feat(pm): a class-(b) finding names its seam — the Seam: line on os-dev.md rule 3, and the H68 patrol row for a (b) card without one

[os-steve]

Fixes #18901

Clause-②: no
Ruling-ref: 5727134819 — director seat batch #157 item 3, letter A′ on #18900 (item ①, this card, as filed); triage handover 5725479658 binds (every governed text at its ceiling; the retired line is named below).
Seat domain:skills#2 · claim 5749402494 · branch claude/issue-18901-seam-line-filing-rule · head 5033c4d (on origin/main 13d5294).

What lands

1. .claude/agents/os-dev.md 六条基本规则 3 — net 0 (403 / 403), every touched line under the 120 B line ratchet

• :42 before (97 B): 「 - (a) 可复现缺陷(复现或失败探针具名);(b) 违背已声明契约(引契约原文);」
• :42 after (113 B): 「 - (a) 可复现缺陷(复现或失败探针具名);(b) 违背已声明契约(引契约原文,带 Seam: 行);」
• :43 NEW (112 B), the grammar in ONE line directly under (b): Seam: spec:KEY → runtime:CALL_SITE | renderer:COMPONENT, then 「,无消费者写 consumer: none。」 — in the file the three placeholders are written in Chinese inside ASCII angle brackets (键或符号 / 调用点 / 组件), spelled out here as words because this body is a GitHub body.
• RETIRED to pay for it — :53 at 13d5294 (115 B): 「 - ⛔ 不用 MCP GitHub 写工具;令牌按会话定:installation ⇒ claude[bot],user-to-server ⇒ 用户。」 Surviving homes, clause by clause: the ban lives on :51 「GitHub 写一律走 REST 代理(curl 带 GITHUB_TOKEN)」 (all writes REST ⇒ no MCP write), :307 「不走 MCP」, :308 「⛔ 不换路重发(curl/MCP/手工)」 and the report template :371 mcp_calls 「a write tool in the list = this report is refused」; the token → login fact lives at .claude/skills/pm-dispatch/references/platform-readings.md:130 「容器 curl 的 REST 通道令牌两类:installation(claude[bot])与 user-to-server(用户)」, and the rule that fact served — 「归属 = 文本里的 session ID,非 user.login」 — stays on :51. Line numbers are pre-edit; after the edit the REST line is :52 and :307 / :308 / :371 are unchanged (one insert and one delete above them cancel). No gate pins the retired line's phrases (grepped scripts/, .claude/, .github/).
• Spelling agreed with the routing half (pm-dispatch: a Seam: card routes by the seam — vertical dispatch by default inside objectstack (file surface names both ends; acceptance = the consumer reads it or the key retires with its ledger row), and rule 2's parent + per-repo sub-issues trigger on a spec↔objectui seam, not only on a feature (item ② of #18900) #18902, in flight on this seat): the key is backticked Seam: in both texts.

2. scripts/pm/check-half-states.mjs — H68, ONE report-only row, net +120 (124 insertions / 4 deletions)

• Predicate h68ClassBWithoutSeamLine(issue) with its docblock, placed after H63's; readers classBDeclaration(body) — a markdown heading naming the class, or a line opening with it (list / quote / bold decoration and a leading Finding tolerated), the shapes measured below — and hasSeamLine(text) = directiveValues(text, 'Seam'), the reader Blocked-by: / Restart-when: share (decoration tolerated, a same-line value required, a bare key with nothing after it is no line).
• Band state, registered beside H67 with the three refusals argued (familyRegistryCoverage: 0 missing / 0 extra; registry 62 / cap 64).
• Sweep call beside H63's in the UNSCOPED listing — the class is declared in the body before any grade, so an ungraded finding carrier is first visible there; two body reads on a row already held, no request, no thread.
• Battery H68 class-(b) seam line beside H63's cases (26 registered, pin 24; ⛔ not at selfTest()'s tail); SELF_TEST_BATTERY_FLOOR 6 → 7 with the three existing exact pins moved.
• ⛔ Untouched: the ownership markers and markerMatches, the H26 region ([finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255), every workflow file, and directiveValues's JSDoc key union (that line sits in [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255's hunk; the reader is untyped at runtime).
• ⛔ Never a half-state verdict by itself: the file's contract is report-only (exit 0 on a completed sweep, no label written); the row lists a card by number and prescribes the line. The VALUE after Seam: is deliberately not judged here — presence at line start is the filed contract; the value is the routing half's reading.

The class-(b) signal, measured (board read 2026-09-20T11:12Z, 537 open non-PR cards, saved listing replayed against the finished row)

• Bodies saying class (b) / 判据 (b) anywhere: 20. DECLARING it — a heading naming the class (## Class (b) — …, ## Why it is class (b), ## The defect — class (b), …, ## 判据 (b):…) or a line opening with it (Class (b): …, **Class (b)** — …, ⇒ class (b): …, Finding class (b), …): 16, one of them the pm:seat post [PM seat] triage (objectstack-wide) — 🟢 session_01KYzNJPin9Ar4oMMkSh4S9h · 就座 2026-09-20 · 自绑 Routine trig_016NBra1kZGHk7UteTziFbrK · ⛔ 接手先读正文「大扫除试运行班收班交接台账」再定级 #6015 (never judged) ⇒ 15 listable. Mid-sentence only: 4 — finding(spec): the react-tier ListView catalogue declares onNavigate's action as 'view' | 'edit' — the implementation sends view and new_window, and edit zero times #19030, [finding] @objectstack/spec has four grammars for a package version, and ManifestSchema refuses the 2.0.0-beta.1 that PackageVersionSchema's own docstring advertises #18697, [finding] the tracked root sdui.manifest.json and the browser dump differ on the fields the parity gate consumes (18 of 57 components; dataSource missing on 15 in the root artefact) — the gate's verdict set is identical under both today, which is the configuration in which the drift is least visible #17735 and os-dev's three-class filing rule: a class-(b) finding names its seam — a Seam: line (producer → consumer, consumer: none allowed) on the card, and the half-state patrol reports a class-(b) card that lacks one (item ① of #18900) #18901, this card, which is ABOUT the rule and is the false positive a mention-reader would list.
• Comment channel: finding cards 20 (5 with domain:*, 6 with any comment) — a comment naming the class on 0 of 20; on the 20 body hits, 8 also carry such a comment; comment-only cards: 0. The body declaration strictly dominates and costs no request ⇒ the row reads the body only; the comment channel is not read.
• Seam: at line start on the whole open board: 0 (6 cards mention Seam:, all about the rule) — every (b) card listed today predates the rule, which is the population the routing half needs the line on.
• Would list today (15): finding(spec,objectql): ruling B's percent storage derivation is NOT live — max_scale has no percent arm and the percentScaleOf docblock is silent about scale #19320 [finding] the duplicate-name refusal on sys_permission_set carries NO machine-readable code — a bare 409, and it fires BEFORE the packaged-set guard, so the Clone dialog's most likely error never reaches NOT_OVERRIDABLE #19307 [finding] 退役的 GIT_* allowlist 还有第三个载体 —— create-objectstack/src/template-consistency.test.ts,分诊 2026-09-08 的普查只点了两个 #19171 finding(scripts/pm): H52 has no clearing act on a pm:on-hold card — the only three null-exits are close, needs-user-decision (which the row itself forbids here) and a newer dev report a held card can never get #19160 skills(pm-dispatch): grading reads the checklist ITEM, not the area — Path: carries the item id, the card inherits the item's P, and 「有人在等」 is not a test #19155 skills: re-key the three 事实层 = references/ spellings the Tier S ruling leaves false (os-dev.md :286–:287, landing-operations.md :27–:28, H48) #19146 [finding] check-doc-authoring 的两条腿都够不到 packages/spec/scripts/** —— 一条印给作者的拒绝消息里带着 404 的编号,而门禁是绿的 #19124 [finding] POST /api/v1/packages installs a manifest with NO version and answers 201, while its published declaration requires one — the door parses nothing #19120 [finding] a key retired through the .strict() guidance-map route leaves no [REMOVED] tombstone — the whole route is invisible to the ledger and to every walk-rooted census, and nothing says so #19118 [finding] the ownership near-miss census still lists Release-landed: … report-line labels as refused release records — four separator lines on two cards, the largest remaining class of the kind #18926 narrowed #19043 finding(skills): the seat's claim-time serial check reads open PRs' file lists only, so a sibling card already claimed and pushed but not yet opened as a PR is invisible to it — one git ls-remote --heads origin | grep <file stem> would have shown #18314's branch before #18844 was claimed #18964 finding(skills): dispatch-runbook.md :61 paraphrases reading three NARROWER than SKILL.md :80 declares — it drops ∪ pm:queue and 同席号, so a seat following the runbook can miss a Claim: on a pm:queue card and fail to self-retire (the two-writer seat, by a second route) #18957 [finding] the landing flow has no legal state for the undraft route's 429 — a seat that hits it can only escalate or invent a retry cadence, while the 429 body names the exact instant to wait for #18743 [finding] driver-memory keeps a THIRD refusal wording for the same dateRange condition — the tree is at two wordings, not one, after #18232 #18277 [finding] check (c) has no proof shape for a strict-schema guidance retirement — on a reachable def such a key can never prove itself, and it passed only because proof 2 was broken #18132.
• Specimen LISTED: finding(scripts/pm): H52 has no clearing act on a pm:on-hold card — the only three null-exits are close, needs-user-decision (which the row itself forbids here) and a newer dev report a held card can never get #19160 (pm:queue · domain:skills · priority:p2) — 「## Class (b) — it violates a contract this repository publishes」, no Seam: line; the row quotes that heading back.
• Specimen NOT listed: the same finding(scripts/pm): H52 has no clearing act on a pm:on-hold card — the only three null-exits are close, needs-user-decision (which the row itself forbids here) and a newer dev report a held card can never get #19160 body with Seam: spec:… → runtime:… appended (a fixture — no live (b) card carries the line; finding(spec): field.relatedListFilter and object.validations are DECLARED by the served schema and omitted by METADATA_FORM_REGISTRY's forms — the generic metadata form never renders them, so an author's only door is the Source tab #19085 is closed and carries neither signal), and os-dev's three-class filing rule: a class-(b) finding names its seam — a Seam: line (producer → consumer, consumer: none allowed) on the card, and the half-state patrol reports a class-(b) card that lacks one (item ① of #18900) #18901 itself (mid-sentence mention).

Reader test

「I found a declared key the runtime never reads — what do I write?」 ⇒ the (b) line — 违背已声明契约(引契约原文,带 Seam: 行): cite the contract text and add Seam: spec:THE_KEY → consumer: none (you measured no consumer). Where a consumer exists but does not honour the key: Seam: spec:THE_KEY → runtime:THE_CALL_SITE, or Seam: spec:THE_KEY → renderer:THE_COMPONENT.

Verification — all on head 5033c4d, every exit code captured before any pipe

• node scripts/pm/check-half-states.mjs --self-test → exit 0 — 「✓ check-half-states self-test: 5107 cases pass. Batteries: H66 released queue card 183/172, H19 judged-set founding 37/34, H65 tier declaration spelling 46/42, H67 queued merged-delivery reading 152/142, H2/H47/H66 decorated ownership marker 213/184, ISSUE_BODY_LIMIT measured cap 55/52, H68 class-(b) seam line 26/24」.
• pnpm check:pm-skill-ratchet → exit 0 — 「✓ check-skill-line-ratchet: .claude/agents/os-dev.md is 403 lines (ceiling 403; headroom 0)」.
• Gate union from node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 5033c4d (47 commands, no paths passed): 47 / 47 ran on 5033c4d, every one exit 0 — --ran reconciliation: 「47 derived famil(ies) accounted for — 47 run, 0 NOT-MEASURED (a DERIVED zero — all 47 recorded an exit code and none of them is 3)」. The list: node scripts/check-changeset-no-major.mjs --base origin/main · node scripts/check-changeset-no-major.mjs --self-test · node scripts/check-ci-filter-parity.mjs · node scripts/check-closing-keyword-parity.mjs · node scripts/check-closing-keyword-parity.mjs --self-test · node scripts/check-comment-mask-corpus.mjs · node scripts/check-declaration-mirrors.mjs · node scripts/check-declaration-mirrors.mjs --self-test · node scripts/check-scripts-symbol-anchors.mjs · node scripts/check-scripts-symbol-anchors.mjs --self-test · node scripts/check-self-test-wired.mjs · node scripts/check-self-test-wired.mjs --self-test · node scripts/check-self-test-workflow-commands.mjs · node scripts/check-self-test-workflow-commands.mjs --self-test · node scripts/check-whole-set-label-write.mjs · node scripts/check-whole-set-label-write.mjs --self-test · node scripts/pm/bare-root-worklist.mjs --self-test · node scripts/pm/board-snapshot.mjs --self-test · node scripts/pm/check-governed-queue-guard.mjs --self-test · node scripts/pm/check-harness-current.mjs --self-test · node scripts/pm/sweep-closed-cards.mjs --self-test · pnpm --filter @objectstack/lint run check:doc-formula-expressions · pnpm check:agent-model-declared · pnpm check:agent-test-spelling · pnpm check:bash32-floor · pnpm check:changeset-gate-self-tests · pnpm check:cli-command-ids · pnpm check:closing-target-claim · pnpm check:commit-card-trailers · pnpm check:cross-package-test-inputs · pnpm check:doc-authoring · pnpm check:driver-memory-census · pnpm check:entry-guard · pnpm check:nul-bytes · pnpm check:parse-guard · pnpm check:partof-closing-keyword · pnpm check:pm-dispatch-gates · pnpm check:pm-governed-merges · pnpm check:pm-half-states · pnpm check:pm-skill-id-lint · pnpm check:pm-skill-ratchet · pnpm check:pnpm-filter-targets · pnpm check:ratchet-remedy-authority · pnpm check:refd-timer-probe · pnpm check:single-claim-paths · pnpm check:skill-frame-sync · pnpm check:watch-hint-literal.
• node scripts/pm/check-governed-merges.mjs --test .claude/agents/os-dev.md scripts/pm/check-half-states.mjs → exit 3 = GOVERNED, Tier S (.claude/** ×1; scripts/pm/check-half-states.mjs not on the register) — the expected reading: this PR stays DRAFT until the seat's ## Contract review record for the current head.
• Build for check:doc-formula-expressions: pnpm --workspace-concurrency=2 --filter '@objectstack/lint...' build under scripts/pm/os-verify-lock.sh — 「VERDICT command-exit 0 · held the lock 220s」.

Acceptance notes (noted, not filed)

• Three live (b) cards declare the class only mid-sentence (finding(spec): the react-tier ListView catalogue declares onNavigate's action as 'view' | 'edit' — the implementation sends view and new_window, and edit zero times #19030, [finding] @objectstack/spec has four grammars for a package version, and ManifestSchema refuses the 2.0.0-beta.1 that PackageVersionSchema's own docstring advertises #18697, [finding] the tracked root sdui.manifest.json and the browser dump differ on the fields the parity gate consumes (18 of 57 components; dataSource missing on 15 in the root artefact) — the gate's verdict set is identical under both today, which is the configuration in which the drift is least visible #17735) and are deliberately not listed; 承接者: the seat that next grades or writes on each — a heading or line-start restatement puts them on the row.
• directiveValues's JSDoc key union does not name 'Seam'; 承接者: [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255's lander (one word on a line already in that hunk).
• SELF_TEST_BATTERY_FLOOR: [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 (in flight) bumps the same 6 → 7 and adds its own roster row; the later lander resolves the roster merge and sets 8.
• PM mechanism assumption 3 refined: the graded-(b) population does not live under finding + domain:* (5 cards, none with a class comment) — finding is transient (「定级即离标」, H63), so the measurement ran over the whole open board.

维护者速读(草稿)

改了什么:给开发 agent 的立卡规则加了一句:凡是「协议里声明了、运行时或渲染器没兑现」这一类(b 类)发现,卡面上必须带一行机器可读的 Seam:,写明这条缝的两端(哪个协议键 → 哪个运行时调用点或哪个渲染组件;实测没有消费者就写 consumer: none)。同时巡检脚本新增一行只报不判的 H68:把已经声明为 b 类却没写这一行的开放卡按编号列出来。文件行数不变(403 / 403),为此退休了一行重复条款(MCP 写工具禁令 + 令牌说明),它的内容在同规则的 REST-only 行、标签写入条款、报告模板和 platform-readings 里都还在。

为什么改:你 9-18 同意的五项攒批第 ① 项。今天缝卡按「修复落地的包」路由,缝的两端各落一层、各修一次;有了这一行,路由那一半(#18902)才能按缝派发,一个键在一次落地里要么被兑现要么退役。板面实测:15 张开放 b 类卡今天都没有这一行,巡检会把它们列出来。

风险与代价(含回滚):只加一行规则、一行巡检;不改标签、不写板面、不判红。误报边界已量:句中顺带提到「class (b)」的卡不列(3 张),席位帖子不列。回滚 = revert 本 PR。巡检脚本的行数配额已用满(+120);自测下限 6 → 7,与在飞的 #19255 撞同一行,后落地者改为 8。

席位意见:

你要做的:受管面(.claude/**,Tier S)⇒ 席内达档复核后由席位落地,不需要你点击。若你希望巡检也校验 Seam: 行的取值格式(今天只查有没有这一行),回一个字,另立卡。

---

Generated by Claude Code

[os-steve]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 5033c4d9a1e3560911bed9922c1954a8ed698257

① Derived judgments

• Published accept set / public surface: none moves — the dev agent's instruction prose (.claude/agents/os-dev.md, net 0 at 403 / 403) and a report-only row in the PM patrol reader (scripts/pm/check-half-states.mjs, +124 / −4, net +120 against the ≤ +120 budget); no published package, no schema, no error code, no label written, no request added. check-widening-tells --declaration no on the diff: 2 files NOT MEASURED (no declared surface covers either) — the no stands on the seat's reading of the diff, ⛔ not on that exit. check-governed-merges --pr 19356: GOVERNED, Tier S (.claude/** ×1; the script is not on the register). Judged correct.
• The rule half (六条基本规则 3): (b) now reads 「违背已声明契约(引契约原文,带 Seam: 行)」 (:42, 113 B) and ONE adjacent line carries the grammar Seam: spec:<键或符号> → runtime:<调用点> | renderer:<组件>,无消费者写 consumer: none (:43, 112 B) — both under the 120 B line ratchet; the spelling (backticked Seam:) agrees with the routing half already on origin/main (SKILL.md :241 「Seam: 卡归 spec 席,默认纵向派发」, PR skills(pm-dispatch): a Seam: card routes by the seam — spec seat, vertical dispatch by default; rule 2 splits a spec↔objectui seam into parent + per-repo sub-issues #19321). Paid by retiring :53 at 13d5294 (the MCP-write ban + token-kind fact); the seat verified every surviving home on THIS head: the ban at :52 「GitHub 写一律走 REST 代理」, :307 「不走 MCP」, :308 「⛔ 不换路重发(curl/MCP/手工)」 and the report template :371 (mcp_calls — a write tool in the list refuses the report); the token → login fact at references/platform-readings.md :130; no gate pins the retired phrases. Judged correct.
• The reader half (H68): one state-band row on the UNSCOPED listing — an OPEN card that DECLARES class (b) in its body (a heading naming the class, or a line opening with it; decoration and a leading Finding tolerated) and carries no Seam: line at line start, read with the Blocked-by: directive reader (directiveValues(text, 'Seam'), same-line value required). The chosen reader is the MEASURED one: 537 open cards at 2026-09-20T11:12Z — 16 declare the class (one a pm:seat post, excluded by NEVER_SWEPT_LABELS), 4 mention it mid-sentence (this card among them — the false positive a mention-reader lists), comments name the class on 0 of 20 finding cards, so the body declaration dominates and costs no request. Closed cards and PR rows null; the VALUE after Seam: is deliberately not judged (the routing half's reading). Registered state beside H63 / H65 with the three refusals argued; familyRegistryCoverage 0 missing / 0 extra; battery 「H68 class-(b) seam line」 26 / pin 24 placed beside H63's cases (⛔ not at selfTest()'s tail); SELF_TEST_BATTERY_FLOOR 6 → 7 with the three existing exact pins moved. The remedy text names the routing rule the line feeds (vertical dispatch for a packages/spec producer; parent + sub-issues for an objectui consumer) as SKILL.md :241–:243 state it. Judged correct — a leg on an existing listing, ⛔ not an invented sweep.
• Evidence, seat-run on this head in a detached worktree: check-half-states.mjs --self-test exit 0 (5107 cases; the new battery 26 / 24); check-skill-line-ratchet exit 0 (「os-dev.md is 403 lines (ceiling 403; headroom 0)」); check-skill-id-lint / check-governed-prose exit 0; git merge-tree against origin/main c334ba0: 0 conflicts, and neither file has moved on main since merge-base 13d5294; CI on 5033c4d: 37 check runs, all success / skipped. The dev's 47-family gate union reconciles 47 / 47 with --ran (a DERIVED zero).

② Semver level

None — nothing published; skip-changeset on the PR is the correct declaration. Judged correct.

③ Boundary flags

open_questions empty. Deviations read and accepted: the run was cut by the account's usage wall and resumed on its own worktree (base merged as a fast-forward); a first H68 draft measured +215 and was restored and re-applied compact to +120 exactly (no over-budget commit exists); one gate batch was moved to the background by the harness and its remaining commands re-run in the foreground, both readings recorded; the --concurrency flag forwarded into a package script on the first build attempt, corrected on the second; the platform dropped one trailing newline of the PR body. Shared line: SELF_TEST_BATTERY_FLOOR 6 → 7 is also made by PR #19351 (#19255) — that PR was converted back to DRAFT at 2026-09-20T14:16Z by an act that is not this seat's, so if this PR lands first the floor is 7 and #19351's lander resolves the roster to 8 (the claim's rule, unchanged). Three noted, not filed items with named carriers: three live (b) cards declare the class only mid-sentence (#19030 · #18697 · #17735 — the seat that next writes on each); directiveValues's JSDoc key union lacks 'Seam' (the line sits in #19255's hunk — its lander, one word); pnpm check:pm-dispatch-gates runs ~25 min (an observation). Mechanism assumption 3 was refined by measurement (finding is transient, so the (b) population is the whole open board, not finding + domain:*) — accepted, the correct consequence of measuring first.

Implemented-by: claude/issue-18901-seam-line-filing-rule
Reviewed-by: session_017ETYWqMQD4qMtZzAGovWNi

VERDICT: PASS

---

Generated by Claude Code

[os-steve]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: ef8358d84b139c8fbc6e1cf2e007484b73947c17

① Derived judgments

• This head = the reviewed head 5033c4d (PASS 5750369816, every judgment there unchanged: the rule half at 403 / 403 with :53 retired and its homes verified; the H68 report-only row; no public surface moved) + ONE merge commit of origin/main 231283a made by the seat, because PR feat(pm): second Unlock-action: value — a label-transition exit on a named card (#19255) #19351 ([finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255) landed first and both PRs raised SELF_TEST_BATTERY_FLOOR 6 → 7 and added a roster row at the same place — the claim's rule that the later lander resolves it, done by the seat since the dev's run ended on the tier's usage limit after its hand-back. Resolution, read line by line: both roster rows kept (H19/H26 label-transition unlock exit: 10, then H68 class-(b) seam line: 24); the floor set to 8; every floor pin moved to EIGHT — the [finding] ISSUE_BODY_LIMIT = 65536 is falsified: a 150,507-char issue body is stored, and the real refusal boundary measured into (257945, 263533] bytes #18664, H67 (now naming both batteries), BATTERY68, [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 and H68 pins; nothing else in the two conflict hunks or outside them touched by hand (the rest of the merge is main's own content). Judged correct.
• Evidence, seat-run on this head in the merge worktree: check-half-states.mjs --self-test exit 0 — 5118 cases, 8 batteries (the two new ones 11 / 10 and 26 / 24 both present); check-skill-line-ratchet exit 0 (「os-dev.md is 403 lines (ceiling 403; headroom 0)」); check-skill-id-lint / check-governed-prose exit 0; check-governed-merges --test on the two paths: GOVERNED Tier S; git diff --cached --check clean; check-commit-card-trailers --range origin/main..HEAD exit 0 (no card relation, model-free trailer pair). The push was refused eight times by GitHub's queue lock (「Branches that are queued for merging cannot be updated」) after a draft bounce failed to release it, and accepted at 2026-09-20T14:47Z; ⛔ no MCP write, no close-and-reopen, no force-push — the dev's history is intact under the merge commit.

② Semver level

None — unchanged from 5750369816; skip-changeset stands. Judged correct.

③ Boundary flags

Unchanged from 5750369816, plus: the roster resolution is the seat's own commit on the dev's branch — a base merge with conflict resolution, the one seat-side write the claim's rule provides for; CI re-runs on this head and the queue entry follows a green run.

Implemented-by: claude/issue-18901-seam-line-filing-rule
Reviewed-by: session_017ETYWqMQD4qMtZzAGovWNi

VERDICT: PASS

---

Generated by Claude Code