Seam-card queues have no named reader: `repo:*` seat probes never scan objectstack's `repo:objectui`/`repo:cloud` labeled cards — a routed p0 sat unclaimed for a day and a half

[os-zhuang]

Filed by the skills seat (session session_017TNzEetykdh7ceZGwuAPLq, 2026-08-16) after the maintainer's direct question in the live PM session (verbatim, untranslated): 「这个问题报了很久没有处理，是因为 objectui 项目经理不会再看 objectstack 仓库是吗？」— the answer is yes, and by protocol design rather than negligence.

Measured

• The Console filterBuilder p0 (priority:p0 + target:v17, filed 2026-08-15 04:21) was triaged correctly within 29 minutes — settled seam placement, repo:objectui + pm:queue, dedup clean, fallback priced, p0 queue-jump noted. It then sat unclaimed ~36h.
• The repo:objectui seat post's 范围 section reads, in full: 「objectui 全仓队列」. The objectstack seam-card queue (label:repo:objectui here) is not in its declared scan scope, and nothing else names it.
• The pm-dispatch skill's seam-card rule (多仓协调 Add metamodel interfaces for ObjectQL/ObjectUI contract #1: seam cards stay in objectstack carrying repo:objectui/repo:cloud) presupposes the repo seat reads this queue — the triage comment on the p0 states that presupposition explicitly. Both sides executed their own text faithfully; the gap is between the texts.
• This violates the skill's own invariant: 「一个标签存在,当且仅当有具名读者」— repo:objectui/repo:cloud on objectstack cards currently have none.

Deliverable (one PR, skills lane)

Give the seam queues a named reader, in the versioned protocol rather than seat-post prose: the repo:* seat's round-start probe (整车道一次读全) explicitly includes the objectstack seam query — list_issues on objectstack with the single label repo:<repo> (single-label per the OR-semantics trap row) — merged into the same candidate pool and priority ordering as the home-repo queue. Landing spot is the dev's call between the 多仓协调 rule in SKILL.md (one clause naming the reader) and references/seat-post-protocol.md (the repo-seat probe checklist); ratchet ceilings apply, fund by compression. The objectui/cloud seat posts' 范围 lines then update to match (one-line edits by their own seats, notified — not this PR's files).

Interim mitigation (already executed)

Cross-seat notice posted on the repo:objectui seat post naming the p0 and asking this shift to include the seam query until the protocol lands.

Constraints

.claude/skills/pm-dispatch/** ⇒ fable-mandatory (derive via --tier); ADR-class human merge; new text in Chinese, machine literals English; no issue numbers in operational text.

[os-zhuang]

Scope extension (skills seat, session session_017TNzEetykdh7ceZGwuAPLq) — the maintainer asked how this problem class gets solved permanently (verbatim, untranslated: 「以后类似的问题要怎么解决」). Two riders added to this card's deliverable, forming the three-layer fix:

1. Stale-p0 detector (mechanical backstop): extend scripts/pm/check-half-states.mjs with a report-only detector — priority:p0 + open + unassigned + older than a threshold since last label/routing activity ⇒ flagged for the triage round brief and the decision box. A no-reader gap should never again be discoverable only by the maintainer asking; p0 volume is tiny, the scan is free. Self-tests extended per the file's convention.
2. One-time dead-letter audit (verification step, not code): as part of this card's verification, enumerate every circulating repo:* / domain:* label across the three repos against the seat posts' 范围 lines, and list any label with open stock but no declared reader in the report — clearing the whole class at once rather than per-incident.

File surface grows to include scripts/pm/check-half-states.mjs (+ self-tests) alongside the protocol-text half; still one card, one PR, fable (protocol half mandates it), ADR-class human merge.

---

Generated by Claude Code

[os-zhuang]

Maintainer ruling recorded (skills seat live PM session, 2026-08-16, verbatim, untranslated) — presented with A (repo:* seat probes add the seam query) vs B (triage transfers cards to the fix-landing repo, seam exception tightened back to its criterion), the maintainer ruled:

用 B

Scope finalized accordingly (supersedes the probe half of the original body):

1. SKILL.md 多仓协调 rule tightening: file-at-destination executed strictly by the existing criterion (正文抽掉 objectstack 还成立 ⇒ transfer)— console/UI defects transfer to objectui at triage time, citing the rebuild recipe (出处头 + full-name refs + close source as moved); the seam-card category shrinks back to genuine coordination cards, which must name their reader in the card body rather than rely on a label queue. The 2026-08-10 file-at-destination ruling stays cited; this is enforcement, not a new rule.
2. Stale-p0 detector (scripts/pm/check-half-states.mjs, report-only, self-tests extended): priority:p0 + open + unassigned + stale since routing ⇒ flagged for the triage brief and decision box — the A/B-independent backstop.
3. Verification-step audit: enumerate circulating repo:*/domain:* labels against seat-post 范围 lines; list dead-letter labels AND the existing seam-card stock re-classified under the tightened criterion (the listing is the deliverable; actual transfers are triage bookkeeping post-merge).

Serial position: SKILL.md is a hot chain — this card queues behind the claim-section PR currently awaiting the maintainer's merge, then dispatches at chain head (fable, ADR-class human merge; the merge event auto-triggers dispatch). The interim mitigation (cross-seat notice + the p0 directive on the filterBuilder card) stands until then.

---

Generated by Claude Code

[os-zhuang]

Bundle member claim — rides the on-hold protocol card as anchor (full claim, finalized scope, and the maintainer's 「用 B」/「确认，先处理 objectstack」 rulings recorded there; session session_017TNzEetykdh7ceZGwuAPLq). Overlapping file surface (SKILL.md + check-half-states.mjs) makes one PR the correct shape. Dispatches when the SKILL.md chain clears.

---

Generated by Claude Code