pm-dispatch: `needs:contract-review` hangs on BOTH the PR and the card — symmetric hang and clear, so the gate is visible in the PR list

[huangyiirene]

Filed by the skills seat (session_01ApyDuQY2fkunMCqXiqvBhR) recording a maintainer ruling from live PM chat, 2026-08-22.

Blocked-by: #10822

(Serialization only, not a semantic dependency: both cards edit the same hot file, .claude/skills/pm-dispatch/SKILL.md's 复审链 bullet, and the fold-or-serial answer on that file is serial. When #10822's PR lands and closes it, this card returns to queue.)

The ruling (maintainer, verbatim, untranslated)

需要复审的 pr 是否应该直接打上 label,这样我在列表中能看出来。也可以直接要求复审。

简化一点是否可以直接挂 PR 侧
两边都挂好

Context — the measured inconsistency this fixes

Practice on 2026-08-21 was carrier-by-habit: the dispatching seat hung needs:contract-review on PR #10800 and its card, but on cards #10716/#10759/#10701 the label went issue-side only. The review sub-round therefore queries both sides defensively, and the maintainer's PR list showed nothing for three PRs that were in fact gated. The clearing side already runs the four-step read-modify-write against both carriers (a PR-side leftover was caught and cleared on #10800); the hanging side is what this ruling makes symmetric.

What the SKILL.md edit does

In the 复审链 / 条款②闸门 bullets of .claude/skills/pm-dispatch/SKILL.md: the parking act hangs the label on both the PR (as soon as it exists) and the card, in the same stroke; a PASS clears both in the same stroke, each side by the standing read-modify-write hard steps. The label remains the gate carrier exactly as the 2026-08-18 载体不迁移 ruling holds — this changes where it is visible, not what enforces it. A dev report arriving before the PR exists parks issue-side first and the dispatching seat completes the PR side at ACCEPT.

Constraints: the file sits at its line-ratchet ceiling (682/682 after the pending clause-② PR) — the wording must be paid for by tightening elsewhere in the same file; id-lint (no issue numbers in operative text); ruling carried as date + verbatim quote; governed face ⇒ draft PR, human merge; protocol-semantics face ⇒ fable dispatch per clause ①.

Discussed, deliberately not commissioned here

The maintainer's first message also raised requesting review natively on the PR (「也可以直接要求复审」) as a notification channel. Left out of this card's scope pending an explicit shape ruling (who the requested reviewer account is, and whether the request is notification-only or the maintainer wants to revisit the 2026-08-18 decision that the gate itself does not migrate to PR review). If ruled later, it is a separate one-sentence addition.

[huangyiirene]

Maintainer ruling recorded (2026-08-22, decision-inbox digest with the skills seat, session_01ApyDuQY2fkunMCqXiqvBhR; batch acceptance verbatim: 「接受所有」): the request-review notification channel is commissioned — notification-only, the gate stays on the label (2026-08-18 载体不迁移 holds). Scope addition for this card's dispatch: when parking needs:contract-review, the dispatching seat also requests review on the PR from the reviewing seat's account; the reviewing seat clears the request (or leaves a review comment) on PASS. One-sentence addition to the same SKILL.md bullet, same ratchet economics.

---

Generated by Claude Code

[huangyiirene]

Unblocked by the skills seat (session_01ApyDuQY2fkunMCqXiqvBhR): the serialization blocker is gone — PR #10936 merged to main (closing #10822), so the 复审链 bullet of .claude/skills/pm-dispatch/SKILL.md is free for this card's edit. Note for the implementer: the clause-② relaxation is now in the file and the ratchet ceiling reads 682/682, so this card's wording must be paid for by tightening elsewhere in the same file, as the card states. The request-review notification channel commissioned by the 2026-08-22 ruling (see the prior comment) is in scope.

State: pm:blocked removed → pm:queue.

---

Generated by Claude Code

[os-zhuang]

Claim: PM loop round 1 (family chain head)
Session: session_01MsbKEG4LtERSLaDrbehM3e
Branch: claude/issue-10972-skillmd-protocol-family
Worktree: objectstack-issue-10972
Domain: domain:skills
File surface: .claude/skills/pm-dispatch/SKILL.md (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: claude-fable-5 — this round's node scripts/pm/dispatch-gates.mjs --tier .claude/skills/pm-dispatch/SKILL.md: "Model tier — MANDATORY: claude-fable-5 … clause ① of the model-tiering ruling"
Clause-②: no
Serial constraints cleared: SKILL.md hot-file queue was empty at takeover (seat post #7623); prior chain occupants #10822 (PR #10936, clause-② relaxation) and #10837 (UNRECOGNISED rows) are MERGED on origin/main (fb0dd00, 35ad326) — this family starts from latest main. No in-flight claim in any lane names this file (skills-lane scan + spec-lane scan, both zero assignees, 2026-08-22T15:4xZ). #10190 shares this file and is explicitly serialized BEHIND this PR (see exclusion list).

Family dispatch — fold-or-serial answered: FOLD. Members #10972 + #10719 + #10313; one dev, one worktree, one PR. Maintainer provenance: live PM chat 2026-08-22, the maintainer asked for combined dispatch of the skills changes and accepted this seat's fold plan.

• Gate ① same defect shape & fix: all three are "SKILL.md protocol text lags a standing ruling/reality — amend the named bullet, paying the ratchet"; not keyword grouping.
• Gate ② same file/area: this file sits at its 682/682 ratchet ceiling — folding lets one dev balance the whole file's add/cut ledger once, instead of three sequential devs each re-fighting it.
• Gate ③ every member adjudicated/graded: this card carries its 2026-08-22 maintainer ruling (incl. the commissioned request-review sentence); pm-dispatch: the sweep's "sister repos have no lane labels" exemption is stale after the objectui three-stream ruling — a queue card without a stream label is invisible to every seat's candidate query #10719 and [finding] A seat restating clause ② on a card reads as a STRONGER floor than clause ② itself — the quota-exhaustion exemption's scope is unstated at the point of use, and it parked two cards for ~11 hours #10313 are triage-graded queue cards with no open decision.
• Gate ④ per-member verifiable, one commit per member, PR body carries one Fixes #<n> line per member:
  • pm-dispatch: needs:contract-review hangs on BOTH the PR and the card — symmetric hang and clear, so the gate is visible in the PR list #10972 — the 复审链/条款②闸门 bullets hang AND clear needs:contract-review on both the PR (as soon as it exists) and the card in the same stroke, plus the ruled one-sentence request-review notification addition.
  • pm-dispatch: the sweep's "sister repos have no lane labels" exemption is stale after the objectui three-stream ruling — a queue card without a stream label is invisible to every seat's candidate query #10719 — sweep disjunction ② sweeps the "queue without lane label" shape on every repo that HAS stream labels (main repo + objectui); the exemption survives only for repos genuinely without lane labels (cloud, objectos). ⛔ Do not widen the sweep to cloud/objectos.
  • [finding] A seat restating clause ② on a card reads as a STRONGER floor than clause ② itself — the quota-exhaustion exemption's scope is unstated at the point of use, and it parked two cards for ~11 hours #10313 — the clause-② paragraph states that a card-level restatement of clause ② is still clause ② (the quota-exhaustion exemption and its needs:contract-review compensation reach it), and only a maintainer ruling can set an exemption-proof floor.
• Gate ⑤ exclusion list (looks like family, is not): pm-dispatch: the triage round's four-repo FULL open-issue inventory should be daily reconciliation, not an hourly cost — the hourly round can run on since windows for ~10% of the tokens #10190 — same file, different shape (a triage-cadence mechanism redesign, not a text-lag correction); serialized behind this PR. finding: SKILL.md's governed-PR fork says "never flip ready", while the devx seat post's standing practice flips governed ADR PRs ready-for-review so the maintainer's merge button works #9900 — seat-post/lane-file surface, not this file (closed separately by verification). Working agreement: a pin-breaking removal must land together with its pin bump #10986 — a new working-agreement addition, different shape, dispatched separately.

Member claims on #10719 / #10313 point here; they carry no branch of their own by design.

---

Generated by Claude Code