[pm:seat] domain:devx @ objectui — ⚪ VACANT (stood down 07:26Z 08-24 on maintainer order) · in flight 0 · queue 45 · 1 PR awaiting the maintainer's release timing

[os-zhuang]

Seat registration — domain:devx @ objectui. Lane scope and standing duties live in the pm-dispatch skill; this post carries current state only.

1 · Current PM

⚪ VACANT — no PM holds this seat.

Stood down 2026-08-24T07:26Z on the maintainer's order, verbatim: 「当前任务处理完合并后就下班。」 The last session (session_01PQ3NihCHE9LUtHoGxo6A9f) finished its in-flight work, landed it, and stopped. In flight 0. No check-ins armed. Nothing is being watched.

The next /pm-dispatch devx@objectui fire takes the seat normally — run the three open-round mutex readings against this post and the closing briefs below.

Lane	domain:devx, repo objectstack-ai/objectui
Last rounds	R3 01:56Z–03:25Z · R4 03:26Z–07:26Z, both closed. (R1/R2 were session_0124Qg8rLvpXnQDwCmpKUmaJ, 08-23)
Landed R3+R4	6 cards / 6 PRs — #5367 (PR #5938) · #3840 (#5943) · #5458 (#5946) · #4477 (#5948) · #5174 batch 1 (#5951) · #4514 (#5960)
Closed as already-delivered	#5313 — shipped by merged PR #5412 on 08-20, never closed
Queue	47 pm:queue unassigned; 45 dispatchable (#5397 and #5775 are held)

2 · ⚠️ Left for the maintainer — nobody is watching this

#5442 / PR #5774 — ready, mergeable_state: clean, re-verified under a moved main. ⛔ NOT enqueued, deliberately: enqueueing publishes the 39-package fixed group at 17.6.0, and release is human-only. It will wait indefinitely; no seat is monitoring it.

3 · Inherited ledger

• [finding] The changeset release PR runs no CI at all — every ci.yml run on changeset-release/main is action_required, so the release commit reaches main unvalidated #5397, [finding] The lane job's declared mirror of readChangesetState names one ignored filename; the installed reader ignores four and also reads .changeset/pre/*.md #5775 — held strictly behind 16 of 90 versioned releases never reached npm — a release-PR merge that lands with changesets pending versions again instead of publishing (npm is currently a release behind: repo 17.6.0, npm 17.5.0) #5442: same file (changeset-release.yml), same region. pm:queue, held by ruling rather than by label. ⛔ Untouched.
• check-doc-snippet-types.mjs collects only .mdx under content/docs — 40 .md guides are uncovered with no UNGATED_DOCS entry, contradicting its own "covered by default" rule #5174 — batch 1 landed (PR docs(guides,tooling): clear ten .md pages off the doc-snippet ledger (#5174 batch 1) #5951). 9 .md guide entries remain, 274 diagnostics; layout (45) and component-registry (59) are 104 of them. ⚠️ Its body and the older comments say "51 .md guides" — wrong, it counted 32 package READMEs. The R4 comment on the card carries the corrected numbers and the live scope. ⛔ Do not dispatch from the body.
• Seven published packages still emit extensionless relative specifiers, so plain Node ESM cannot load them #5214 — stale assignee (os-support-ai), mis-graded pm:queue. Tracking parent, not dispatchable. ⛔ Not silently re-labelled.
• finding: 未使用的具名导入/局部变量在本仓结构上不可能让 CI 变红 —— no-unused-vars 是 warn(无 --max-warnings),且 38 个包 extends 的根 tsconfig 关着 noUnusedLocals(base 里开着的那条 0 个包在用) #4806 — assigned to yinlianghui, a non-loop identity. ⛔ Never touched.
• Install the parameterised half-state patrol (sweeper + workflow pair from objectstack PR #11294) #5791 — the half-state patrol is not installed here (.github/workflows/half-state-patrol.yml absent), so this seat's "read the patrol anchor first" criterion is vacuous. A measurement for its implementer is on the card.
• 合并队列声称「已强制」却从未产生过一次 merge_group 构建(repo-wide 0),必需集实测不含 4 个 shard / Type Check / Lint —— #3523 的第 3 步从未落地,而 AGENTS.md §9 已按「队列会替你兜住」反转了 auto-merge 禁令 #4986 — dated instance attached in R1. ⛔ Evidence only.
• @object-ui/app-shell still cannot take the nodenext pin: @monaco-editor/react lazy import is TS2345 under Node16/NodeNext #5440 pm:blocked · six cards pm:on-hold ([test-harness] await act(async …) never settles when a test renders the console AppContent surface #5194 合并队列声称「已强制」却从未产生过一次 merge_group 构建(repo-wide 0),必需集实测不含 4 个 shard / Type Check / Lint —— #3523 的第 3 步从未落地,而 AGENTS.md §9 已按「队列会替你兜住」反转了 auto-merge 禁令 #4986 metadata-admin DOM tests: two surfaces that never settle under the light dom project, with no diagnostic #4249 28 个 vite/vitest config 用 __dirname,Vite 8 已警告它在未来 major 默认的 configLoader: 'native' 下不受支持 #3592 [finding] 类型门收紧 PR 与旧式抑制指令的合并窗口相撞已两例(第二例弄红了 main)—— 值得一道结构性防线 #3535 [ADR-0057] P4 follow-ups — live handoff verification + context-carrying upgrade (cloud#817) #2443).

4 · Notes

Standing constraints

• Governed surface (.claude/**, docs/adr/**, skills/**, AGENTS.md, CLAUDE.md): draft only, ⛔ never ready / enqueued / auto-merged; request review from os-zhuang and report under "awaiting a human merge".
• Enqueue resource = every check green, ⛔ not the required subset. Gates read by name (合并队列声称「已强制」却从未产生过一次 merge_group 构建(repo-wide 0),必需集实测不含 4 个 shard / Type Check / Lint —— #3523 的第 3 步从未落地,而 AGENTS.md §9 已按「队列会替你兜住」反转了 auto-merge 禁令 #4986).
• ⚠️ Landing method flagged to the maintainer, unanswered. All six PRs landed via GitHub auto-merge after full-green. The skill says the merge queue is the only sanctioned path, but 合并队列声称「已强制」却从未产生过一次 merge_group 构建(repo-wide 0),必需集实测不含 4 个 shard / Type Check / Lint —— #3523 的第 3 步从未落地,而 AGENTS.md §9 已按「队列会替你兜住」反转了 auto-merge 禁令 #4986 records that objectui's queue has never produced a merge_group build. Raised twice; no ruling. The next seat should re-raise it rather than assume this is settled.
• Ref deletion is unavailable to every agent seat (403 via the proxy). Strays needing a maintainer: refs/heads/probe/write-check-5442, claude/issue-5409-delete-dead-changelog-workflow, claude/issue-5367-readtsconfig-jsonc (0 commits).
• objectui has no scripts/pm/ — dispatch-gates.mjs is objectstack-resident, so tiers are explicit seat decisions and gate lists are derived from the diff by hand.
• ⚠️ pm:dispatched sits on 816 closed cards while this seat strips it on landing. Unresolvable from inside this repo; recorded on Install the parameterised half-state patrol (sweeper + workflow pair from objectstack PR #11294) #5791 for the patrol's implementer to settle before it emits 816 rows on day one.

Method notes worth keeping

• list_issues labels is a UNION, ⛔ not an intersection. Measured: ["domain:devx"] → 61, ["domain:devx","pm:queue"] → 135. A lane∩state count from one query is inflated 2–3× and looks entirely plausible. Read the lane on its single domain:* label and intersect locally. Filed as objectstack#11559.
• git ls-remote before judging a dev dead. Mounting an I18nProvider in a test file silently changes what every LATER provider-less render in that file resolves #4514 showed no PR and no report at 50 minutes — GitHub's PR list was empty — but the branch existed with two commits and 350 lines of correct work. The PR list is not the evidence; the branch is.
• A card can be finished and still sit in pm:queue (finding: assertCanonicalVitestInvocation does not cover the 11 packages that own a vitest.config.ts — measured: a package-cwd run in plugin-grid is NOT refused #5313, four days). The decision re-read plus a git grep on origin/main catches it; the candidate query never will.
• An empty grep is not a reading. A mis-escaped predicate returned nothing this session and would have read as "already fixed". Always pair a zero with a term certain to be present.
• Census line numbers rot faster than the defect (Four live adapter.find calls pass an unprefixed query option — and no-query-params-under-options gates only the sibling half of the class #5458: one of four sites already fixed, two moved 20–120 lines).
• A green block can be green for the wrong reason ([finding] object-form 与 object-master-detail-form 在 binding-reach 探针的自动 fixture 下渲染成错误卡(reading 'map')—— 断言仍成立,但两个分支的崩溃守卫因此没法一起开 #3840: a sample outside the prop's vocabulary routed around the crashing gate).
• A fixture can assert the defect (Four live adapter.find calls pass an unprefixed query option — and no-query-params-under-options gates only the sibling half of the class #5458: expect.objectContaining({ top: 200 }) passed while pinning the dead spelling).
• Timer text is delivered stale. A fired check-in carried pre-edit text still saying "then open R5" after the stand-down order had superseded it. Re-read state on every fire; ⛔ never execute a timer's prose.
• A squash-merged PR's head sha is not an ancestor of main — read the log or the PR state.

PM premises falsified by devs this session — both worth the habit

1. "51 .md guides" (check-doc-snippet-types.mjs collects only .mdx under content/docs — 40 .md guides are uncovered with no UNGATED_DOCS entry, contradicting its own "covered by default" rule #5174) — counted package READMEs; the real population was 19, now 9. It was in the deferral note and the dispatch order and would have propagated.
2. "Mechanically clearable now The getting-started guides teach ~14 symbols the packages do not export — one decision per symbol, applied everywhere (the #5160 class, on content/docs/guide/**) #5343 landed" (check-doc-snippet-types.mjs collects only .mdx under content/docs — 40 .md guides are uncovered with no UNGATED_DOCS entry, contradicting its own "covered by default" rule #5174) — measured false: zero fabricated-export diagnostics remain anywhere in the guide tree. ⚠️ The answer was already in the file the dev was told to edit — the gate's own docblock says so. An assumption was asserted where a reading was available.
3. "Reset the global instance's language" (Mounting an I18nProvider in a test file silently changes what every LATER provider-less render in that file resolves #4514) — the leak is the pointer: initReactI18next.init runs setI18n(instance), so a provider replaces the global wholesale. A language-only reset would have looked like it worked.

Lane baseline — 12 dispatches. claim → draft PR: ~5 · ~11 · ~13 · ~17 · ~25 · ~30 · ~19 · ~29 · ~43 · ~14 · ~51 · ~85 min. claim → merged: 34 · 45 · 72 · 51 · 79 · 238 min. ⛔ Still not a liveness threshold — the 238 is #4514, which was alive and working the whole time. The deciding reading for death is an absent report plus an absent branch, ⛔ never an empty PR list.