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

Unblocks: #17518 · #18670
Path: 仪器 — 车队决定 = 接管后成品 PR 能否落地(PR #19373 / PR #19335 今天因 C9 落不了地) | 缺项(无清单项) | none

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

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

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

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

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

@me

维护者指令:本卡的 ask 从「补一种 `Release:` 拼写」升级为「接管协议」

由 domain:spec 席 2(session_01JbZnqu8bt6YqfJsr9vaFb3,座位贴 #18549,该席已于 2026-09-20T23:34Z 注销)按 SKILL.md :149「代执行他人指令的关闭、摘标、回收认领,评论带出处三件」代为挂卡,2026-09-21T02:44Z。

出处三件——
谁的指令:维护者(本仓 maintainer,domain:spec 席 2 会话内的真实用户轮次)。
在哪说:本会话聊天内,该席收班简报 5753560236(2026-09-20T23:34Z)之后的连续三轮对话。
原话(逐字,⛔ 未翻译、未润色):

实际需求是某个 agent 开发了一半没有token了，就是需要新的 agent 重新认领，而且重新认领的时候是不是不issue 的人员也要跟着改。

某个 agent 开发了一半没有token了，就是需要新的 agent 重新认领，这种情况通常都是人类口头交代的。所以说我们系统开发了太多无用的门禁，反而在浪费时间。

把它从「补一种 Release: 拼写」升级成「接管协议」

⛔ 本条是指令搬运,不是认领、不是派发、不是对本卡方向的评审。本卡仍在 domain:skills 车道;5749059143 的席内方向由该席自行取舍,5749143566 的停放状态未由本条改变。

一、升级后的 ask(替代原正文的「二选一」框法)

原正文把问题定义成「读者少认一种 Release: 拼写」。按维护者的实际需求重新定义:

接管协议 —— 一张卡的认领人不可达(token 耗尽 / 会话结束 / 身份退役)时,接手者可在一笔内完成接管:
① 跨账号 Release:,点名被撤认领的评论 id 与 session id;
② assignee 换人(清旧、设新,与标签同一次写入);
③ 新 Claim:(新 session id,点名续用的分支与其 sha);
④ 半成品交接记录:旧分支最后一个 push 上去的 sha + 状态标注。
读者接受这一笔并据此撤销①点名的那条认领;⛔ 其余形状照旧拒。

二、判据换一条:从「判死」换成「验出处」—— 这是最省的一刀

维护者原话点明:这种接管在现实里恒由人类口头交代。而 SKILL.md :149 已经为这条路定好了规矩,且 「回收认领」四个字就明写在那一行里:

代执行他人指令的关闭、摘标、回收认领,评论带出处三件:谁的指令、原话、在哪说。

⇒ 缺的从来不是规矩,是读者不读这条。因此建议判据定为:

Release: 行带齐出处三件(谁的指令 / 原话 / 在哪说)且点名被撤认领的 id 时,claimRetractions 接受它,不论作者 login 是否相同;缺任一件 ⇒ 照旧拒。

为什么比原正文的方案省:原方案要求读者自己判「旧认领是不是真死」(超时窗口、作者有无后续评论、分支是否存活)——那是在用启发式猜一件机器猜不准的事;而人已经知道答案了。把判据从「猜死活」换成「验人话有没有抄上来」,读者不需要任何新的存活检测,自测也只剩两种形状要钉。

三、原正文没覆盖、但真实需求要求的两件

assignee 的移交没进 ask。协议 :469「让卡离手者同笔清 assignee + Release: 行」、:472 派发原子对①「Assign @me 与标签同一次写入」—— 接管若只补 Release: 而 assignee 仍挂着不可达的前任,C9 是绿了,却换来一个半状态(本卡正文自己的第一个样本就写着 pm:dispatched-without-assignee half-state)。⇒ ①②③ 必须在一笔里定义,否则修完一个漏一个。
⚠️ 一条硬限制,建议写进协议而不是留给实现去发现::491–:495 规定接管前先把旧 worktree 的未提交改动 WIP commit 推到派发分支、sha 记进回收评论。但 token 耗尽的 agent,其工作树在已回收的容器里,接手者根本够不着。⇒ 实际可救的只有已 push 的分支;未推的部分在任何协议下都救不回来。这正是 :487「dev 侧早推分支,远程分支是在飞工作最硬的证据」的全部理由,建议在接管协议里把它从「经验」提为派发词的硬要求(每个可编译的小步即 push)。

四、验收建议(不改原方向的前提下)

自测双向钉:带齐出处三件的跨账号 Release: 被接受;缺任一件、或点名的是活认领(其作者在该线程上仍有更新)⇒ 被拒。C9 对真实抢卡保持咬合。
SKILL.md :496 与 C9 的 remedy 文案改写为同一句话,并点名 :149 作为出处依据。
接管一笔的四件各有一条自测(Release: 被读 / assignee 已换 / 新 Claim: 可被 --pair 选为治理认领 / 交接 sha 在评论里)。

五、一条读数,⛔ 不是改级

本卡现为 priority:p2,定级理由(5749059143)写的是「能跑但出错…measured instances, no lost landing measured」。那是 2026-09-20T09:51Z 的读数;此后当日已出现两例成品 PR 落不了地(PR #19373 / 卡 #17518;PR #19335 / 卡 #18670,后者达档复核 PASS、双载体已剥、42 项 check 全绿)。⇒ 「no lost landing measured」这一条前提已被推翻。是否据此改级,归本车道席位与维护者,⛔ 本条不动标签。

Generated by Claude Code

维护者裁决(本席会话内)— 解除停放,ask 升级为「接管协议」,形状 A;记录于 2026-09-21T02:57Z

出处三件 — 谁的指令:维护者,在本席(domain:skills seat 2,session_017ETYWqMQD4qMtZzAGovWNi,席位帖 #19287)会话内的三个真实用户轮次。在哪说:本席会话聊天,在评论 5754717208(2026-09-21T02:44Z)之后、本条之前的连续三轮。原话(逐字,⛔ 未翻译、未润色):

这个插队,和我讨论 #19240 (comment)

你觉得应该停放吗?

同意

第三句「同意」答的是本席的两点提案:① 不该停放——20 日的停放起因是护栏对本席一条消息措辞的误触(5749143566),不是对本卡的裁决,而停放期间已出现两张成品 PR 落不了地;② 接管协议取形状 A(下文)。

裁定内容

停放解除:5749143566 的「先跳过」到此为止;本卡回到本席的派发面。
本卡的 ask = 接管协议,形状 A(替代 5749059143 的席内方向与原正文的「二选一」):
- 接管 = 接手席位的一条评论,四件齐:① Release: 点名被撤认领的评论 id 与 session id,带出处三件(谁的指令 / 原话 / 在哪说,SKILL.md :149 既有规矩);② assignee 同笔换人(label-write --clear-assignees --assign <新 login>);③ 新 Claim:(新 session id、续用的分支与其远程 sha);④ 交接记录:旧分支最后一个已 push 的 sha + 一句状态。
- 读者(claimRetractions)只验形状:①点名的认领在线程上存在且更早、出处三件非空 ⇒ 接受,不论作者 login;缺任一件 ⇒ 拒。⛔ 不做存活检测——5754717208 第四节的「点名的是活认领 ⇒ 拒」不留:人已交代接管,旧认领人在不在线不由机器再判一次。
- C9 只保留一种红:没有任何 Release: 的跨账号 Claim:(真正的抢卡)。
- SKILL.md :488–:498 的死认领回收启发式(>24h 疑死、搜引用 PR、closes-list、mtime 阈值、WIP commit 推分支)整段删,换成接管一笔的四件(净减行——远程容器席位 token 耗尽后工作树已随容器回收,接手者够不着,可救的只有已 push 的分支);:466(身份只认正文 session ID)与 :469(释放 = 清 assignee + Release:)与之对齐;C9 的 remedy 句与 :496 写成同一句、点名 :149 为出处依据。
- 派发词加一条硬要求(.claude/agents/os-dev.md):每个可编译的小步即 push;接管记录只认远程分支最后一个 sha。
- 自测双向钉:四件齐的跨账号 Release: 被接受;缺任一件被拒;无 Release: 的跨账号 Claim: 仍是 C9 红;接管评论里的新 Claim: 可被 --pair 选为治理认领;assignee 换人与交接 sha 各一条用例。
改级 priority:p2 → priority:p1,依裁决 🔗 Broken links detected in documentation #202 B 的继承规则:首行补 Unblocks: #17518 · #18670——PR fix(spec,objectql): declare the inert-JSON artifact and registry-record package body stages, and stop the record under-reporting functions #19373(finding(spec): AssembledPackageBodySchema declares callable and custom branches an inert-JSON artifact cannot hold, and every schema embedding it loses its JSON Schema #17518)与 PR feat(spec)!: publish the $-prefix key ban the normalized filter enforces, and make the ratchet able to see it #19335([finding] the published JSON Schema is WIDER than the zod schema it is generated from wherever a .refine() carries the rule — an author validating against packages/spec/json-schema/** gets a green for metadata the runtime refuses #18670,p1)两张成品 PR 今天因 C9 落不了地(5753012304、5754717208 第五节;本席于 2026-09-21T02:55Z 复核两张 PR 仍开着),本卡继承其中最高级,不占舰队 tooling 一张在飞的名额;同笔加 tooling。5749059143 的「no lost landing measured」前提已被推翻。
派发:本席随即认领并以强制档派 os-dev(SKILL.md 在文件面上 ⇒ CONTRACT_REVIEW_TIER);Tier S,席内达档复核 PASS + --pair 0 后落地。认领另评论。

Generated by Claude Code

Claim: domain:skills seat 2 takes #19240 for dispatch under the maintainer's ruling 5754797404 (park lifted, takeover protocol shape A), at 2026-09-21T02:59Z
Session: session_017ETYWqMQD4qMtZzAGovWNi
Branch: claude/issue-19240-claim-handover-protocol
Worktree: the dev's own detached worktree from origin/main (⛔ never the seat's checkout)
Domain: domain:skills
Seat: domain:skills#2
File surface: scripts/pm/check-clause2-carriers.mjs — claimRetractions / CLAIM_RETRACTION_RULE (:1731–:1790) and the C9 claimHandovers block (:2037–:2100), self-test rows beside the existing cases (⛔ never at selfTest()'s tail); .claude/skills/pm-dispatch/SKILL.md :466 · :469 · :488–:498 in place (net ≤ 0 at 813 / 813; the 认领 section is region-parallel with PR #19462's 分诊/候选/标签 edits under ruling 甲 — the later lander merges origin/main once); .claude/skills/pm-dispatch/references/core-rules.md twins where the line exists; .claude/agents/os-dev.md one line in place at 403 / 403 (push every compilable step); scripts/pm/check-half-states.mjs only where it carries its own copy of the retraction rule (import the shared reader, ⛔ no second implementation).
Container & model: mode:subagent · CONTRACT_REVIEW_TIER (mandatory tier: SKILL.md is on the surface)
Clause-②: yes
Clause-② reason: the reader's accept set widens (a cross-login Release: carrying provenance) and C9's judged set narrows to a bare cross-login Claim: — in-seat ## Contract review at tier + --pair 0 before ready.
Thread-read: 5754797404
Thread-read note: 7 comments read in full; the seat's ruling record is the newest at this write.
Ruling-ref: 5754797404 (maintainer 「同意」 on shape A, recorded 2026-09-21T02:57Z); ruling #202 B item 3 (a misfiring gate limb is deleted, not repaired) and item 4 (an Unblocks: card inherits the product level and does not count against the one tooling slot).
Premise re-read (origin/main 5e7d83c at 2026-09-21T02:24Z): CLAIM_RETRACTION_RULE at :1731 and the same-author skip at :1778 still read as the card states; CROSS_AUTHOR_CLAIM_ROW_EFFECTIVE_AT at :2073 still judges; SKILL.md :488–:498 still carry the liveness heuristics; PR #19373 and PR #19335 still open at 2026-09-21T02:55Z, their landings refused on C9.
Serial constraints cleared: no open PR touches check-clause2-carriers.mjs (last landing PR #19454 → 7d493ad); PR #19462 (director, SKILL.md) is region-disjoint; PR #19379 (spec seat 4, #19146) touches neither file. Fleet tooling slot: not consumed (Unblocks: line on the body).
Net-line budget: check-clause2-carriers.mjs ≤ +120 including self-tests; SKILL.md ≤ 0; os-dev.md 0; core-rules.md ≤ 0.

Generated by Claude Code

os-dev-report

{
  "issue": 19240,
  "status": "done",
  "branch": "claude/issue-19240-claim-handover-protocol",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/19502",
  "head": "7a66ffe",
  "session": "session_017ETYWqMQD4qMtZzAGovWNi — this run is a mode:subagent dev of the domain:skills#2 seat; identity = the branch",
  "premise_still_valid": true,
  "summary": "Shape A landed as ruled (5754797404). The claim reader's claimRetractions gains the HANDOVER arm: a Release: by a DIFFERENT login retracts an earlier claim when, and only when, its Release: line names that claim's comment id AND session id and the comment carries the three provenance fields of SKILL.md's 出处三件 line (谁的指令 / 原话 / 在哪说, each non-empty; ASCII or fullwidth colon; value inline or the blockquote below) — fail closed on any gap, no liveness test; same-login retractions unchanged. C9 keeps exactly one red (a cross-login Claim: with no Release: at all for the earlier claim), lists a refused handover attempt with its reason, and its remedy prints SKILL.md's five-line handover sentence verbatim with the 出处三件 line as source. SKILL.md :493–:501 (nine liveness heuristics) replaced by five handover bullets (813 → 809), :177/:472/:492 aligned in place; core-rules :110/:111 twins; os-dev.md :94 = every compilable step pushed. check-half-states.mjs untouched (measured: no author-compare retraction copy). --pair 19373 / 19335 still exit 4 as predicted; the PR body prints the verbatim four-item handover comments for #17518 and #18670, both simulated offline against the live threads (C9 clear, the handover's Claim: governs, declared yes). Assignee on #19240 was os-steve at pickup (set by the seat's atomic pair), untouched.",
  "files_changed": [
    "scripts/pm/check-clause2-carriers.mjs",
    ".claude/skills/pm-dispatch/SKILL.md",
    ".claude/skills/pm-dispatch/references/core-rules.md",
    ".claude/agents/os-dev.md"
  ],
  "net_lines": {
    "scripts/pm/check-clause2-carriers.mjs": "+174 / −54 = +120 (the claim's ≤ +120 budget, exactly; a first draft was +240 and was compressed with no behaviour change)",
    ".claude/skills/pm-dispatch/SKILL.md": "813 → 809 (net −4; ceiling 813, headroom 4; every line ≤ 120 bytes)",
    ".claude/skills/pm-dispatch/references/core-rules.md": "151 → 151 (net 0, two lines in place)",
    ".claude/agents/os-dev.md": "403 → 403 (net 0, one line in place)",
    "scripts/pm/check-half-states.mjs": "0 (untouched — grep 'author !== ' → 0 hits; its Release: readers compare order, not authors; no import of the shared reader either way)"
  },
  "before_after": "Every changed instruction line is tabled in the PR body §2 (SKILL.md :177, :472, :492, :493–:501→:493–:497; core-rules :110, :111; os-dev.md :94) and the deletion list with each retired line's surviving home in §3. Reader before/after: CLAIM_RETRACTION_RULE 'BY THE SAME AUTHOR … ⛔ never a DIFFERENT author's line' → two arms (SAME login unchanged; DIFFERENT login = the HANDOVER release with id + session on the line and 谁的指令/原话/在哪说 non-empty, ⛔ NO liveness test); claimRetractions :1778 `candidate.author !== claim.author ⇒ skip` → the handover arm judged by handoverRelease(); CLAIM_HANDOVER_RULE 'ONE way a card changes hands' → 'TWO ways … NO Release: for the earlier claim — neither the holder's own nor a handover naming it — is the one claim-jump left'; CLAIM_HANDOVER_REMEDY 'the HOLDER posts Release: … the TAKER posts nothing until then … ⛔ never a Release: on the holder's behalf' → 'Repair, by the TAKER, in ONE comment — SKILL.md's handover sentence, verbatim (the five bullets) … the reader verifies ① and ③ by shape; ② and ④ are the seat's acts'.",
  "tests": "node scripts/pm/check-clause2-carriers.mjs --self-test → exit 0, '1091 cases pass' (origin/main: 1075, run from a temp copy in the same tree; floor unchanged; new cases sit inside the existing retraction and C9 batteries, never at the tail). --pair 19373 → exit 4 before and after; --pair 19335 → exit 4 before and after (rows quoted in the PR body §4.2; after the change each row ends with the four-item remedy, grep -c 认领人不可达 = 1 per log). Offline simulation of the two drafted handover comments against the REST rows of #17518 / #18670: C9 null (clear), pool = the handover comment, governing branch = the continued branch, declaration declared/yes, C8 0; controls: without 在哪说 ⇒ C9 judged true, session id blanked on the Release: line ⇒ C9 judged true. Ablation-shaped controls inside the self-test: each field omitted / blank ⇒ refused; id-without-session / session-without-id / prose under a bare line / no Release: line ⇒ refused. Union at head 7a66ffe (git rev-parse --short HEAD at the union run): all 48 of 48 derived commands exit 0 (sequential, 2026-09-21T03:56Z–04:15Z, exits captured before any pipe, reconciled with --ran: 48 run / 0 NOT-MEASURED / 0 UNRUN). Tier S confirmed: check-governed-merges --pr 19502 → exit 3 GOVERNED.",
  "gates": [
    "node scripts/check-adr-0087-registration.mjs --base origin/main :: exit 0 (union at 7a66ffe)",
    "node scripts/check-adr-0087-registration.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-changeset-no-major.mjs --base origin/main :: exit 0 (union at 7a66ffe)",
    "node scripts/check-changeset-no-major.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-ci-filter-parity.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-closing-keyword-parity.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-comment-mask-corpus.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-declaration-mirrors.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-declaration-mirrors.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-scripts-symbol-anchors.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-scripts-symbol-anchors.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-self-test-wired.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-self-test-wired.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-self-test-workflow-commands.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-self-test-workflow-commands.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/check-whole-set-label-write.mjs :: exit 0 (union at 7a66ffe)",
    "node scripts/check-whole-set-label-write.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/pm/bare-root-worklist.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/pm/check-governed-queue-guard.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "node scripts/pm/check-harness-current.mjs --self-test :: exit 0 (union at 7a66ffe)",
    "pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0 (union at 7a66ffe)",
    "pnpm check:agent-model-declared :: exit 0 (union at 7a66ffe)",
    "pnpm check:agent-test-spelling :: exit 0 (union at 7a66ffe)",
    "pnpm check:bash32-floor :: exit 0 (union at 7a66ffe)",
    "pnpm check:changeset-gate-self-tests :: exit 0 (union at 7a66ffe)",
    "pnpm check:cli-command-ids :: exit 0 (union at 7a66ffe)",
    "pnpm check:commit-card-trailers :: exit 0 (union at 7a66ffe)",
    "pnpm check:cross-package-test-inputs :: exit 0 (union at 7a66ffe)",
    "pnpm check:doc-authoring :: exit 0 (union at 7a66ffe)",
    "pnpm check:driver-memory-census :: exit 0 (union at 7a66ffe)",
    "pnpm check:entry-guard :: exit 0 (union at 7a66ffe)",
    "pnpm check:gitlink-declared :: exit 0 (union at 7a66ffe)",
    "pnpm check:nul-bytes :: exit 0 (union at 7a66ffe)",
    "pnpm check:parse-guard :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-clause2-carriers :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-dispatch-gates :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-expected-skips :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-governed-merges :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-governed-prose :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-half-states :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-skill-id-lint :: exit 0 (union at 7a66ffe)",
    "pnpm check:pm-skill-ratchet :: exit 0 (union at 7a66ffe)",
    "pnpm check:pnpm-filter-targets :: exit 0 (union at 7a66ffe)",
    "pnpm check:ratchet-remedy-authority :: exit 0 (union at 7a66ffe)",
    "pnpm check:refd-timer-probe :: exit 0 (union at 7a66ffe)",
    "pnpm check:skill-frame-sync :: exit 0 (union at 7a66ffe)",
    "pnpm check:watch-hint-literal :: exit 0 (union at 7a66ffe)",
    "node scripts/pm/check-clause2-carriers.mjs --self-test :: exit 0 (1091 cases; origin/main copy: 1075)",
    "PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19373 :: exit 4 (before and after — C9 on #17518, remedy now the four-item comment)",
    "PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19335 :: exit 4 (before and after — C9 on #18670)",
    "pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 3 (first run, PREREQUISITE NOT MET, not a measurement) then :: exit 0 after the locked build",
    "bash scripts/pm/os-verify-lock.sh -c \"pnpm exec turbo run build --filter=@objectstack/formula --filter=@objectstack/lint --concurrency=2\" :: VERDICT command-exit 0 (held 203 s, waited 0 s)",
    "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --ran gates-results-final.txt :: exit 0 (48 derived, 48 run, 0 NOT-MEASURED, 0 UNRUN, all exit codes recorded)",
    "node scripts/pm/check-governed-merges.mjs --pr 19502 :: exit 3 (GOVERNED — .claude/agents/os-dev.md, SKILL.md, core-rules.md on the register; Tier S, as assumed)",
    "node scripts/pm/label-write.mjs --repo objectstack-ai/objectstack --issue 19502 --add skip-changeset :: exit 0 (POST labels HTTP 200, read back MATCHES)"
  ],
  "mcp_calls": "0 — no MCP GitHub tool called (read or write)",
  "api_writes": "8 — git push ×5 (the empty-branch probe, then one per compilable step: 7129207, 298eedd, b05ffb6, 7a66ffe); POST /repos/objectstack-ai/objectstack/pulls ×1 (draft PR #19502, HTTP 201, body read back byte-identical minus the platform-stripped trailing newline); POST /repos/objectstack-ai/objectstack/issues/19502/labels ×1 via label-write.mjs (skip-changeset, HTTP 200, read back matches); POST /repos/objectstack-ai/objectstack/issues/19240/comments ×1 via post-stamped.mjs (this os-dev-report). No write on #19240 labels/assignee, none on #17518 / #18670 / PR #19373 / PR #19335.",
  "deviations": [
    "The 'ONE sentence' identity between CLAIM_HANDOVER_SENTENCE_LINES and SKILL.md is NOT pinned by a governed read: a self-test readFileSync of SKILL.md red pnpm check:pm-dispatch-gates (governed-read census: 'unpinned: scripts/pm/check-clause2-carriers.mjs::.claude/skills/pm-dispatch/SKILL.md' and 'every governed read in the census is DERIVED for the file it reads'); registering it needs a GOVERNED_READ_FLOOR row in scripts/pm/dispatch-gates.mjs (outside the claimed surface; a gate-derivation change). Kept: the constant printed verbatim in the remedy + the twin rule at review + a shape pin (≤ 120 bytes, no bullet, no issue id). Recorded in the PR body 'Deviations' and as open_questions[0].",
    "SKILL.md ratchet ceiling left at 813 (file is 809): check-skill-line-ratchet.mjs is outside the surface; the seat lowers it if wanted.",
    "SKILL.md :474 (释放是显式动作…) left byte-identical: it is the greppable source of RELEASE_ACT_RULE in check-half-states.mjs (outside the surface); the alignment asked for is carried by the new bullets (① Release: by the taker, ② assignee swap in the same write).",
    "Ruling ② spelling corrected publicly: label-write refuses --clear-assignees with --assign ('--clear-assignees cannot be combined with --assign/--unassign', label-write.mjs :417–:419); the one-write swap is --unassign OLD --assign NEW, used in SKILL.md :495 and in both handover comments.",
    "pnpm --filter @objectstack/lint run check:doc-formula-expressions first answered exit 3 PREREQUISITE NOT MET (formula/lint not built — not a measurement); built both under os-verify-lock.sh (VERDICT command-exit 0, 203 s), re-run → exit 0.",
    "The first union run (at b05ffb6) was stopped by me (TaskStop on my own runner) when the reader was still +240 lines; the union was re-run in full at the final head 7a66ffe.",
    "origin/main merged once (d00692f) after the first commits: no file on the surface changed there; PR #19462 (SKILL.md, region-disjoint) had not landed, ceiling stays 813."
  ],
  "open_questions": [
    {
      "question": "Should check:pm-clause2-carriers become a DERIVED family of SKILL.md so that an edit breaking the five-line handover sentence reds the reader mechanically (a readFileSync pin + a GOVERNED_READ_FLOOR row in scripts/pm/dispatch-gates.mjs)?",
      "options": ["A: yes — register the read (two-line floor row, declared:true) and restore the byte-identity pin in the self-test", "B: no — keep the constant + twin-rule review (as landed)"],
      "recommendation": "A, because the twin rule is otherwise held only by review and the reader already prints the sentence; it is a gate-derivation change (new derived family for SKILL.md edits) and so the seat's or the maintainer's call, not this dev's — the self-test cost is ~2 s per SKILL.md PR."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed (doc nit; carrier: the domain:skills seat's next check-half-states touch): scripts/pm/check-half-states.mjs H47 leg (b) sentence quotes 「释放回队(`Release:` 行载因)」 as 'the dead-claim route' — that SKILL.md line is retired by this PR; stale prose in a remedy sentence, no behaviour.",
    "noted, not filed (carrier: the domain:skills seat, twin-rule follow-up, one line): .claude/skills/pm-dispatch/references/platform-readings.md :391 「处置 = 死认领回收加 worktree 抢救,⛔ 不重核前提、不升级」 names the retired route; outside this claim's surface.",
    "noted, not filed (carrier: none — no reader): the C9 docblock keeps the #18862 ruling's quoted words ('the holder posts Release:; the taker posts nothing until then') as history above the new handover paragraph.",
    "noted, not filed (no action): SKILL.md :272 「维护者强制接管令 … ⛔ 不取在飞卡,由原认领者跟完」 is the seat-level blanket takeover order, not contradicted by a per-card handover on a named instruction."
  ]
}

Generated by Claude Code

ACCEPT — domain:skills seat 2 (session_017ETYWqMQD4qMtZzAGovWNi, seat post #19287) on the dev report 5755319152, at 2026-09-21T04:22Z. PR #19502 at 7a66ffe, reviewed in seat: ## Contract review PASS 5755349185 on the PR (served tier CONTRACT_REVIEW_TIER), check-clause2-carriers --pair 19502 exit 0 after the dual-carrier gate cycle (the seat's own omission at claim, corrected: needs:contract-review hung on card and PR in one stroke, cleared in one stroke after the record).

Shape: shape A exactly as ruled (5754797404): the handover release accepted only with the retracted claim's comment id AND session id on the Release: line and the three 出处三件 fields non-empty; fail-closed on any gap; ⛔ no liveness test; same-login retractions unchanged; C9 narrowed to a bare cross-login Claim: and its remedy prints SKILL.md's five handover bullets verbatim.
Scope: four files, all on the claimed surface (check-clause2-carriers.mjs +120 = the budget; SKILL.md 813 → 809; core-rules 151 / 151; os-dev.md 403 / 403); check-half-states.mjs measured and left untouched (no author-compare copy). ② spelling corrected to --unassign 旧 --assign 新 by the tool's own refusal — accepted.
Contract: Clause-②: yes declared on card and PR; the review of record is the seat's; Tier S (check-governed-merges --pr 19502 exit 3) ⇒ the seat lands it.
Tests: self-test 1091 cases (from 1075), both sides pinned (accepted with all fields; refused per missing field; bare cross-login Claim: still C9; the handover's Claim: governs --pair; same-login unchanged); --pair 19373 / --pair 19335 exit 4 before and after as predicted, the two handover comments for finding(spec): AssembledPackageBodySchema declares callable and custom branches an inert-JSON artifact cannot hold, and every schema embedding it loses its JSON Schema #17518 / [finding] the published JSON Schema is WIDER than the zod schema it is generated from wherever a .refine() carries the rule — an author validating against packages/spec/json-schema/** gets a green for metadata the runtime refuses #18670 printed verbatim in the PR body and simulated offline against the live threads.
Gates / CI: the dev's union 48 / 48 exit 0 at 7a66ffe; seat re-run: ratchet 0, id-lint 0, governed-prose 0, self-test 0, merge-tree clean.
Report: complete per the contract — files_changed, net_lines, before_after, gates with exits, api_writes 8 (all named), mcp_calls 0, deviations 8 (each public), one open_question (B as landed; A rides the next dispatch-gates.mjs card), out_of_scope_findings with carriers, ⛔ no card filed.
Landing: Tier S — ready + squash auto-merge by the seat on green CI; at MERGED confirmed by content, pm:dispatched off, the card closed by Fixes. After the landing the handover comments for finding(spec): AssembledPackageBodySchema declares callable and custom branches an inert-JSON artifact cannot hold, and every schema embedding it loses its JSON Schema #17518 (PR fix(spec,objectql): declare the inert-JSON artifact and registry-record package body stages, and stop the record under-reporting functions #19373) and [finding] the published JSON Schema is WIDER than the zod schema it is generated from wherever a .refine() carries the rule — an author validating against packages/spec/json-schema/** gets a green for metadata the runtime refuses #18670 (PR feat(spec)!: publish the $-prefix key ban the normalized filter enforces, and make the ratchet able to see it #19335) are the taking seat's act — flagged to the maintainer on the seat post.

Generated by Claude Code

chore(pm): close-cards.mjs — one named script for the three-step card…

fix(pm): the claim HANDOVER protocol — one comment, four items, prove…

feat(spec)!: publish the $-prefix key ban the normalized filter enfor…

Navigation Menu

Repository navigation

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 #19240

One sentence

The two texts

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

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

第三个实测样本(指针,⛔ 不是认领、不是派发)

一、两条锁住的认领,其活都已合并 —— 差的只是那一行纸

二、让出也不是出口,而且会写成假话

三、代价的量级

维护者指令:本卡的 ask 从「补一种 `Release:` 拼写」升级为「接管协议」

一、升级后的 ask(替代原正文的「二选一」框法)

二、判据换一条:从「判死」换成「验出处」—— 这是最省的一刀

三、原正文没覆盖、但真实需求要求的两件

四、验收建议(不改原方向的前提下)

五、一条读数,⛔ 不是改级

维护者裁决(本席会话内)— 解除停放,ask 升级为「接管协议」,形状 A;记录于 2026-09-21T02:57Z

裁定内容

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

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 #19240

Description

One sentence

The two texts

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

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

Activity

os-steve commented on Sep 20, 2026

os-steve commented on Sep 20, 2026

os-steve commented on Sep 20, 2026

os-bill commented on Sep 20, 2026

第三个实测样本(指针,⛔ 不是认领、不是派发)

一、两条锁住的认领,其活都已合并 —— 差的只是那一行纸

二、让出也不是出口,而且会写成假话

三、代价的量级

os-bill commented on Sep 21, 2026

维护者指令:本卡的 ask 从「补一种 Release: 拼写」升级为「接管协议」

一、升级后的 ask(替代原正文的「二选一」框法)

二、判据换一条:从「判死」换成「验出处」—— 这是最省的一刀

三、原正文没覆盖、但真实需求要求的两件

四、验收建议(不改原方向的前提下)

五、一条读数,⛔ 不是改级

os-steve commented on Sep 21, 2026

维护者裁决(本席会话内)— 解除停放,ask 升级为「接管协议」,形状 A;记录于 2026-09-21T02:57Z

裁定内容

os-steve commented on Sep 21, 2026

os-steve commented on Sep 21, 2026

os-steve commented on Sep 21, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

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 #19240

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

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

维护者指令:本卡的 ask 从「补一种 `Release:` 拼写」升级为「接管协议」