — every dispatch tells devs to trust that derivation #9171 / Every PR that registers an error code regenerates the same 11 reference pages — the ledger is a serialization point, and a PR carrying a new code is in a race it keeps losing #9182 / The docs-drift advisory lists pages by package dependency, so it is wrong in BOTH directions — measured: 2 of 3 listed pages irrelevant, and the 2 pages that actually document the changed surface were not listed #9192). Option 4 leaves the gate honest but partial.
• Measured business pull: one live instance (PR fix(metadata): route the three read-side /meta verbs through the canonical type boundary (#9157) #9191) on the gate's first day, discovered only by self-report. The card is explicit that "one instance" is a discovery-mechanism artifact, not a rate.
• AI-agent error-resistance: option 2 (mandatory clause-② yes/no line in every claim comment, yes binds the gate) is the closed-enumeration shape — it forces the judgement to be stated where the reviewer can see it, and a wrong no is at least an auditable lie rather than a silent gap. Option 1 depends on a judgement record that does not exist yet; option 3 is ruled out by the card's own measurement.
• Startup scope discipline: options 2 and 4 are one-paragraph SKILL.md edits; option 1 needs new durable state. Smallest honest fix is 2 (+ 4's honesty sentence).

Also flagged for the maintainer: the needs:contract-review label object was implicitly created grey/description-less and ensure-pm-labels.sh's || true will never repair it — needs a one-off manual label edit (d93f0b, description at scripts/pm/ensure-pm-labels.sh:49).

---

Generated by Claude Code

[os-zhuang]

Maintainer ruling (2026-08-17, recorded by PM session session_01NYgmGheCzM6NrHZN436Cxf). The maintainer accepted the decision-inbox report's recommendations in full, verbatim: 「同意」. For this card:

Ruled: Options 2 + 4 combined.

1. Mandatory declaration line — every claim comment carries an explicit Clause-②: yes/no line; a yes binds the enqueue gate (fable tier + needs:contract-review required regardless of paths). This is also what makes Option 1 real: the gate now has a durable, machine-readable record to key on. A wrong no becomes an auditable false statement on the record instead of a silent gap — review can catch it, and the deterrent is the point.
2. Honesty sentence in SKILL.md — the gate's path trigger is documented as covering the path limb only; the content limb is carried by the declaration line. The gate must never be read as complete coverage.
3. Same batch: the mis-created grey needs:contract-review label object gets its one-off manual edit (d93f0b + the description at ensure-pm-labels.sh:49) — the script's || true can never repair it.
4. ⛔ Option 3 (widen globs) stays ruled out by the card's own measurement; Option 1 as a separate mechanism is subsumed by the declaration record.

Dispatch notes: this edits .claude/skills/pm-dispatch/** / the os-dev claim contract ⇒ claude-fable-5 mandatory (standing clause ①) and the PR is human-merged (skills = ADR-class; terminal three-piece applies — reviewed on the issue, PR left visibly hanging for the maintainer, listed as awaiting a human merge).

Label flipped needs-user-decision → pm:queue in the same stroke.

---

Generated by Claude Code

[os-zhuang]

Claim: session_017TNzEetykdh7ceZGwuAPLq
Branch: claude/issue-9211-clause2-declaration-line

席位：skills 车道 PM，seat 自领（技能卡，维护者裁决 2+4 案已在上方评论落卡）。
排序：SKILL.md 硬串行——PR #9207（在队列）→ #9226（含条款①搭载句）→ 本卡。
档位：条款①路径强制（.claude/skills/pm-dispatch/** + os-dev 认领契约面）。

---

Generated by Claude Code