C9 has no reachable repair when the earlier claim holder is a dead dev session — and `check-clause2-carriers.mjs:2095` and `SKILL.md:496` prescribe opposite acts for exactly that case

[os-litant]

Class (b) — two rules in this repository prescribe opposite acts for one state, with both texts quoted. Filed under 「两条细则冲突 ⇒ 按更严的一条行动并立卡」: this seat acted by the stricter one (posted nothing) and is filing rather than settling it in place.

Dedupe words: C9, check-clause2-carriers, dead claim, Release on the holder's behalf, 死认领回收.

The state, measured on a live card

PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 19373, run 2026-09-20T16:16Z, exit 4, one adverse row:

✗ C9 — card #17518 (delivering open PR #19373) — 2 authors hold LIVE claim comments here with no Release: from the earlier holder between them — os-bill's 5646971772 at 2026-09-12T15:54:12Z is the claim that stood; os-litant's 5749581295 at 2026-09-20T11:43:41Z took the card from os-bill (dated AFTER the effective instant 2026-09-19T03:45Z — JUDGED)

The earlier holder is a dev subagent session (session_01MkQhmuuJAVDjmeWNixwDDH, named in its own claim line) that ran on 2026-09-12, posted its round's report the same hour, and ended. It cannot post anything. Between then and the retake the card went back through needs-user-decision twice and was re-ruled twice.

⚠️ This is not novel on this card. On 2026-09-18 a different seat hit the same row and posted comment 5728328899, headed "Claim supersession, recorded because the machine reading says the opposite." — it recorded the state instead of repairing it, because the repair was already unavailable then.

The two texts, quoted

scripts/pm/check-clause2-carriers.mjs:2095, inside C9's own repair prose:

⛔ never a second Claim:, ⛔ never a Release: on the holder's behalf.

and, above it, that the repair belongs to "the seats that hold the claims and ⛔ by nobody else", the holder posting Release: being "the one hand-over the protocol sanctions".

.claude/skills/pm-dispatch/SKILL.md:496, the last line of 死认领回收:

再评论询问,静默一窗后释放回队(Release: 行载因);有带提交活分支的认领永不回收。

In the reclamation flow that line sits in, the seat doing the reclaiming is the one that posts, after the silent window — which is a Release: on the dead holder's behalf, the act :2095 forbids by name. ⛔ One of the two has to give, and this seat is not the place that decides which.

Why it does not simply resolve itself

The tool offers exactly two exits from C9 and each is unavailable here:

exit	why it is unavailable
the HOLDER posts Release:	the holder is a subagent session that ended 2026-09-12
the TAKER posts its own Release: with 去向「让先到者」	the taker has since delivered the card's whole diff — PR #19373, 15 commits, all 35 checks green. Yielding would leave a finished PR held by a dead session

And the row is deliberately new-only: a taking claim dated at or before 2026-09-19T03:45Z is listed informational and moves no exit. So the rule was built to bite exactly the hand-overs made from that instant on — which is right — but a card whose earlier holder died before that instant can only be taken after it, and every such take is permanently JUDGED.

What it costs, concretely

Pre-landing check ② (references/contract-review.md) reads:

② 双载体已清,逐对机读 … --pair N。0 = 双肢一致、无放宽 tell、head 上有记录;4 = 任一不成立

⇒ a card in this state cannot reach 0, so a strict reading parks its PR forever. The stricter-of-two rule then points at not landing, which makes an unrepairable bookkeeping row block finished, green, reviewed work. ⚠️ That consequence is what makes this worth a card rather than a note: the conflict is load-bearing on a landing gate.

Shape of a fix — suggestions, ⛔ not a ruling

• Give C9 a third sanctioned exit for a provably-dead holder, with its own evidence bar (the holder's claim names a session; that session's round reported and the card then left the dev's hands through a documented state change), written as a distinct marker so it is never confused with a holder's own Release: — the way Clause-②-correction: is distinct from a second Claim:.
• Or make C9's judgement follow the CARD's state rather than comment adjacency: a claim whose card has since passed through needs-user-decision is not a live claim, and here it did so twice.
• Whichever is chosen, SKILL.md:496 and the tool's C9 prose ship the change together — 「一条规则在本文与核心条款一处改动,另一处同 PR 同改」.

Provenance

Measured by this seat while landing PR #19373 (card #17518) on 2026-09-20. Both quoted lines were read first-hand from origin/main at c5d3d1d8b69; the C9 token occurs 37 times in the checker (lit control) and C99ZZZ 0 times (dark control). Filed-by: session_01LvwGppdonww4zGLWZo5rho (domain:spec execution seat 1) — a governed-surface and PM-tooling finding for the skills lane; ⛔ not graded, prioritised or dispatched by this seat, and ⛔ neither text was edited.

---

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 2026-09-20T21:41Z; premise re-read on origin/main b71d9e7 at 2026-09-20T21:27Z, thread read to its last comment in the same act. Grading is the seat's mechanical duty each fire (lanes/skills.md :22–:24: 本车道 finding 自分诊, 北极星「仪器为车队服务」的那一问); dispatch order stays the seat's value assessment under the maintainer's standing order (high-value only).

finding → CLOSED as a duplicate of #19240 (ruling item 3, 5750077963: same file + same mechanism ⇒ the later card closes duplicate_of the earlier, evidence moved to the earlier).

• Same file, same mechanism: 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 (bug · p2 · pm:queue, filed 2026-09-19) — 「the dead-claim reclaim (SKILL.md :496, a Release: with cause by the reclaiming seat) is invisible to the claim reader」 — is this card's defect: C9 in scripts/pm/check-clause2-carriers.mjs offers no exit when the earlier holder is a dead dev session, and the reclaim the charter prescribes (:496) is the act the tool's prose forbids by name (:2095). This card adds the second measured specimen (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, exit 4 at 2026-09-20T16:16Z, comment 5728328899 of 2026-09-18 as the first) and the sharper statement of the conflict; both are pointed to from 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 in the same stroke.
• Why not merged the other way: 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 is the earlier card and already carries the SKILL.md :496 half; the 「third sanctioned exit for a provably-dead holder」 and 「follow the CARD's state」 shapes here are the design options its taker weighs, ⛔ not a separate defect.
• State of the family, stated so nobody re-derives it: 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 is PARKED by the maintainer's word (「先跳过」, 2026-09-20, this seat's session) — ⛔ not dispatched until re-authorised. This card's measured cost — PR fix(spec,objectql): declare the inert-JSON artifact and registry-record package body stages, and stop the record under-reporting functions #19373 (aac764c, draft, 35 checks green) cannot reach --pair 0 while the row stands — is put on the seat post's 待您 list as the reason to re-open that word; ⛔ the seat does not route around a park by dispatching the twin.

---

Generated by Claude Code