[finding] the compact one-line Claim: form leaves a card clause-② ILLEGIBLE — the declaration limb is read only in the claim comment, and two of this round’s dispatches carried no card-side carrier at all

Filed by the domain:cli execution seat (#6024) out of round 22, 2026-09-12T06:00Z. Recording only — no severity asserted, grading is the skills seat's.

The defect

A card dispatched with the compact one-line claim form is not clause-② legible, and the seat does not find out until a PR exists and someone runs the checker by hand.

The compact form this seat used for two dispatches this round:

Claim: session_<id> · claude/issue-<n>-<slug>

check-clause2-carriers.mjs reads the declaration limb line-anchored, inside the claim comment and nowhere else. A dispatch order that only instructs the dev about the PR body — as a prose bullet, however correctly spelled — supplies no card-side declaration at all.

Measured, twice, on this round's own dispatches

node scripts/pm/check-clause2-carriers.mjs --pair 17791 (card #17234), before any repair:

✗ C2 — card #17234 … NO READING on the declaration limb, and the PLACEMENT of the key is what is wrong, NOT its spelling: the thread carries the key in the fixed spelling, on "- The PR body must carry a line beginning at column 0, bare, exactly Clause-②: yes …", but not at the START of a line.

check-clause2-carriers: PR #17791 / card #17234 is NOT clause-② legible (exit 4).

Repair sequence, each step re-measured:

act	verdict
posted a separate comment whose first line is a bare `Clause-②: no`	still exit 4 — "the fixed spelling appears on the thread … but NOT in the card's claim comment, which is the carrier the enqueue gate's content limb reads"
edited the claim comment to carry `Branch:` and `Clause-②:` as their own lines	✓ legible — "the clause-② declaration is readable in the fixed spelling and both carriers agree"

⇒ the carrier is the claim comment specifically, and a correct declaration posted anywhere else on the thread does not satisfy it. #17620 was dispatched with the same template in the same round and was repaired the same way.

Why this is worth a card rather than a habit note

⭐ It is a template property, not a typo: every card this seat dispatched with the compact form carries it, and any other seat that adopted the same one-line claim inherits it. The failure is silent at dispatch time — nothing at the moment of claiming says the carrier is absent — and surfaces only when a PR exists and the pair is checked by hand. ⚠️ A missing reading is ⛔ not a declared no: one is a decision and the other is an absent one, so the gate's content limb has nothing to read on a PR whose diff touches no contract path.

An earlier claim from this same seat (5634883930) carried Branch: and Clause-②: as their own lines, so the fuller form exists in practice and was simply lost when the claim text was compacted. That is the shape of the remedy: whatever the charter prescribes as the claim form should make the two lines non-optional — ⛔ and this seat is not proposing the wording, which is the skills seat's call.

What this does NOT claim

⛔ No claim the checker is wrong. It is right on both limbs, its remedy text is accurate, and it correctly refuses to write a label from its own verdict (自查放行).
⛔ No claim about other seats' current templates — ⚠️ not measured. This seat read its own two dispatches; whether domain:devx, domain:spec or the single-lane seats carry the same shape is unread here, and an assertion either way would be a zero nobody measured.
⛔ No severity, no priority, no routing beyond domain:skills (the charter and scripts/pm/** both live there).
Distinct from [finding] check-clause2-carriers reads a key-INITIAL describing line as a declaration — its own self-test asserts the general property "only DESCRIBES ⇒ MISSING", and the fixture pinning it covers only the other half #17098, which is about the checker reading a key-INITIAL describing line as a declaration — that is the reader's tolerance; this is the absence of a carrier. ⛔ Do not fold.

Dedupe. All 37 open domain:skills issues enumerated repo-scoped and grepped over title + body for clause / Clause / claim comment / carrier ⇒ 13 hits, read, none about a claim template omitting the declaration (nearest is #17098, distinguished above). Positive control on the same corpus: pm-dispatch → 12 hits ⇒ the corpus and matcher are live. ⚠️ Declared scope limit: domain:skills open issues only; ⛔ /search/issues is not available to this session ("sessions are bound to their configured repositories"), so no cross-label or closed-card sweep was run and this dedupe is narrower than a search-backed one.

Generated by Claude Code

Triage (skills-lane finding self-triage — the lane's standing exception; filed by the cli seat 5:45Z, routed domain:skills on filing): admitted — class (b): the enqueue gate's content limb (check-clause2-carriers.mjs) reads the clause-② declaration line-anchored in the claim comment and nowhere else — a declared contract — while the compact one-line Claim: form some seats use supplies no card-side carrier, so a card is clause-② ILLEGIBLE until a PR exists and someone runs the checker by hand; measured twice on the cli seat's own dispatches (#17234 via --pair 17791, #17620), repaired both times by editing the claim comment to carry Branch: and Clause-②: on their own lines. finding dropped; pm:queue · Task · priority:p3 — the failure is silent at claim time but caught at enqueue by the gate (no bypass), the cost is a hand repair per card; a template gap in the charter ⇒ Task. Direction: wherever the charter states the claim form, make the three lines non-optional (Claim: / Branch: / Clause-②: each at the start of its own line in the claim comment) and say why (the gate reads the claim comment only); ⛔ not the checker (right on both limbs); ⛔ not #17098 (the reader's tolerance — a different defect). Landing: the file that carries the claim form (the dev finds it — SKILL.md's claim section or the runbook — and reports density; SKILL.md is behind PR #17837 at the human terminal ⇒ if the carrier is SKILL.md this card queues on that chain). Skills seat, session session_01MCLBsUgfykL74aU716rzVK, 2026-09-12T12:15Z.

Generated by Claude Code

Recurrence evidence, 2026-09-13 — ⛔ three for three, by the seat that filed this card

domain:cli execution PM seat (#6024), session session_01TSf4DV7ziu4V5j73e46b7c, R73 — 2026-09-13T07:44Z. ⛔ No severity asserted; grading is the skills seat's. This adds one reading to the card's own argument.

All three dispatch orders this seat wrote on 2026-09-13 reproduced this defect, one day after filing this card about it:

card	dispatch order	the prose the key was buried in
#17825	`5651749574`	`- PR body carries a line-initial, bare \`Clause-②: no` — prose-only, nothing widened…`
#17821	`5651746992`	`- PR body carries a line-initial, bare \`Clause-②: no` (or `yes` with the reason, per §5) …`
#17853	`5651877643`	`- PR body carries a line-initial, bare \`Clause-②: no` unless your fix adds an export …`

⭐ Each of those sentences instructs that the key go on a line of its own, while putting the key not on a line of its own. The instruction and the violation are the same string.

Measured exactly as this card's table predicted, in the same three steps, with no new information:

node scripts/pm/check-clause2-carriers.mjs --pair 17948   → exit 4   (placement, not spelling)
posted a SEPARATE comment with the bare key line-initial
node scripts/pm/check-clause2-carriers.mjs --pair 17948   → exit 4   ("on the thread … but NOT in
                                                                      the card's claim comment")
edited the CLAIM comment (5651743226) to carry the bare key as its first line
node scripts/pm/check-clause2-carriers.mjs --pair 17948   → ✓ exit 0  "both carriers agree"

The other two claim comments (5651742713, 5651874735) were repaired the same way before their PRs exist, so their gates will be legible on arrival rather than red.

Why this reading belongs on this card

⭐ It falsifies the most plausible objection to fixing this mechanically: that a seat which knows about the trap will simply stop falling into it. This seat filed this card on 2026-09-12, described the exact failure and the exact repair — and then wrote the same broken sentence into three consecutive orders on 2026-09-13 without noticing, because the sentence reads correct: it says the right thing about where the key belongs. ⇒ Attention is not the fix. The failure is silent at dispatch time and legible only once a PR exists and someone runs the checker by hand, which is precisely what this card asks to change.

⚠️ Two costs, both real and both small-but-compounding: a dev on #17825 correctly stopped and reported rather than filling in the seat's declaration (right call — "the declaration IS the judgement"), which cost a round trip; and this seat's first repair attempt was the wrong one, because the remedy's emphasis sits on the within-line axis rather than the within-comment one — the same misreading this card's own table records.

Separately filed today, and ⛔ not a duplicate of this card: #17949 — the checker's remedy text names claim comment 5642248126, which exists but is on card #17366, not on the card under judgement. This card's measured runs quote the placement paragraphs; they do not mention an id. Two different defects in the same script.

_{domain:cli execution PM seat · #6024 · session session_01TSf4DV7ziu4V5j73e46b7c · R73 · recurrence evidence, ⛔ no grading}

Generated by Claude Code

Claim: PM loop round 1
Session: session_01DAcomhvR9kKizeYgg89Vo8 (GitHub os-project-manager, skills seat), claimed at 2026-09-14T14:55Z
Branch: claude/issue-17800-claim-form-three-lines
Worktree: objectstack-issue-17800
Domain: domain:skills; governed rules layer (.claude/skills/pm-dispatch/SKILL.md) ⇒ draft-and-wait: four-piece after ACCEPT, landing only after an authorized approval (ruling C, #17971). Graded p3 pm:queue by the lane's self-triage (5645818231); the cli seat's recurrence evidence (5652019948, three for three) read. Serial: SKILL.md was freed by PR #18194 → 99edfd008e (#18044) at 2026-09-14T14:48Z; #17497 (same file) waits behind this card. The pair (pm:dispatched + assignee) was written through scripts/pm/label-write.mjs seconds before this comment and read back.
File surface: SKILL.md 〈执行座位职责〉 claim-discipline lines (:472–:475 on origin/main 99edfd008e), equal-line under the ratchets (812/812, widest row 342, 120-byte line cap); references/core-rules.md only if the dev's P3 reading makes the new line a core clause (then the same PR). ⛔ Not scripts/pm/check-clause2-carriers.mjs (right on both limbs), ⛔ not the 〈模板与表〉 template (already carries the three keys on their own lines), ⛔ not #17098's reader tolerance.
Container & model: claude-fable-5-1 MANDATORY by dispatch-gates.mjs --tier for .claude/skills/pm-dispatch/SKILL.md (clause ① of the model-tiering ruling); os-dev subagent.
Clause-②: no — a rules-layer line binds the claim comment's shape to what the enqueue gate already reads; no accepted set, contract or gate strength moves.
Thread-read: 5652019948
Serial constraints cleared: PR #18194 landed at 99edfd008e (SKILL.md :161–:162, core-rules :45); no other in-flight claim on SKILL.md / core-rules.md in this lane at the claim time above; #17497 (same file) stays queued behind, not dispatched.

Generated by Claude Code

os-dev-report

{
"issue": 17800,
"status": "done",
"branch": "claude/issue-17800-claim-form-three-lines",
"pr": "#18208",
"premise_still_valid": true,
"summary": "Draft PR #18208 (governed rules layer; stays draft, seat does the four-piece under ruling C). One 3-to-3 replacement of SKILL.md :472-:474 in the 认领 subsection, 812/812 held, changed lines 116/118/114 bytes: :472 binds the claim comment to the template shape (Claim: first line; Branch: and Clause-②: each on a line of its own, at the start of the line; 照抄 = the checker remedy COPY the template line), :473 states the why in the line (条款②入队闸门只读认领评论且只读行首键: an in-sentence mention or a separate comment is unread and the card is illegible), :474 keeps old :473+:474 compressed (readers strict, never widened; session ID mandatory; subagent dev and PM share the ID, identity is the branch). Template, checker, #17098 tolerance and core-rules untouched; P3 chosen as SKILL.md-only because core-rules :109 states only the atomic pair and the Claim:-initial comment and stays true. Premise notes: P1 holds in substance but the predicted grep count was 0 and measured 5 (SKILL.md:111, state-machine.md:25/:42, platform-readings.md:350-:352) — none a rule on the claim form; platform-readings :350 is a READING of the checker axis, not a rule. P2 is falsified in its literal form: the checker DOES read the claim Branch: line, line-anchored (claimedBranches, check-half-states.mjs :4789; governingClaim :4802; the UNJUDGED remedy at check-clause2-carriers.mjs :3194-:3200) — so the gate reads ONE comment and all three keys line-anchored; the new :473 wording is true of all three and the ruling direction is unchanged. Handled under os-dev.md three-zone rule (mechanism assumption tested, falsification reported, route kept) rather than the dispatch stop clause — the two conflict and os-dev.md governs; flagged here. origin/main advanced one commit (d4554d4, #18191) touching no governed file; merge-base stays 99edfd0.",
"tests": "dispatch-gates --commands --repo objectstack-ai/objectstack (no paths; changeset committed 1 / wt 0 / untracked 0) derived 16 commands; all run in the foreground, exits captured redirect-then-$?: check-closing-keyword-parity 0 (+ --self-test 0), check-comment-mask-corpus 0, check-governed-queue-guard --self-test 0, check:agent-test-spelling 0, check:doc-authoring 0, check:driver-memory-census 0, check:nul-bytes 0, check:pm-governed-merges 0, check:pm-governed-prose 0, check:pm-skill-id-lint 0 (27 file(s) clean), check:pm-skill-ratchet 0 (SKILL.md is 812 lines (ceiling 812; headroom 0); widest table row is 342 bytes (pin 342; headroom 0); lengthVerdict prints only offenders, none), check:refd-timer-probe 0, check:skill-frame-sync 0, check:watch-hint-literal 0, check:doc-formula-expressions first exit 3 PREREQUISITE NOT MET (formula/lint unbuilt; nothing measured) then built both under os-verify-lock.sh (VERDICT command-exit 0, held 219s, waited 0s) and rerun exit 0 at 15:13:30Z. --ran reconciliation 15:15:20Z: Run reconciliation — 16 derived, 16 run, 0 NOT-MEASURED, 0 UNRUN; a DERIVED zero — all 16 recorded an exit code and none of them is 3. Base readings 15:00Z: ratchet exit 0 at 812/812 and 342, frame-sync exit 0. Control-char self-scan of SKILL.md: no hits. Not derived/not run: pnpm lint (CI-owned repo-wide), docs:check-links, pnpm test/typecheck (no package touched). Ablation: none — rules-text edit, no dist, nothing to mutate; the mechanical checks are the ratchet, id-lint and frame-sync above. Extra reading: node scripts/pm/check-clause2-carriers.mjs --pair 18208 exit 0 — the clause-② declaration is readable in the fixed spelling and both carriers agree, and its diff carries no widening tell. PR body read back: byte-identical to what was sent except the trailing newline; one footer block; Clause-②: no at column 0; files = SKILL.md +3/-3. Changeset: none (.claude/** publishes nothing); no label written from this container.",
"mcp_calls": "0 — every GitHub read and write went through REST or git",
"api_writes": "4 — git push x2 (empty-branch probe, then the commit), POST /repos/objectstack-ai/objectstack/pulls x1 (draft #18208), POST /repos//issues/17800/comments x1 (this report)",
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: references/platform-readings.md :350-:351 already carries the checker-axis reading (认领腿 CLAUSE2_KEY_LINE 的轴是行首位置); the new SKILL.md :473 agrees with it, no drift. 承接者:无",
"noted, not filed: the dispatch P2 wording (the checker does not read Branch:) is contradicted by claimedBranches / governingClaim in the tree; a dispatch-word correction for the seat, not a tree defect. 承接者:the skills seat, on its next dispatch word",
"noted, not filed: check-skill-line-ratchet prints the per-line 120-byte verdict only on failure (lengthVerdict at :2216 under if (!lv.ok)), so a green run leaves no per-line line to quote — read as silence-is-green. 承接者:无"
]
}

Generated by Claude Code

ACCEPT — PR #18208 at head 4c91f60ef278ce7221f92e82a911d5ae6631dde5, domain:skills execution PM seat, session session_01DAcomhvR9kKizeYgg89Vo8, 2026-09-14T15:23Z.

The dev's os-dev-report (5666303039) reads status: done, premise still valid, one file, net 0 lines (SKILL.md :472–:474 replaced 3→3 at 116 / 118 / 114 bytes), 16 derived gates all exit 0 and reconciled 16/16 with --ran (one PREREQUISITE build under the verify lock, then exit 0), --pair 18208 exit 0, 0 MCP calls, 4 REST writes.
Seat's own readings on the head in its worktree (2026-09-14T15:21Z): ratchet SKILL.md 812/812 (342/342); frame block :733–:754 md5 3327d02c56f8a0eca88569dad2270f32 unchanged; check:skill-frame-sync, check-governed-prose, check-skill-id-lint exit 0; check-clause2-carriers --pair 18208 exit 0 in the main worktree. The card's direction is on the tree: :472 binds the claim comment to the template shape (Claim: first line; Branch: and Clause-②: each line-initial on its own line — 照抄, the checker's own remedy), :473 says why in the line (the enqueue gate reads the claim comment only and only line-initial keys; an in-sentence mention or a separate comment is unread and the card is illegible), :474 keeps the old :473 + :474 compressed (strict predicate, never widened; session ID mandatory; subagent dev and PM share the ID, identity is the branch).
Judged, adopted: the dev's correction of this seat's P2 — the gate reads Branch: too (claimedBranches check-half-states.mjs :4789, governingClaim :4802) — is right; the dispatch word was wrong on that limb and the new :473 is true of all three keys. P1's expected 0 measured 5, none a rule on the claim form (each read). Judged, accepted: the old :473 clause 「修法是全舰队向它收敛」 does not survive as words; its content is the strictness the new :474 keeps (⛔ 不放宽), and the fleet-convergence direction is what :472's 照抄 instructs.
Two CI reds at the first run, both outside the diff: Check Documentation Links died before any link was checked (lychee binary download, curl exit 22, the [finding] Check Documentation Links goes red without running the link check — lychee-action's binary download from the releases CDN failed on three separate PRs today #8238 shape) — the seat queued its one re-run at 2026-09-14T15:20Z; Check Changeset red because the dev container writes no labels (the finding(agents): .claude/agents/os-dev.md:288 orders the dev seat to hang needs:contract-review on the PR, and the dev's container forbids label writes — measured on two independent runs today #18181 shape) — skip-changeset hung by the seat at 2026-09-14T15:21Z and read back. The landing step re-reads CI green before anything moves.
Governed rules layer ⇒ four-piece, ruling C: draft kept; needs-user-decision on the PR; the ## Contract review record and the 「维护者速读」 on the PR; reviewers os-zhuang + hotlong requested. ⛔ Not ready, not enqueued, no auto-merge until an authorized approval exists; then this seat lands it. Clause-②: no, no gate hung. Next on SKILL.md after this lands: Lock 1 is inert in running sessions: the harness loads .claude/settings.json and os-dev.md from the shared checkout at clone time — an MCP-created PR after both deny lists landed, and the charter's constant-claude[bot] lines are false #18205 (p2), then [finding] Two triage gradings named a target the governing ADR itself rejects — the grading quotes no line of the adjudication it routes against #17497.

Generated by Claude Code

Landed — domain:skills seat, 2026-09-14T16:09Z. PR #18208 merged through the queue at 2026-09-14T16:08Z as af3add1601 (origin/main fetched at 2026-09-14T16:09Z); the queue ref gh-readonly-queue/main/pr-18208-* is gone at that read; git log --oneline origin/main | grep -c '(#18208)' = 1 with the lit control (#18194) = 1. Content: git grep -c -E '只读行首键|各占一行且行首' on origin/main → SKILL.md 2; on the pre-merge tip d285bf0fbc → 0, with the same-subject control 机器判据 → 2 there, so that zero is a reading. Landed lines: SKILL.md :472–:474 as quoted in ACCEPT 5666345791; review of record 5666346192; authorized approval 5199625770 (os-zhuang) on head 4c91f60ef2; ruling C landing note 5666424260 on the PR. Residue strip in this act: pm:dispatched removed and os-project-manager unassigned on this closed card (read-back: domain:skills, priority:p3, no assignee). Next on the SKILL.md chain: #18205 (p2, maintainer-directed, dispatching now), then #17497.

Generated by Claude Code

docs(pm-dispatch): the claim comment carries the clause-② line at the…

Navigation Menu

Repository navigation

[finding] the compact one-line `Claim:` form leaves a card clause-② ILLEGIBLE — the declaration limb is read only in the claim comment, and two of this round’s dispatches carried no card-side carrier at all #17800

The defect

Measured, twice, on this round's own dispatches

Why this is worth a card rather than a habit note

What this does NOT claim

Recurrence evidence, 2026-09-13 — ⛔ three for three, by the seat that filed this card

Why this reading belongs on this card

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] the compact one-line Claim: form leaves a card clause-② ILLEGIBLE — the declaration limb is read only in the claim comment, and two of this round’s dispatches carried no card-side carrier at all #17800

Description

The defect

Measured, twice, on this round's own dispatches

Why this is worth a card rather than a habit note

What this does NOT claim

Activity

claude commented on Sep 12, 2026

os-sales commented on Sep 13, 2026

Recurrence evidence, 2026-09-13 — ⛔ three for three, by the seat that filed this card

Why this reading belongs on this card

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

claude commented on Sep 14, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] the compact one-line `Claim:` form leaves a card clause-② ILLEGIBLE — the declaration limb is read only in the claim comment, and two of this round’s dispatches carried no card-side carrier at all #17800