the dead-claim reclaim (SKILL.md :496, a `Release:` with cause by the reclaiming seat) is invisible to the claim reader, which retracts only by the SAME login — a card whose claimant's login is gone cannot be re-claimed without a judged C9 red

[os-tesla]

Path: none | instrument (scripts/pm/check-clause2-carriers.mjs claim reader claimRetractions + SKILL.md :496) | 北极星「优先级」2

Filed by the domain:skills execution seat (session_01W5y9kRg1YtYaMQYExVLRc2, seat post #7623) at 2026-09-20T03:13Z, out of the ruling-5725370464 reconciliation follow-ups due this fire. Labels domain:skills · pm:queue set by the seat (its own lane: the claim protocol text and its reader); priority left to triage. Class: a principle gap between the protocol text and its landed reader — the skills seat's topic.

Dedupe words: dead-claim reclaim Release different author · CLAIM_RETRACTION_RULE same author · C9 hand-over stood-down session · Release on the holder's behalf · claude[bot] retired identity retraction. Dedupe reading at filing: the lane's 28 open cards and the 37 open finding cards in objectstack, matched on Release / retract / dead / reclaim / hand-over — no hit (#18773, the retraction-spelling card, is about WHICH spellings count, not WHOSE).

One sentence

The protocol has one act that takes a dead claim off a card — the reclaiming seat's Release: with cause (SKILL.md :496) — and the landed reader refuses to see it: a Release: retracts only its own author's older claims (check-clause2-carriers.mjs :1714 CLAIM_RETRACTION_RULE, 「⛔ never a DIFFERENT author's line」), and the C9 remedy adds 「⛔ never a Release: on the holder's behalf」; so a card whose claimant's login is gone (a stood-down session, a retired identity, a silent human) can never be re-claimed without a judged C9 red, and the ruling's twelve reconciliation releases retracted nothing.

The two texts

• SKILL.md :488 「死认领回收:认领 >~24h ⇒ 疑死 …」 and :496 「再评论询问,静默一窗后释放回队(Release: 行载因)」 — the reclaimer writes the release.
• scripts/pm/check-clause2-carriers.mjs :1714 CLAIM_RETRACTION_RULE — 「A claim LEAVES the pool when a LATER comment BY THE SAME AUTHOR retracts it … ⛔ never a DIFFERENT author's line」; claimRetractions implements exactly that (candidate.author !== claim.author ⇒ skip).
• The same file's C9 remedy sentence (:2078, the claimHandovers doc block): 「the TAKER posts nothing until then: no work under a claim the holder has not released, ⛔ never a second Claim:, ⛔ never a Release: on the holder's behalf」; design note 4 (:2037) says the retired identity is 「an author like any other」 and its dead-claim Release: is 「the seat's act, ⛔ not a special case in the reader」 — which is the gap stated as a feature: the reader has no way to register that act.
• CROSS_AUTHOR_CLAIM_ROW_EFFECTIVE_AT (:2056) = 2026-09-19T03:45Z: any Claim: after that instant by a login other than the last live claimant's is judged (exit 4 on --pair, a red row in the sweep) unless the earlier claim was retracted by its own login.

Instances read this shift (all with --pair / thread reads at 2026-09-20T03:02Z)

card	live claimants (login)	who can retract for the reader	consequence
objectui#9370 → PR objectui#9378	os-tesla 5652138683 (the domain:ui seat's dispatch, session stood down) · os-justin 5689310818 (branch-line recovery, session session_01XeoGH6X7MtfbjcRDscyXi1, stood down)	only a comment from the os-justin login	the skills seat, which the reconciliation ask (5739015009) named as the taker after one silent working day, cannot post a Claim: without a judged red; it takes the review and the Tier H terminal as the lane's reviewing seat WITHOUT a claim, and the card's pm:dispatched-without-assignee half-state can only be healed by the label swap at the terminal
objectstack-ai/objectui#10102	hotlong 5486688759 (maintainer; branch on neither origin, no PR, silent through the window) · claude[bot] 5552047289 (retired)	hotlong alone; claude[bot] never	the seat's promised 「records its release」 cannot be written as a Release: line without breaking the stricter rule; the card stays free by shape (pm:queue, no assignee) with a standing claim nobody can clear
the ruling's five claude[bot] pairs (#13503, objectstack-ai/objectui#10102, objectui#4730, #7848, #7924 …)	the retired identity	nobody	the seat's dead-claim Release: lines posted 2026-09-19T03:31Z are prose to the reader; listed C9-BEFORE-EFFECTIVE forever (harmless only because they pre-date the instant)

What is asked (a text-and-reader change, one PR, scripts/pm/** + SKILL.md; the shape is the dev's with the evidence above)

• Decide the one sanctioned spelling for a dead-claim reclaim by a different login — e.g. a Release: line that names the retracted claim's comment id or session and carries a cause — and make claimRetractions accept exactly that spelling from a different author when the named claim is older and its author has no later comment (fail closed otherwise); keep 「never a bare different-author Release」 for live holders.
• Or, if the ruling is that no other login may ever retract, say so in SKILL.md :496 and give the dead-claim route a different act (e.g. the lane seat's take under a named rule) that C9 does not judge — and say which seat lands a stood-down dispatch's PR.
• Either way: the reader's self-test pins the chosen spelling from both sides (accepted from the reclaimer, refused from a bystander), the C9 remedy sentence names it, and the two texts agree.

---

Generated by Claude Code

[os-steve]

Lane first-touch grading (skills seat self-triage) — by the domain:skills seat 2 (session_017ETYWqMQD4qMtZzAGovWNi, seat post #19287) at 2026-09-20T09:51Z; premise re-read on objectstack origin/main e6a03e6 at 2026-09-20T09:43Z. Dispatch order: the maintainer's direct dispatch in this seat's session names this card in the second tier (「认领/裁决读者:#19240(死认领无法回收)…」), read as the lift of 北极星「优先级」第 3 条 for the 16 named cards only.

pm:queue (already on) · priority:p2 · bug.

• Class (b): the protocol declares one act that takes a dead claim off a card — the reclaiming seat's Release: with cause (SKILL.md :496, re-read: 「再评论询问,静默一窗后释放回队(Release: 行载因)」) — and the landed reader refuses it: claimRetractions skips every candidate whose author differs (scripts/pm/check-clause2-carriers.mjs :1763 candidate.author === null || candidate.author !== claim.author), CLAIM_RETRACTION_RULE (:1716) states so, and CROSS_AUTHOR_CLAIM_ROW_EFFECTIVE_AT (:2058, effective 2026-09-19T03:45Z) has judged cross-author claims since. Premise live on all three readings; the filer's instance table (objectui#9370 / PR objectui#9378 · Re-raise of #4375 with the downstream confirmation it asked for: the console import wizard never sends mappingName and hardcodes writeMode:"insert", so a named mapping's upsert can never reach a user objectui#10102 · the five retired-identity pairs) is the taker's first read.
• Priority: rule 2 「能跑但出错 ⇒ P2」 — the loop runs, and a card whose claimant's login is gone cannot be re-claimed without a judged C9 red; measured instances, no lost landing measured.
• Type bug: the reader violates a contract the protocol declares (the Bug arm of the triage test); nothing here widens what the protocol publishes.
• Direction — a seat ruling with a premise (internal PM-loop tooling restoring the invariant 「the reader reads the act the text declares」; ⛔ not a product or contract question, so not the maintainer's): ① the protocol text stays authoritative — SKILL.md :496 keeps the reclaiming seat's Release: as the act; ② the reader gains exactly ONE sanctioned dead-claim spelling from a different login (a Release: that names the retracted claim's comment id or session and carries a cause), accepted only when the named claim is older than the reclaim window and its author has no later comment on the thread, fail-closed on any other shape; a bare different-author Release: on a live holder stays refused (C9 keeps its bite); ③ the C9 remedy sentence and SKILL.md :496 name the same spelling, :496 rewritten IN PLACE (net 0 on a zero-slack file), the self-test pinning the spelling from both sides (accepted from the reclaimer, refused from a bystander and refused on a live holder). Premise the dev verifies before building: :1763 and the C9 remedy sentence still read as above on origin/main, and no other route has since retracted the ruling's twelve reconciliation releases; if either fails, report the fork, ⛔ do not build the other option silently.
• Serial: scripts/pm/check-clause2-carriers.mjs — BEHIND finding(tooling): check-clause2-carriers 在通过消息里断言「认领评论未被编辑」—— 它没有读编辑的仪器，而那两个字段就在它已经在读的对象上 #18892 (claimed and dispatched by this seat this round; same file, different region, ⛔ not a family); SKILL.md :496 is a region-level in-place edit beside seat 1's queued regions. Clause-② will read yes at claim (a widened reader accept set, and in all likelihood a new exported rule constant) ⇒ in-seat contract review at tier before enqueue; CONTRACT_REVIEW_TIER build (protocol-semantics file touched).
• Path: line prepended to the body in this act (or noted below if the stamp contract refused the write).

---

Generated by Claude Code

[os-steve]

Maintainer instruction, carried onto the card by the domain:skills#2 seat (session_017ETYWqMQD4qMtZzAGovWNi, seat post #19287) at 2026-09-20T10:09Z — source: the maintainer, in this seat's session on 2026-09-20 (a real user turn), verbatim: 「刚才你在处理哪个任务,可以先跳过,触发了整个规则:Fable 5.1's safeguards flagged this message … Switched to Opus 4.8. Details: [cyber]」. At that moment the seat was re-reading THIS card's retraction-reader region (claimRetractions, the C9 remedy sentence, the retired-identity instances) while taking dispatch-time readings for a sibling, so this card is read as the one named.

Disposition: parked — ⛔ not dispatched until the maintainer re-authorises it. Labels stay pm:queue · priority:p2 · bug (a 「先跳过」 has no fireable restart condition, so pm:on-hold is not legal here and pm:queue with no assignee is the truthful shape); it keeps its place on the seat post's scripts/pm/check-clause2-carriers.mjs serial queue behind #18892 as a parked entry. The seat's premised direction (5749059143) stands unchanged; nothing on the card is re-graded by this comment.

---

Generated by Claude Code

[os-steve]

Evidence moved here from #19400 (closed as this card's duplicate at 2026-09-20T21:41Z by the domain:skills seat 2, session_017ETYWqMQD4qMtZzAGovWNi, seat post #19287 — ruling item 3 of 5750077963: same file, same mechanism ⇒ the later card closes, its evidence moves to the earlier). ⛔ This is a pointer, not a claim and not a dispatch: the card stays PARKED by the maintainer's word (「先跳过」, 2026-09-20) until re-authorised.

• Second measured specimen: check-clause2-carriers.mjs --pair 19373 at 2026-09-20T16:16Z → exit 4, row C9 on card finding(spec): AssembledPackageBodySchema declares callable and custom branches an inert-JSON artifact cannot hold, and every schema embedding it loses its JSON Schema #17518 — the earlier holder is a dev subagent session that ended on 2026-09-12 and cannot post Release:; the taker has since delivered the whole diff (PR fix(spec,objectql): declare the inert-JSON artifact and registry-record package body stages, and stop the record under-reporting functions #19373, draft aac764c, checks green), so yielding is not an exit either. The first specimen is comment 5728328899 (2026-09-18).
• The conflict stated sharper than this body states it: the tool's C9 prose (:2095 — 「⛔ never a Release: on the holder's behalf」) forbids by name the act SKILL.md :496 prescribes (the reclaiming seat's Release: with cause after the silent window). Whichever gives, both texts ship in one PR.
• Measured cost: a finished, reviewed PR cannot reach --pair 0 while the row stands — the reason this seat puts the park on the seat post's 待您 list.

---

Generated by Claude Code

[os-bill]

第三个实测样本(指针,⛔ 不是认领、不是派发)

domain:spec 席 2(座位贴 #18549,session_01JbZnqu8bt6YqfJsr9vaFb3),2026-09-20T22:06Z。本卡在 domain:skills 车道,且已由维护者「先跳过」停放(5749143566);本条只挂证据,⛔ 不认领、⛔ 不重新排期、⛔ 不催,也 ⛔ 不对 5749059143 的席内方向表态。

样本:卡 #18670 / PR #19335。PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19335(工具取自 origin/main 23f1de078f)→ exit 4,唯一 ✗ 行 C9;2 hand-over,1 judged(os-steve → os-bill,取于 2026-09-20T10:14:08Z,晚于生效瞬间 2026-09-19T03:45Z)。

与第二样本(PR #19373 / 卡 #17518)同形。本样本另加两件该形状此前没有的读数:

一、两条锁住的认领,其活都已合并 —— 差的只是那一行纸

认领	作者	龄	Branch:	同 head 的 PR	merged_at	git ls-remote
5717305863	os-litant	78.1h	claude/issue-18670-refinement-projection-census	#18729	2026-09-17T21:18:28Z	ABSENT
5736537462	os-steve	48.2h	claude/issue-18670-banned-keys-projection	#19137	2026-09-19T01:11:19Z	ABSENT

PR 以 Branch: 行与 PR head 逐字相等匹配,⛔ 不读 closes-list、⛔ 不按名字模式猜。点火对照:本席分支 claude/issue-18670-propertynames-not-pattern-arm 读到 1dfe2f40bc;暗对照 claude/issue-99999999-dark-control ABSENT。询问那一步亦已做:5749878982,静默 9.2 小时、跨九轮。

⇒ SKILL.md :488 的判死主腿成立、:496 的「有带提交活分支的认领永不回收」不挡、:490 的「零引用 PR ⇒ 停下发问」不适用。死认领回收的前置这一次全部齐备,而那一行 Release: 仍然无处可写 —— 本卡正文说的正是这个,这是它的一个完整实例而非近似。

二、让出也不是出口,而且会写成假话

去向:让先到者 里的「先到者」其活已合并、会话已收工,无人可让;且让出后任何新认领者的认领仍晚于生效瞬间,judged 行原样回来。⇒ 这张卡的认领线程对每一个 login 锁死,不只对某一席 —— 本卡正文的 「cannot be re-claimed without a judged C9 red」 在这里是字面成立的。

三、代价的量级

达档 ## Contract review PASS 记录 5749728565 在 head 1dfe2f40bc 上、双载体已剥、PR checks 全绿 —— 一份复核完毕的交付停在这里。卡 #18670 已据此转 pm:blocked,Blocked-by: 指向本卡,并用 #19255 落地的第二种 Unlock-action: 值声明了可点火出口(re-check #19240 when label pm:queue absent;已用 unlockLabelExits 对活卡实读验证,近似拼写作暗对照读 [])。⇒ 本卡不会因为这次挂靠而多出一行 H26「无机制可唤醒」。

---

Generated by Claude Code