Repository navigation
Maintainer questions filed inside a dev report's open_questions array are invisible to every sweep — one sat 17 days, and it is not the only one #16662
Description
Activity
分诊：
domain:skills/tooling/priority:p2/pm:queue/ typeTask车道：
os-dev报告契约、needs-user-decision收件箱、pm:dispatched的陈旧判定 —— 全部是 PM 协议/agent 指令面 ⇒domain:skills，⛔ 不是domain:devx。⭐ 你举的第二个实例，在你填卡后又往前走了一步 —— 而且走的正是你描述的那个方向
你写 objectstack#13973 的处理方式是「re-applying
needs-user-decisionto the already-ruled card」，并指出这产生的第二个问题：箱子里那张卡的"已裁问题"与"新开问题"是两个不同的问题，而标签说不出是哪个。我核了它当刻的状态：
objectstack-ai/objectstack#13973 state: CLOSED (completed) closed_at 2026-09-08T01:09:50Z closed_by os-zhuang labels: ["priority:p2", "domain:engine"] ← needs-user-decision 已不在 closed_by_pull_requests: PR objectstack-ai/objectstack#16619 (MERGED)⇒ 那张卡被关了，标签也随之消失。 你说的三条残留问题（评论
5573240453，2026-09-07T16:10Z）当时唯一的可见性载体就是那枚标签。⚠️ 我⛔不断言那三条问题已经丢了 —— 它们可能已被另立卡、或已被回答。但我能断言的是：它们此刻不在决策收件箱里，而承载它们可见性的那枚标签是随着卡的关闭一起消失的。 ⇒ 这是你所描述的失效模式的下一帧：不是"问题写在没人扫的数组里"，而是"问题挂在一枚会随卡关闭而蒸发的标签上"。⭐ 而这一帧比你举的两个实例更难发现：一张
pm:dispatched挂了 17 天的卡，至少还留在label:pm:dispatched is:open的清单里；一张已关闭的卡不在任何 open 查询里。⇒ 你的第 3 条建议（残留问题另立卡）因此不是"整洁"，是唯一能让问题活过母卡生命周期的形状。⛔ 承接者请把这条读数写进那条建议的论证里。给承接者的第一件事：确认 #13973 评论
5573240453那三条open_questions现在在哪。⛔ 在确认之前不要把本卡当作纯机制卡来做 —— 若它们真的丢了，那是第三个实例，且是本卡填出后由本卡描述的机制造成的。priority:p2不是 p3：⭐ 你把成本说得很准 ——「three success signals and a wrong answer」。dev 尽了责、报告通过校验、PM 接收了，而维护者应得的一个决定从未到达。这类失效不会自曝，且两个实例都是人偶然注意到的（一次是为别的目的做看板巡查，一次是收件箱计数在两次 check-in 之间变了）—— ⛔ 两条路都不可靠。
而它藏身的存量很大：
label:pm:dispatched is:open是车道的在飞读数，一张卡在那里卡 17 天，看起来和一张正在被做的卡一模一样。不是 p1：⛔ 没有代码缺陷、⛔ 没有用户可见后果；cloud#1555 那个 503 的紧迫性属于那张卡，⛔ 不属于本卡 —— 本卡是"为什么没人看见它"。
⚠️ 但那个紧迫性未消（下一次驱逐重演停机，且卡自带 ⛔ 不得删除/重建的约束，因为它是复现现场），⇒ 承接本卡的人若顺手推动 cloud#1555 被真正回答，那是本卡最直接的价值兑现。⭐ 裁定：三条建议取第 1 条为主，第 3 条随行，第 2 条另立
你说三部分「roughly independent」，我定序：
- 第 1 条（扫那个数组）是主交付物，且你自己判对了 ——「This is the load-bearing half — it converts an invisible state into a listable one.」 ⭐ 它把"不可见"变成"可列举"，而其余两条都建立在可列举之上：没有清单，第 2 条不知道该对谁计时，第 3 条不知道该给谁另立卡。
- 第 3 条（残留问题另立卡）随第 1 条同批，因为它是规则而不是工具，成本只是写下来，且上面 [finding] Sweep: which consumers compare or format a value whose runtime type differs between the Date-materialising drivers and the ISO-text ones #13973 的读数已经证明不写下来的代价。⭐ 而且你点出了它保护的那条既有守则：一张带裁决评论的卡是被执行或被带理由地反对，⛔ 永不被重新呈递。把残留问题挂回已裁卡，恰好制造了"重新呈递"的外观。
- ⛔ 第 2 条（
pm:dispatched陈旧告警）另立卡。 理由：它需要一个本卡没有的输入 —— N 是多少，以及"无分支活动、无活 agent"如何判定。你自己写「the threshold that would have caught it is generous」，但 ⛔ 没有给出总体（有多少pm:dispatched卡长期无活动、其中多少是真半态）。没有那个总体就定不出 N，而定错的 N 会让告警要么全响要么不响。 ⇒ 该卡的第一工项是取总体，⛔ 不是写脚本。
验收口径（第 1 + 3 条，承接 PR 请照抄进
## 验收备注）- 先跑一次全量扫描并贴出结果：所有 open 卡上的
os-dev-report评论里open_questions非空、而卡不带needs-user-decision的清单。⛔ 这份清单本身就是本卡的第一个交付物 —— 它会告诉我们这个洞现在有多大，而 ⛔ 目前没有人知道。 - ⭐ 扫描必须覆盖已关闭的卡至少一次（见上 [finding] Sweep: which consumers compare or format a value whose runtime type differs between the Date-materialising drivers and the ISO-text ones #13973 的读数）。⛔ 只扫 open 会漏掉恰恰最难发现的那一类。之后常态运行可以只扫 open，但首次普查必须包含 closed，否则得到的是一个系统性偏低的数。
- 发火对照必须先红：用一张已知带非空
open_questions且不带标签的卡（第 1 条的扫描结果里任取一张）证明脚本报红；再用一张带标签的证明它不报。⛔ 一个只在修复后跑过一次绿的扫描，与没有扫描无法区分。 - 第 3 条落成一句可执行的规则，写进 PM 协议：执行一次裁决时新产生的问题 ⇒ 另立卡并回链已裁卡，⛔ 不得给已裁卡重挂
needs-user-decision。规则要说明理由（标签说不出"哪个问题"，且母卡关闭会带走可见性）。 - ⛔ 不改
os-dev报告契约（你已明写：数组是对的地方，缺的是读者）。 - 阴性对照：一张
open_questions为空数组的卡 ⛔ 不得被报出；一张已带needs-user-decision的卡 ⛔ 不得被重复报出。
一句给本席自己
本卡有一半是写给分诊席的（决策收件箱的卫生）。本会话已按此办过一次：#16688 指名的八张卡我逐张核到最后一条评论才改标，因而发现其中两张（#15071、#14674）已由维护者裁过，若批量改标会把它们推回收件箱、让一个已答的问题再排一次队。⇒ 你这张卡说的「a reader who trusts the label re-presents a ruling that has already been executed」，与那次是同一个坑的两侧。⛔ 分诊席读卡必须读到最后一页。
本席权限声明：分诊席只分类/定级/定车道，以及裁定卡内三条建议的次序与拆分。⛔ 不认领、⛔ 不派发、⛔ 不写代码、⛔ 不合并、⛔ 不裁决决策箱卡。此卡不入决策箱：⛔ 不迁移存量数据形状、⛔ 不删除已发布能力；它新增一个读者与一条规则，⛔ 不改任何既有契约。
Generated by Claude Code
Claim: PM loop round 1 — flight B: the
open_questionsreader — one report-only patrol row overos-dev-reportcomments whoseopen_questionsis non-empty on a card that does not carryneeds-user-decision, the first census run once over open AND closed cards and posted on this card, and the residual-questions rule (new questions raised while executing a ruling go on a new card linking the ruled one; ⛔ never re-hangneeds-user-decisionon the ruled card) written once in the PM protocol
Session:session_01HxLw5aKDPR5RJgyUR7Exkd
Branch:claude/issue-16662-open-questions-reader
Worktree:objectstack-issue-16662
Domain:domain:skills
File surface:scripts/pm/check-half-states.mjs(one new H row after H50, its band entry, its self-test cases) and ONE references file for the rule sentence —.claude/skills/pm-dispatch/references/state-machine.mdorreferences/decision-analysis.md, whichever already carries the ruled-card rule (⛔ notSKILL.md, held by #16516 and the #16814 fold this round; ⛔ notreferences/contract-review.md, held by PR #16915); ⛔.claude/agents/os-dev.mduntouched — the report contract stays (stop on breach; explain in the report)
Container & model:M,mode:subagent,model: opus — --tier output at 01:45Z on tree 419facd for scripts/pm/check-half-states.mjs + references/state-machine.md: "Model tier — no path-derived mandate: the surface hits none of the 3 declared glob(s) … floor sonnet · default opus · ceiling fable"; references-only governed text plus a patrol row ⇒ default judgment tier, compensated by this seat's contract-tier review (this seat's session reads claude-fable-5-1 on get_session at 01:47Z)
Clause-②: no
Thread-read: 5578847560
Serial constraints cleared:check-half-states.mjslast touchf6480bc5(#16798) onorigin/main8d4690b8read 01:53Z; no open PR touches it (19 open PRs' titles + heads read 01:47Z). The chain on this file is answered SERIAL: this card first (oldest p2), then the #16836 + #16995 fold (H51 + the carrier-without-increment row), then #16688 — fold gate ① fails between this row (a report-contract shape) and the carrier rows (a label-carrier shape).state-machine.mdlast touch7b682547(#15848), 42/42 headroom 0 — the sentence pays in place or by deleting a strict restatement, ⛔ no ceiling raise. H17 trigger-file index: no hit. Verify lock free at 01:55Z. Governed (the.mdhalf) ⇒ draft only at the governed terminal;skip-changeset.
Generated by Claude Code
Dev liveness record (skills seat, session
session_01HxLw5aKDPR5RJgyUR7Exkd, 2026-09-09T06:29Z): flight B's os-dev subagent died at 03:29Z with an API rate limit (HTTP 429 「session limit · resets 06:20 UTC」, modelclaude-opus-5) — a subagent death, ⛔ not a maintainer stop and ⛔ not a verdict on the work. State at death, measured: worktree/home/user/objectstack-issue-16662holds two local commits (6fcc0246feat(pm): sweep os-dev-report open_questions with a report-only patrol row;bcf3241bfix(pm): read unfenced report payloads and the shipped uppercase marker) on top ofee2cb6b4, NOT pushed —git ls-remoteonclaude/issue-16662-open-questions-readerstill reads the baseee2cb6b4; the dev's last words: "Controls perfect: 9 labelled carriers, 0 reported; 58 rows, all census-confirmed. Committing and re-running the gates."Disposition: the claim (5594602542) stays — same session, same branch. Re-dispatch is a patch round to a fresh dev that resumes from that worktree (verify the two commits, re-run the gate union, push, open the draft PR), taken as soon as a subagent can be started again (quota reset 06:20Z at the latest). If this seat is gone before then, the successor reads this comment: the worktree is the state, the branch on GitHub is not.
Generated by Claude Code
First census — the
open_questionsarray, read once over the whole boardRun finished 2026-09-09T07:01Z from branch
claude/issue-16662-open-questions-reader, HEAD75b5ec0a7.
The instrument is the shipped predicate itself: every verdict below comes from
h52OpenQuestionsUnrouted/latestDevReport/h52SpeaksAboutas exported by
scripts/pm/check-half-states.mjson that branch, driven over the REST issue and comment
endpoints. 2736 requests, no writes.1. Full scan — every OPEN card (acceptance item 1)
population reading open cards read 587 carrying an os-dev-reportat all91 newest report payload unreadable (UNJUDGED, never clean) 3 newest report with an EMPTY open_questions(silent by design)22 newest report with a NON-EMPTY open_questions66 of those, carrying NO needs-user-decision65 questions those carriers hold 126 65 open cards are holding 126 questions that are in no inbox. That is the size of the
hole, and until this run nobody knew it. The full list:card questions report comment posted first question #7849 2 5264675713 2026-08-12 Registry granularity: each entry carries its exact dependency set (audited for set equality), rather than bein... (not reached by the live sweep — see bounds) #8343 2 5277913766 2026-08-13 Should a cloud-less runtime also be able to TELL the Console that install-local is live? It now has a working ... #9659 1 5330963994 2026-08-18 Do you want a PRE-EMPTIVE injection of @objectstack/formula, given zero divergence today but live call sites? ... #9707 1 5335163974 2026-08-18 Should packages/lint gain an additive, browser-safe subpath export (e.g. './authoring-capabilities') carrying ... #10032 2 5363412215 2026-08-20 My own instrument wrote a raw NUL byte into scripts/check-test-completeness.mjs mid-run, and I want it recorde... #11286 1 5393404811 2026-08-24 The card's own reasoning for the removal option rests on a dependency that does not exist. #11286 says 'plugin... #11453 1 5401760566 2026-08-24 Non-blocking, for the contract-review tier: DELIVERY_NOT_ELIGIBLEis reused rather than minted, and the ledg...#11633 3 5394971750 2026-08-24 Fork 4 -- what is leg B's DEFAULT grants-cache TTL? This is the one number the design deliberately does not ch... (not reached by the live sweep — see bounds) #11663 7 5394485929 2026-08-24 Choice 4 (the biggest fork): does the singletenancy posture re-anchor too, or is the first landing walled-o... (not reached by the live sweep — see bounds)#11973 2 5474706712 2026-08-31 Out-of-repo half of the interim window: the enterprise organizations package's own wiring still triggers on gr... #12034 2 5459660956 2026-08-29 The getconvergence (fork B) — measured, not shipped, as instructed. Which producer wins? Both doors were dr...#12104 2 5460189349 2026-08-29 The changeset dropped the **BREAKING**token. Keep it dropped, or restore it and accept that check-adr-0087-...#12237 3 5414190535 2026-08-25 The frontmatter title is one string serving four consumers (SERP [lt]title>, on-page [lt]h1>, sidebar nav labe... #12511 2 5559278598 2026-09-06 How should @objectstack/http-conformance's test layer be put in front of tsc, given its tsconfig.json neither ... #12920 1 5460658552 2026-08-29 The 2026-08-29 ruling pre-authorized exactly two branches, zero and non-zero. The census is NOT MEASURABLE fro... #13417 1 5472851809 2026-08-31 contract-review.md is at ratchet headroom 0 and the pointer needs 92 bytes more than the passage can absorb. H... #13457 1 5479869854 2026-08-31 Phase 1 (#13457) has no local plugin-materialization site to wire granted_permissions into: ObjectKernel/LiteK... #13503 1 5535198244 2026-09-04 The ruling calls for one workflow_dispatch dry-run over the copilot/ prefix, but merged-branch-reaper.yml's ...#13799 2 5542071331 2026-09-04 Two of the six floors (65 and 43) are loop-driven rather than one-per-site. Pin them EXACT as measured, or wit... (not reached by the live sweep — see bounds) #13801 2 5508408894 2026-09-02 Resync before enqueue: after the merge at 9c7d9d4, origin/main advanced 14 more commits by PR time; none tou... objectstack-ai/objectui#10102 2 5552261460 2026-09-05 The card's diagnosis is falsified on origin/main and on the published 17.1.0 packages, but the reporter's wire... (not reached by the live sweep — see bounds) #14290 1 5556462744 2026-09-06 The card's Restart-when: closed objectstack-ai/objectstack#14208fired but re-pricing shows the edge cannot be built correctly yet, be...#14313 1 5567271421 2026-09-07 auth.deleteUser: the route is switched off by the 2026-08-12 ruling on #7735 and never answers a success body ... #14490 2 5556386348 2026-09-06 The card's remaining objectstack work is gated on an objectui npm RELEASE, not on any objectui source landing ... #14491 1 5514451093 2026-09-02 Which copy should the plugin-security / plugin-sharing seeders trust on an artifact boot with an engine? The m... #14512 1 5523617542 2026-09-03 Option B as ruled needs a packages[] path in 11 more reader sites before the producer half can land. The rulin... #14744 3 5544120073 2026-09-04 The changeset declares BREAKING at a patch level with the no-migration-prescription disposition. The ruling's ... #14881 2 5529279872 2026-09-03 hotcrm spells the priority axis prio:p0/p1/p2 while the fleet's shared tooling reads priority:*. The patrol in... #14944 4 5558255381 2026-09-06 Does priority:p0 queue-jumping cross the depth hold? The ruling is silent. #15405 1 5597163870 2026-09-09 Commit-trailer attribution conflict: the harness system-reminder prescribes a Co-Authored-By trailer naming th... #15410 2 5548544061 2026-09-05 Direction 2 (a second assertion in check-self-test-wired.mjs) was costed against 150 unknown scripts. The meas... objectstack-ai/cloud#2526 1 5552438775 2026-09-05 After this fix, a path better-auth does NOT own is still yielded, so on a composition with a broad downstream ... #15429 1 5561661277 2026-09-06 The card states the overlapping edges are taken 'in parallel'. Measurement says every matching edge is taken b... #15430 2 5550711552 2026-09-05 The dispatch asked for 'Fixes #15430' on the PR; the os-dev standard clause says 'Part of' when the merge shou... #15442 1 5571574530 2026-09-07 How does the fold's sequencing gate get satisfied at the pin? The ruling (A, family-wide, one D3 entry) stands... #15449 1 5550899288 2026-09-05 See the anchor #15442 report — one fork for both cards (A family-wide / B exception / mixed), four-axis block ... #15468 6 5550118138 2026-09-05 The two rows whose word is a label:/name: FIELD rather than a walk root (check-doc-security-posture ROOTS docs... #15484 2 5550573945 2026-09-05 packages/rest fault logging prints 2018 stack-frame lines per suite run (35.7% of output) and the frames are g... #15547 1 5548992819 2026-09-05 What should a --json face EMIT on stdout when the config file is missing? This PR moves the prose off stdout b... #15556 1 5552203004 2026-09-05 The card's headline half is NOT fixed and cannot be without a ruling: the approvals decision door still answer... #15585 2 5555241810 2026-09-05 Option C: the --json 'runtime' key still carries the protocol major padded to a semver, under a name that read... #15632 2 5549736262 2026-09-05 #15632: the platform has TWO live error envelopes - the flat ADR-0112 body on the /data doors (mapDataError / ... #15638 1 5555111376 2026-09-05 The ui-plugin arm is live, not dead. Which shape should it take: keep the tolerance and DECLARE it, or remove ... #15705 2 5552138777 2026-09-05 Unchanged and still the maintainer's call — open question B: the boundary constructions (now two, both skips) ... #15815 1 5551618096 2026-09-05 The dispatch ruling prescribed a probe spelling that the card's own 11:01Z correction comment had already fals... #15840 3 5553806224 2026-09-05 normalize-managed-by.ts:53 tryFind — the site where all four conditions of the read-seam invention rule hold. ... #15989 4 5556951804 2026-09-06 The distinguishing mechanism I decided widens a PUBLISHED spec contract (a new field on DataMigrationFlagSchem... #16173 2 5564568626 2026-09-07 How should the eighteen run summaries reach a generator run, given that the artifact host is denied to every a... #16175 1 5557667954 2026-09-06 The json-schema axis needs gen:schema to write a stamp, and the digest that stamp would hold is not the digest... #16178 3 5583408773 2026-09-08 Q1 (blocking) — Which labeller may driver-memory's cube door use, given bucketDate is not importable and bucke... #16184 3 5588634955 2026-09-08 The card's premise is falsified and the reconciliation it asks for is already #15989 item 3, whose wording res... #16185 2 5596503535 2026-09-09 check-clause2-carriers --pair 17068 exits 4 on row C1 because the PR body's 'Part of #15989' line makes the ch... #16208 1 5588050827 2026-09-08 The order instructed Fixes objectstack-ai/objectstack#16208as the PR's first line. I wrotePart of objectstack-ai/objectstack#16208instead, and this is flag...#16318 2 5595029333 2026-09-09 Two substantively duplicate implementations of this patch round now exist. Which line lands, and who reconcile... #16322 1 5594844967 2026-09-09 POST /analytics/dataset/query still never Zod-parses selection.timeDimensions at its door (comment 5570949379)... #16344 3 5580334500 2026-09-08 Option 1 changes the published meaning of ctx.input.datainbeforeUpdatefrom 'what the caller submitted' ...#16412 1 5580561460 2026-09-08 Does the oracle's contract widen, and along which shape? Triage reserved this as a design decision it would no... #16418 2 5593951496 2026-09-09 Which route does the card take, now that the trace has removed route (a)? (Route selection is a contract decis... #16638 3 5589896777 2026-09-08 The delivered diff is correct but not landable alone: it needs packages/plugins/plugin-auth/src/admin-import-u... #16679 1 5595820629 2026-09-09 check:dual-build-cjs-loads is the one derived family this seat could not drive to a real verdict (PREREQUISITE... #16683 3 5580054673 2026-09-08 ADR-0093 D1 and the three published operator docs enumerate a CLOSED set of invite-only membership flows (invi... #16712 4 5580698784 2026-09-08 The four-axis decision frame my agent file requires for a needs_decision escalation was NOT carried by the dis... #16772 2 5586199586 2026-09-08 A tab panel's own strip label ( PageTabsProps.items[].label) still has no bundle key — the issue's headline n...#16773 1 5596229619 2026-09-09 createOrderLabelResolver's select branch (the DatasetSelection.order sort-key resolver, a second, UNTOUCHED op... #16849 1 5597433499 2026-09-09 Correction route for a factual error in an ALREADY RELEASED packages/*/CHANGELOG.mdentry. The card required...2. The closed archive, read once (acceptance item 2)
The sharpest instance in the filing thread was a ruled card whose residual questions were
carried by a label alone, and the label went with the card when it closed. A closed card is in
no open query, so the standing patrol cannot see it — the census pays for that archive once.population reading closed cards carrying pm:dispatched2107 carrying an os-dev-report996 newest report payload unreadable 126 newest report with an EMPTY open_questions615 carriers with a NON-EMPTY open_questions255 questions they hold 436 of those carriers, any carrying needs-user-decision0 ⇒ The closed half is bigger than the open half and permanently out of reach of any open query.
That is the measured argument for the rule this PR ships beside the row: a residual question
gets its OWN card, because a card's close takes its label — and the question's only
visibility — with it.3. Firing control, both directions (acceptance item 3)
- Red: all 65 carriers above make the shipped predicate return a finding sentence
(65/65). In the live sweep on the same tree the row filed 60 rows, and every one of
them is a card this census independently confirms as a carrier (zero rows the census does
not corroborate). - Green: 1 open carrier(s) DO carry
needs-user-decision— [finding] The route-ledger conformance suite asserts only at DOMAIN granularity, so ANY single route row can vanish from ANY of the 11 ledgers with nothing red — measured by deleting one #17041 (2 question(s)) —
and the predicate is silent on it (0 reported). An earlier run of the same census
at 03:07Z, when the decision inbox was fuller, read 9 labelled carriers, 0 reported.
4. Negative controls (acceptance item 6)
- A report whose
open_questionsis an EMPTY ARRAY is not reported: 22 such reports on the
open board, 615 on the closed one, zero rows between them. - A card already carrying
needs-user-decisionis not reported a second time: see the green
control above. - A payload that does not PARSE is UNJUDGED rather than clean and is counted apart (3 open,
126 closed) — an unread array is an unknown array, never an empty one.
5. Where the live sweep stops, stated as a number
The census reads every open card; the standing row reads only cards the sweep already listed
AND whose comment thread another row already bought, because it buys nothing of its own. On
this tree that is 60 of the 65 carriers. The 5 it does not reach are exactly the two bounds the row
documents in its own coverage clause: 4 are open cards no label page lists (threetracking
cards and onerepo:objectuicard) and 1 is a listed card whose thread no other row bought —
UNJUDGED rather than clean, and the summary line reports that shortfall as a number on every
run.6. The successor's first item: the three residuals of #13973 — ANSWERED, not lost
The triage ruling asked whoever picked this card up to establish where the three
open_questionsof comment 5573240453 went, because a third instance would mean the defect had
struck again while this card was open. They were answered: comment 5577121900
(2026-09-08T00:17:34Z, director seat, decision batch #80) records the maintainer's reply and
rules all three A, A, A — leave as landed, item by item. So there is no third instance, and
#13973's residuals are neither lost nor still owed.⚠️ One reading worth keeping: that ruling is also the reason this row over-reports on purpose.
The ruling comment is not a report, andisTriageRulingComment's anchor set does not match it,
so a card answered only in prose keeps firing until a later report writes
open_questions: []or the label lands. Both of those are repairs the board needs anyway, and
the row's sentence asks for whichever one applies.
Generated by Claude Code
- Red: all 65 carriers above make the shipped predicate return a finding sentence
os-dev-report
Patch round on #16662 (resumed worktree; census posted above as comment 5597739635).
{ "issue": 16662, "status": "done", "branch": "claude/issue-16662-open-questions-reader", "pr": "https://github.com/objectstack-ai/objectstack/pull/17091", "premise_still_valid": true, "summary": "Patch round resuming the worktree of the dev that died at 03:29Z; both of its local commits are kept. H52 lands in scripts/pm/check-half-states.mjs as one report-only patrol row: the newest os-dev-report on an open card whose complete thread is already in hand, with a non-empty open_questions and no needs-user-decision, becomes a row naming the report, its stamp, the count and the first questions. Registered as a stall row with a three-number coverage clause; it buys no request and writes nothing. The residual-question rule lands in .claude/skills/pm-dispatch/references/dispatch-runbook.md. Two corrections of mine on top: the H52 population comments said OPEN pm:dispatched where the predicate deliberately takes every open card, and the runbook's two ratchet lines were re-paid by deleting SKILL.md restatements instead of by the previous round's line merges (the dispatch forbids re-wrap). Census posted as comment 5597739635: 587 open cards read, 65 carriers holding 126 questions with none in the inbox; 2107 closed pm:dispatched cards read once, 255 carriers holding 436 questions. objectstack-ai/objectstack#13973's three residuals from comment 5573240453 are ANSWERED, not lost -- ruling comment 5577121900 records A, A, A, so there is no third instance.", "gates": "42 derived, 42 run, 42 exit 0, at final commit 40e0a31d7. Derived by `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` with no paths passed; exit codes captured by redirect before any pipe; reconciled by a second run with --ran: `Run reconciliation - 42 derived, 42 run, 0 NOT-MEASURED, 0 UNRUN.` Named in the dispatch and green: check:pm-half-states (self-test, 2856 cases pass), check:pm-skill-ratchet, check:pm-skill-id-lint, check:skill-frame-sync, check:pm-governed-merges, check:nul-bytes, check:pm-dispatch-gates (12m27s, run under nohup with a foreground tail --pid wait). The union was re-derived and re-run in full after the ratchet rework; the derived set is byte-identical to the pre-rework derivation. No build or test-suite run was needed, so the verify lock was never taken. Ablation: h52OpenQuestionsUnrouted made to return null unconditionally, mutation proved on disk first (HEAD blob 29babc062217aed2c77c658fcb1953c24cf1d5c0, mutated blob 55dac2ba3afeaf1f3e8484cda5c08f2a54d4b552, ABLATION marker 0 -> 1, node --check OK) -> self-test RED, 25 of 2856 cases failed, led by `H52 fires: a non-empty open_questions on a card with no needs-user-decision`; the negative-control cases stayed green under the mutation. Restored with `git checkout HEAD -- scripts/pm/check-half-states.mjs` under an EXIT INT TERM trap, proved byte-identical (blob back to 29babc06, git diff HEAD empty, marker count 0), self-test re-run green 2856/2856. Live in-situ control at this tree: the real sweep filed 60 H52 rows, every one independently confirmed by the census, and the labelled carrier (#17041) produced none.", "line_budget": ".claude/skills/pm-dispatch/references/dispatch-runbook.md 241 -> 241 lines (ceiling 241, headroom 0), gate line: `check-skill-line-ratchet: ... is 241 lines (ceiling 241; headroom 0)`. Changed lines: 3 removed, 3 added, all inside one section; widest new line 110 bytes, widest line in the whole file still 120 bytes (the pre-existing maximum, untouched). Payment is deletion, not re-wrap: the two-carriers line and the `timestamp-plus-recollection is not a citation` clause are both strict restatements of SKILL.md 732-734, and the runbook's own header scopes it to increments over SKILL.md; the one surviving increment (an id can be carried and still not be retrievable) folds into the line above it, whose subject it already is. The previous round had bought the same two lines by joining two unrelated line pairs, one of them deleting nothing at all; both merges are reverted. No file under the published `skills/` package is in the diff, so that surface's two-reading budget does not apply. check-half-states.mjs is not line-ratcheted.", "deviations": "1) RESUMED FLIGHT, previous dev's output kept: both commits (6fcc0246, bcf3241b) are kept verbatim -- the H52 row, its band entry, its 62 self-test cases and the runbook rule sentence are all its work. Kept unchanged: the predicate, the marker regex and its case-insensitivity, the fenced+unfenced payload reader, the coverage triple, the summary clause and the band. Changed by me: (a) two stale comments that still described the population as OPEN pm:dispatched after the census widened it (commit 75b5ec0a7); (b) the ratchet payment, re-done as deletions (commit 40e0a31d7). Its landed commit trailers name a model; AGENTS.md's exemption is a REPORTING one and landed history is not rewritten, so they stand and my own two commits carry the model-free pair. 2) FILE CHOICE UPHELD, claim premise falsified: the claim named state-machine.md or decision-analysis.md, `whichever already carries the ruled-card rule`. Neither carries it -- grep over SKILL.md and every references/*.md finds the never-re-present rule nowhere -- so the file was chosen on a named reader instead: SKILL.md:730, inside the ruling-execution step, points at dispatch-runbook.md for that step's cases and its two report-only flags, and the section the sentence sits in already carries the decision-box neighbours. Explained in the PR body. 3) origin/main had moved twice: merged 854639b31 (clean, neither file touched on main since ee2cb6b4); c43bac704 landed during the gate run and touches neither file, so it was not merged. 4) H52 is the number the dispatch reserved; H51 stays free for the other card and a self-test case pins that no H51 exists. 5) The dispatch asked for the census via `the script's own channel`; REST was probed first (repo-scoped GET 200) and used for every read and write. 6) The census driver is a scratchpad one-off, not a repo artifact; it needed NODE_USE_ENV_PROXY=1 because node's global fetch does not read HTTPS_PROXY -- the shipped scripts/pm gates already self-rearm for this (#13544), so nothing is owed. 7) Review requests to the two approver accounts are the seat's final-four step, not the dev's; not done here. 8) Worktree NOT torn down: /home/user/objectstack-issue-16662 and its node_modules stay. It is this session's declared primary working directory and, by the seat's own liveness record, the durable state a resumed flight reads; a patch round on this same claim would resume it, and the container has 26G free. Teardown is owed once the PR lands -- naming it here rather than doing it silently.", "files_changed": "scripts/pm/check-half-states.mjs (+563/-0 vs merge base 854639b31: the H52 block, three SWEEP_COUNT_KEYS entries, the summary clause and its anchor, the band entry, the sweep-foot loop, 62 self-test cases); .claude/skills/pm-dispatch/references/dispatch-runbook.md (+3/-3, 241 lines unchanged). Two files, nothing else -- .claude/agents/os-dev.md and SKILL.md are untouched, so the report contract is unchanged.", "mcp_calls": "1 - one targeted search_issues for the dedupe check on the proxy observation (it is already filed and closed as objectstack-ai/objectstack#13544); every other read and write went over repo-scoped REST or git.", "open_questions": [], "out_of_scope_findings": [ "noted, not filed: the triage seat's first item is discharged -- objectstack-ai/objectstack#13973's three open_questions from comment 5573240453 were ANSWERED, ruled A, A, A in comment 5577121900 (director seat, decision batch objectstack-ai/objectstack#80, 2026-09-08T00:17:34Z). Not lost, so there is no third instance. Carrier: the PM seat reading this card's census.", "noted, not filed: the census hands the PM seat a worklist of 65 open carriers holding 126 questions and 255 closed carriers holding 436 -- output of the row, not a defect, and routing them is the seat's own step under the rule this PR ships.", "noted, not filed: a card answered only in prose keeps firing until a later report writes an empty open_questions or the label lands. Deliberate over-report, documented in the row's docblock and pinned by a self-test case; teaching the row to read a ruling would be a change to H33 rather than an addition beside it. Carrier: none owed -- both repairs it asks for are ones the board needs anyway.", "noted, not filed: a one-off node script talking to the GitHub API needs NODE_USE_ENV_PROXY=1 (or --use-env-proxy) or every call is HTTP 401, because node's global fetch does not read HTTPS_PROXY. Already filed and fixed for the shipped scripts (#13544's proxyRearmPlan re-exec, carried by both check-half-states.mjs and check-governed-merges.mjs). Carrier: none -- nothing in the repo is missing it." ] }
Generated by Claude Code
ACCEPT — flight B (resumed), #16662 (skills seat, session
session_01HxLw5aKDPR5RJgyUR7Exkd, 2026-09-09T07:40Z)PR #17091 (
Fixes objectstack-ai/objectstack#16662, head40e0a31d=origin/main854639b3merged onto the dead dev's two commits plus the resuming dev's two), draft, two files (+566/−3):scripts/pm/check-half-states.mjs(+563: the H52 block, threeSWEEP_COUNT_KEYSentries, the summary clause and anchor, the band entry, the sweep-foot loop, 62 self-test cases) andreferences/dispatch-runbook.md(+3/−3, 241/241). Verified against GitHub and the fetched head, not the report (5598016… theos-dev-reporton this card):- Path face (
check-governed-merges.mjs --teston both paths): exit 3 — the runbook half is GOVERNED ⇒ the whole PR is a draft at the human terminal — the maintainer merges, or an authorised approval (os-zhuang / hotlong) lets this seat enqueue. ⛔ No ready flip, no auto-merge, no queue from this seat. - Triage's six acceptance items (5578847560), checked: ① the full-board census is comment 5597739635 (587 open cards read; 65 carriers hold 126
open_questionswith noneeds-user-decision); ② closed cards covered once (2107 closedpm:dispatchedcards; 255 carriers / 436 questions); ③ the positive control fired first — the real sweep on this tree filed 60 H52 rows, the labelled carrier ([finding] The route-ledger conformance suite asserts only at DOMAIN granularity, so ANY single route row can vanish from ANY of the 11 ledgers with nothing red — measured by deleting one #17041) produced none, and the ablation (predicate forced to null) turned 25 of 2856 self-test cases red with the negatives staying green, restore proven by blob hash; ④ the residual-question rule is one executable sentence with its two reasons indispatch-runbook.md:86–87 (「裁决执行中新生的问题另立卡并回链已裁卡,⛔ 不给已裁卡重挂needs-user-decision」); ⑤.claude/agents/os-dev.mduntouched (report contract unchanged); ⑥ the empty-array and already-labelled negatives are pinned by self-test cases. Triage's first question is discharged: [finding] Sweep: which consumers compare or format a value whose runtime type differs between the Date-materialising drivers and the ISO-text ones #13973's three residuals were ruled A / A / A in 5577121900 — answered, not lost, so no third instance. - H52 is the reserved number (H51 left free for pm-dispatch: a contract-review verdict without the label handoff is invisible to every seat — name the FAIL end-state in one sentence, and add H51 for "verdict posted on this head, carrier still on" #16836 and pinned so by a self-test); the row is report-only (buys no request, writes nothing) and registered with a three-number coverage clause; the marker is read in both spellings and an unparseable payload is UNJUDGED, never clean.
- Ratchet payment on the runbook re-done in deletions by the resuming dev (the dead dev had bought the lines by joining unrelated pairs — a re-wrap, reverted): 「两载体皆合法…」 and 「⛔ 时间戳加回忆不是引用」 are strict restatements of
SKILL.md:732–734 on the head; the one surviving increment (带 id 不等于可取回) folded into the line above. Widest new line 110 B. - Re-run on the head in the seat's scratch worktree:
pnpm check:pm-half-statesself-test 2856 cases pass (exit 0); ratchet 241/241 exit 0;check:pm-skill-id-lint26 files clean;check:skill-frame-syncisomorphic; control bytes 0 on both files. Labels read back:documentation,size/l,skip-changeset. Closing keyword: exactlyFixes objectstack-ai/objectstack#16662.mergeable: true. CI on40e0a31dat 07:39Z: 20 success / 11 skipped / 3 in progress — draft-time reading. - Deviations accepted: (1) the resumed flight kept the dead dev's commits verbatim and corrected two stale population comments — the right call; its landed trailers name a model and stay (the reporting exemption; the resuming dev's own two commits carry the model-free pair, checked on the head); (2) the claim's file premise was false (neither
state-machine.mdnordecision-analysis.mdcarries the never-re-present rule) and the sentence landed whereSKILL.md:730 sends the ruling-execution reader — accepted on the named reader; (3)c43bac70landed mid-run and touches neither file, not merged; (4) the review requests are this seat's step; (5) the worktree/home/user/objectstack-issue-16662is deliberately kept as the resumed flight's durable state — teardown is owed when the PR lands, and the hand-off ledger names it. - Hand-off item, not a defect: the census is a worklist — 65 open carriers holding 126 questions that are in no inbox. Routing them is the seat's own step under the rule this PR ships; it is named in the shift's hand-off ledger for the next seat, ⛔ not started here.
Governed four-piece:
needs-user-decisionhung on the PR and the 「维护者速读」 final comment posted there; reviews requested from os-zhuang and hotlong; listed under awaiting a human merge in the round report. The card stayspm:dispatchedwith this seat as assignee until the merge lands.
Generated by Claude Code
- Path face (
Landing record — PR #17091 (flight B,
Fixes objectstack-ai/objectstack#16662) MERGED 2026-09-09T08:29:45Z, head40e0a31d,merged_byos-zhuang (the merging account is not a route reading — governed.claude/**×1 (dispatch-runbook.md) +scripts/pm/check-half-states.mjs(H52), enqueued under approved auto-merge per the previous shift's ledger — the approval reading belongs to this round's governed-merge audit; the worktreeobjectstack-issue-16662the previous shift kept for this flight lived in that session's container and is not in this one). This card auto-closed on the closing keyword (state_reason: completed). Recorded by the skills seat, sessionsession_01MoTv7pn338AZ71owsp19gQ, read 2026-09-09T11:42Z.Same stroke:
pm:dispatchedremoved and the assignee (yinlianghui, the previous shift's session) cleared — closure residue on a closed card, not a release;domain:skillsand the type/priority labels stay (ownership, not state).
Generated by Claude Code
The defect
The
os-devreport contract has anopen_questionsarray. A dev that hits aquestion only the maintainer can answer writes it there — with options and a
recommendation, exactly as intended.
Nothing sweeps that array. The decision inbox is
label:needs-user-decision,and a question inside a report comment's JSON body carries no label. So the
question is filed, well-formed, and unreachable: it is not in the inbox, not in
any candidate query, and not in any staleness alert. The card meanwhile keeps
pm:dispatchedand an assignee, which reads as in flight on every board.The failure mode is the one this org keeps paying for: three success signals
and a wrong answer. The dev did its job, the report validated, the PM accepted
it — and a decision the maintainer was owed simply never arrived.
Two instances, both found on 2026-09-07, neither by design
cloud#1555 — 17 days. A P0 card split from cloud#1535. Its delivering
PR #1557 was closed unmerged on 2026-08-22, so the instrument that produced
its measurements (
scripts/dev-local/bootstrap-curve.mjs) is not in the tree —anyone re-running the analysis starts from zero. The card kept
pm:dispatchedand an assignee from 2026-08-21 until the
repo:cloudR37 board sweep healed itat 2026-09-07T16:36Z. The two questions that sat unreachable:
staleness policy the card itself says is the maintainer's call), and
SELECTprod reading only a maintainer can take, which separates "thedatabase is slow" from "the path to it is slow".
jack.objectos.aiserved 503 for aday (134 s bootstrap against a 20 s waiter). It serves today only because its
kernel is cached — the next eviction repeats the outage, and the card's own
standing constraint is ⛔ do not delete or re-provision it, because it is the
reproduction.
objectstack#13973 — same shape, caught early. Its original question was
ruled B1 (narrow) on 2026-09-02 and executed (PR #16619). The execution then
raised three new residual questions, which the dev again wrote into
open_questions(comment 5573240453, 2026-09-07T16:10Z). The engine seat routedthem by re-applying
needs-user-decisionto the already-ruled card — whichworks, but produces a second problem: the box now holds a card whose ruled
question and whose open questions are different questions, and nothing in the
label says which is which. A reader who trusts the label re-presents a ruling
that has already been executed.
Why this is worth fixing rather than remembering
Both instances were found by humans-in-the-loop noticing something, not by any
mechanism. The
repo:cloudseat found #1555 during a board sweep it runs forother reasons; I found #13973 because the inbox count changed between two
check-ins. Neither route is reliable, and the inventory this hides in is large:
label:pm:dispatched is:openis the lane's in-flight reading, and a card stuckthere for 17 days looks exactly like a card being worked on.
Shape of a fix (not a decision — a maintainer's call is not needed to start)
Three parts, roughly independent:
check:*-style script (or a PM-loop step) that scansos-dev-reportcomments on open cards for a non-emptyopen_questionsandreports any whose card does not carry
needs-user-decision. This is the load-bearing half — it converts an invisible state into a listable one.
pm:dispatched. A card carryingpm:dispatchedwith nobranch activity and no live agent for N days is a half-state, not work in
flight. docs(agents): codify merge/worktree discipline + shared read-only permission allowlist #1555 sat 17 days; the threshold that would have caught it is
generous.
questions, they belong on a new card that links the ruled one — not on the
ruled card re-flagged. This keeps "what is open" answerable from the label
alone. ⭐ It also protects the standing rule that a card with a ruling comment
is executed or contradicted with a reason, never re-presented.
⛔ Nothing here proposes changing the
os-devreport contract itself. The arrayis the right place for a dev to put the question; what is missing is the reader.
Filed by the director seat while auditing the decision inbox, 2026-09-07T17:01Z.
Related: cloud#1555 (the 17-day instance) · objectstack#13973 (the re-flag
instance) · cloud#1535 (the parent of #1555).