pm-dispatch: no seat gives objectstack-ai/cloud cards a first touch, frozen cards slip back into the queue, and the cloud unlock scan has no pin-ancestry probe #31332

repo:cloud#2 · os-zhuang · session_0123kRqqdpAJWYMgsgdAJX4G

维护者速读

• 11-08 盘点 cloud：110 张开放卡（cloud 88 + objectstack domain:cloud 22），能派的却很少。查下来有三个 skills 缺口，不是哪个座位偷懒。
• ① 没人给 cloud 仓的新卡做首次分诊：分诊职责写「两仓」，那是 objectui 还在时的写法，分诊例行轮只跑 objectstack。cloud 有 6 张卡一直没有 pm:* 标签，最早的是 11-03。
• ② 冻结只写在评论里、没有标签：ADR-0025 M2–M5 和 SDK 契约卡 10-28 冻结（「M1 做完，其余冻结」），后来一轮分诊把它们放回了 pm:queue（p1），差一点就被派发。
• ③ 跨仓解锁没有可执行的探针：规则写「消费方可安装」才算解锁，但没写 cloud 怎么验。实际卡住的就是这一步：上游已关，修复却还不在 cloud 的 pin 里。
• 改动只动 pm-dispatch 的 skills 文本，三条各一两行，不改脚本。

Filing gate: ① a defect with a named location (the skill text below). Raised by the maintainer, verbatim, in the cloud seat-2 thread, 2026-11-08 02:02Z: 「这些问题问什么分诊没处理？你处理完要和分诊沟通」 / 「是 skills 的问题吗？」.

Reader who acts: the domain:skills lane, as one PR on .claude/skills/pm-dispatch/**.

Dedupe: I searched label:domain:skills triage cloud (8 open, none on this) and 两仓 triage-duties (#15671 recorded the alternating-repo pickup and is closed; #25490 is closed). There are no hits for a cloud first-touch or a pin probe.

The three gaps, as measured on 2026-11-08

1. Cloud cards get no first touch

• Where: references/triage-duties.md:8 says 「每日一 fire 跑两仓全量对账」, and lines 12–13 define the bare-card disjunct ① (全裸). Neither line names the repositories. The triage seat post [PM seat] triage (objectstack-wide) — 🟢 Routine · session_01AavokzJ5DndAwitDXvKy4U · trig_01NcnCtMS2tH46nUg1a5TaMJ (hourly) · state = body + the round records newer than it #6015 and its round records cover objectstack only. Every record says "Next fire: objectstack", and R973 and R974 are objectstack rounds.
• Measured: these cloud cards had no pm:* label until the seat self-triaged them today:
  • cloud#3519, open since 11-03. It was a ruled decision whose execution cards were all closed.
  • cloud#3549, open since 11-03.
  • cloud#3552, open since 11-04. Its blocker cloud#3551 had been closed since 11-05.
  • cloud#3567, cloud#3635 and cloud#3659.
• Fix: name the repositories in triage-duties.md:8: objectstack-ai/objectstack and objectstack-ai/cloud (objectui is archived). Then pick one of two:
  • (a) Run disjunct ① over cloud too. Triage only routes; it adds priority:* and pm:queue, because every cloud card belongs to the cloud lane.
  • (b) Say in SKILL.md that the cloud lane patrol owns first touch for its own repository.
    The cloud seat runs (b) from today in its hourly patrol, so writing (b) down costs nothing. (a) adds a second reader.

2. A freeze recorded only in a comment is undone by a later round

• Where: references/state-machine.md:14-15 says a hold is legal only with pm:on-hold plus a Restart-when: line. The 10-28 freeze of objectstack#27124, ADR-0025 M3 — third-party tier: T1 sandbox dispatch to QuickJS, publish-side secret / vulnerability scans, secret-refs to the settings/KV store, the fs / network enforce-or-remove decision, the deferred §7 answers #27125, ADR-0025 M4 — the administrator plugin shelf: ADR-0019 D2 amended, packageType: plugin listings behind manage_platform_settings, install / upgrade / uninstall from Setup through the M2 path, one Installed list with a Kind column #27284, ADR-0025 M5 — hot attach / detach without restart: kernel attach/detach with per-plugin resource accounting, a mutable plugin route table, plugin:loaded reconcile, cluster fan-out; restart kept for sign-in, driver and core-service plugins #27285 and ADR-0025 SDK contract — the three public seams external plugins build on (messaging channel provider, identity-object writes, PluginContext subset), a conformance testkit, the peer-dependency build check, and the objectstack-ai/plugins repository #27287 was posted as a Frozen comment, with no label and no Restart-when:. The cards kept status:parked, and a later triage round put them in pm:queue.
• Measured: five p1 cards sat in pm:queue against a standing freeze. The seat moved them to pm:on-hold with Restart-when: today.
• Fix:
  • (i) One line in state-machine.md: a freeze or 冻结 ruling lands as pm:on-hold + Restart-when: in the same write. A comment alone is a half state.
  • (ii) One line in triage-duties.md: before triage puts a card into pm:queue, it reads the latest state-transition comment (dispatch-runbook.md:59 already defines how to find it). If that comment is a hold, freeze or ruling, triage files pm:retriage to the lane instead of re-queueing.

3. Cross-repo unlock for cloud has no probe

• Where: references/state-machine.md (「跨仓解锁判据是消费方可安装 …」) names the rule, and SKILL.md:109 says a cross-repo close is never released automatically. Neither says how the cloud lane checks it.
• Measured:
  • cloud#3654 and cloud#3693 had every Blocked-by: closed. Their fixes, objectstack#31243 40eaae7f and feat(runtime): export the package disable door's dependent-rule verdict and its sentence on the package entry #31157 4bf30da1, are not ancestors of cloud's pin 90523379, so cloud cannot install them yet.
  • cloud#3413's blockers are in the pin, and it was unlocked today.
• Fix: one line for the cloud lane's unlock scan. The probe is git merge-base --is-ancestor <fix-merge-sha> $(cat .objectstack-sha) in the objectstack checkout. Until it passes, the card's Blocked-by: moves to the pin-round card that will carry the fix.

Acceptance

• One PR on .claude/skills/pm-dispatch/** with the three edits, and any --self-test or lint the skill tree runs is green.
• The PR body quotes this card's gap numbers.
• The triage seat post [PM seat] triage (objectstack-wide) — 🟢 Routine · session_01AavokzJ5DndAwitDXvKy4U · trig_01NcnCtMS2tH46nUg1a5TaMJ (hourly) · state = body + the round records newer than it #6015 gets one line naming the new scope.

Refs: objectstack#6015 (triage seat post)