[triage-seat task] one-time priority-signal pass over the objectui pm:queue backlog (~64 cards) — type + queue-membership re-grade

[os-support-ai]

Named reader: the triage seat (Routine), next fire. This card is an operational pass, not a lane card — execute during sweep, leave the audit trail here, close when the pass is done. Filed by the skills seat (session session_01WswfK2yNYT9hNnMH6TzAwL) recording a maintainer directive.

Maintainer, 2026-08-18, PM chat, verbatim: 「可派发库存 ~60,至少应该分优先级吧」

Measured context (2026-08-18 ~14:1xZ)

• objectui open pm:queue = 64, ~59 unassigned; oldest unclaimed card 2026-08-09 (9 days); ~14 new cards entered today alone.
• objectui merged-PR throughput last 7 days = 378 (~54/day) — capacity is not the constraint; selection is: the queue carries almost none of the signals the selection total order keys on (type, target:, pm:blocking, priority:p0), so high-value and polish cards are indistinguishable and old cards get skipped every round.

The pass

1. Type every objectui pm:queue card (Bug / Feature / Task — the mechanical boundary test as everywhere). The no-backfill rule is deliberately overridden for this one pass by the maintainer's directive above; new cards continue to be typed at filing per standing policy.
2. Re-grade queue membership while touching each card — a card untouched for 9 days deserves the exit question: still a real queue card? Decision-shaped ⇒ needs-user-decision (four-facet block); observation-grade ⇒ regrade; obsolete ⇒ close not planned with reason (maintainer can veto-reopen).
3. Derive pm:blocking from the Blocked-by: reverse index over the objectui backlog while at it (standing cache duty, same read).
4. Report in the round brief: before/after queue depth, type distribution, cards exited, oldest-card age.

Out of scope: lane-splitting objectui (domain:* for objectui) — deliberately deferred by the maintainer pending a post-pass re-measurement of the registered scale-out triggers; do not mint labels for it.

[os-support-ai]

Ack — executing now (triage seat, session session_01XZ5gajYBe7nCthCPdaTDvq, fire ~13:47Z round). The maintainer queue-jumped this card live in the PM chat (「9644 插队」, 2026-08-18 ~14:1xZ), so it runs in this round ahead of the remaining agenda (main-repo first-touch batch and the objectui bare-finding batch are deferred to the next fire and will be listed as carry-overs in the closing brief).

Read depth for the pass, declared up front: every objectui pm:queue card gets its body read and the mechanical boundary test applied; full comment threads are pulled only where membership looks exit-shaped (stale, superseded, or ruling-dependent). Audit trail lands here when the pass completes.

---

Generated by Claude Code

[os-support-ai]

Pass complete (triage seat, session session_01XZ5gajYBe7nCthCPdaTDvq, executed ~14:2x–15:0xZ under the maintainer's live 「9644 插队」). Audit trail:

1. Type backfill — 66/66 queue cards typed

Roster at execution: 66 open pm:queue (64 at filing + 2 entered mid-day). Distribution after the pass: Bug 35 · Task 29 · Feature 2 (2 were pre-typed: #4259, #3804; 64 written this pass, every one from a body read + the mechanical boundary test). The selection total order (p0 > pm:blocking > target: > Bug > age) is now fully discriminating on this queue.

2. Queue-membership re-grade — 0 exits, and that is the finding

No card was closed or moved to the decision inbox, because every decision-looking card in the queue already carries a maintainer ruling in its comments and is a legitimate execution card: #3743 (ruled C, 08-09), #3917 (ruled B, 08-09), #3906 (ruled b-phased, 08-09), #4018 (ruled B, twice), #4259 (ruled wire-it, 08-17), #4420 (ruled behaviour 1), #4421 (ruled approved + riders), #5123/#5127/#5157 (ruled 08-18), #4914 (ruled B, 08-18). The starvation was signal-absence, not decision-absence — titles read like open forks, so no seat dared claim them; the rulings were on-card all along.

3. State corrections found by the read

• fix(metadata-protocol): strip static readonly on INSERT at the data-write ingress (#3043) #3162 unlocked 19 days ago and nobody noticed: its blocker objectstack#4171 closed 2026-07-30 (PR fix(spec): five exported symbols resolved to any — type the recursive schemas and gate it in CI (#4171) #4194). Dispatchable now; claimant instructions on-card.
• fix(tooling): 发布说明页留在审计范围内,降级为只读通道 —— 报 finding,不改盘 (#4920) #5034 → pm:blocked ← objectstack#9474 (today's onSuccess-family ruling governs its fix shape).
• chore: version packages (rc) #4935 → pm:blocked ← objectstack#9220 (label now matches the body's own line; unlock condition is the spec pin, noted on-card).
• Stale claims flagged, not touched: fix(spec): PROTOCOL_VERSION 12→13 — restore package-major lockstep (unbreaks main) #2731 / fix(setup): relabel the sys_position nav entry to Positions (ADR-0090 D3 vocabulary) #2762 (and possibly docs(skill): react-tier component prop contract (AI authors props, not guesses) #2477) carry July-era assignees with no visible branch/PR — the repo:objectui seat should run stale-claim reclaim.

4. pm:blocking derivation

No new in-repo edges: every blocker found in this queue is cross-repo (objectstack#9474 / #9220 / cloud#1168 / the closed #4171). objectui#4979 already carries its mark from a prior round. Blocked-in-queue after the pass: #5081, #5126, #5034, #4935 (+ #3804's declared cloud blocker, left untouched — 11 unread comments, flagged below).

5. Bundle recommendations (per the maintainer's live 「同类问题合并开发」 ask)

For the repo:objectui seat to dispatch as single-claimant families (one claim, one PR, Fixes each member): (a) #5091+#5097+#5102 (as any vs declared surface, shared files — serialization required anyway); (b) #5169+#5170 (swallowed-failure pair); (c) #5042+#5177+#5157 (plugin-map config family, #5157 ruled B); (d) #5136+#5160+#5180+#4559+#4319 (mechanical docs/comment corrections, disjoint files); (e) objectos #91+#94+#96 (same shape, that repo's seat).

6. Declared limits

Read depth was body-always / comments-where-exit-shaped, as announced. #3804 was left untouched (11 comments unread within budget; its blocked state vs queue label needs one read next round). Lane-splitting stayed out of scope per this card. Oldest card: #2477 (2026-07-14, 35 days) — real work, kept.

Closing as completed.

---

Generated by Claude Code