docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)

[os-litant]

Fixes#14146

Five semantic minors in the axis-weighting / batch-presentation encoding, caught by the at-tier compensating review of PR #14142 and fixed exactly as the reviewer prescribed. Every edit is in place: zero net lines in all four ratcheted files, and the published entry shrinks.

Positions are re-measured on today's main (224f8ea4a); the card's line numbers predate two merges, so they are restated below with the old number in parentheses.

Item 1 — stale cap-form restatement replaced by a pointer

Landing:.claude/skills/pm-dispatch/SKILL.mdL454 (card: L441)

before	…照旧现场呈报,走每批 ≤5 与详细写法纪律。
after	…照旧现场呈报,走常设决裁批流程与详细写法纪律。

The class-2/3 live-presentation route carried an inline 每批 ≤5 while the single home now rules 恰好 5 张 — a seat following the old line could present a 2-card batch and claim compliance. The number is gone; 常设决裁批流程 is the verbatim section heading in references/decision-analysis.md, so the pointer is greppable and there is now exactly one place that states the batch size. Line is 119 bytes (limit 120), zero lines added.

Item 2 — the final partial batch gets its legal move

Landing:.claude/skills/pm-dispatch/references/decision-analysis.mdL13

before	- 每批**恰好 5 张,呈完即停等回批**(2026-09-01
after	- 每批**恰好 5 张(末批不足照呈),呈完即停等回批**(2026-09-01

恰好 5 张 plus 收件箱读数为零才收工 (L17) forbade both presenting fewer than 5 and holding the remainder — the last batch of any queue that is not a multiple of five had no legal move. Inserted in place on the existing line: 57 B to 77 B, file stays at 54 lines (ceiling 54, headroom 0).

Item 3 — count-free wording, in every copy

Chosen: count-free wording. NOT a MENTION_PATTERNS extension. The reasoning, since the card left the choice open:

• The gate's own charter is "blind to prose and sharp on shape", and its %N% / %Q% design exists so that an anchor never spells the value it compares. A count − 1 mention check re-introduces a derived arithmetic relationship that the regex must know in two languages — and a regex can only ever see the spellings it enumerates. Miss one (另外三轴, 其余三条, the remaining three) and the gate is green while the text is stale: the exact defect being repaired, one layer up.
• Count-free wording removes the failure mode instead of policing it. After this PR there is no derived count anywhere in the four copies, so there is nothing for a future axis-count change to leave behind.
• A MENTION_PATTERNS extension landed alongside count-free wording would have zero live matches — a dormant pattern, which is the shape this repo's gates treat as rot.

frameCountMentions still sees and checks the 4 real count mentions (gate output below); this change adds nothing to it and removes nothing from it.

Six sites, all count − 1 hard-codings (the card named five weighting clauses; the sixth, L553, is the same defect class in the axis-① worked example — same file already in this claim's surface, same gate family, mechanical, no new verification surface):

landing	before	after
.claude/skills/pm-dispatch/SKILL.md L905	另三轴合起来投不翻它	其余各轴合起来投不翻它
references/decision-analysis.md L51	另三轴合起来投不翻①	其余各轴合起来投不翻①
.claude/agents/os-dev.md L418	推荐以它领起,另三轴合起来投不翻它	推荐以它领起,其余各轴合起来投不翻它
skills/objectstack-pm-dispatch/SKILL.md L553	On the other three axes they	On the other axes they
skills/objectstack-pm-dispatch/SKILL.md L593	the other three together cannot outvote	the others together cannot outvote
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it, the other three	lead with it, the others

另三轴 and the other three are now at 0 occurrences across all four copies.

Item 4 — published entry: dangling anchor and the dropped term

Landing:skills/objectstack-pm-dispatch/SKILL.mdL594 (card: L703) and L660-661 (card: L859)

site	before	after
L594	Read that weight through axis ②'s own definition — shrink special-cases and contract accretion, never grow them	Read that weight as shrinking special-cases and contract accretion, never growing them
L660-661	read it as shrinking special-cases rather than as licence to expand speculatively	read it as shrinking special-cases and contract accretion rather than as licence to expand speculatively

The L594 citation pointed at that file's axis-② definition for the shrink-not-grow reading, which that definition nowhere states (the real anchor deliberately does not ship under the route-B generalization) — it is now self-contained and asserts the reading directly. L660-661 regains and contract accretion, the term the zero-pull no-conflict argument turns on.

Item 5 — the lead-with-long-term duty, unconditional in all five encodings

The card's premise held on today's tree: 3 unconditional, 2 scoped to axis conflict.

encoding	before	after
.claude/skills/pm-dispatch/SKILL.md L904-905	四轴冲突时推荐以长远的读数领起,另三轴合起来投不翻它 — conditional	推荐以长远的读数领起,四轴冲突时其余各轴合起来投不翻它 — unconditional lead
skills/objectstack-pm-dispatch/SKILL.md L592-593	**Axis ② carries the highest weight — at least 50%** when the axes conflict: lead with the long-term reading, and… — conditional	**Axis ② carries the highest weight — at least 50%.** Lead with the long-term reading; when the axes conflict, the others…
references/decision-analysis.md L50-51	①恒 ≥50%,推荐以①领起 — already unconditional	unchanged (item 3 rewording only)
.claude/agents/os-dev.md L417-418	推荐以它领起 — already unconditional	unchanged (item 3 rewording only)
skills/objectstack-pm-dispatch/SKILL.md L659	lead with it — already unconditional	unchanged (item 3 rewording only)

Conflict-scoping is kept, in both rewritten copies, on the cannot-outvote consequence only — which is where it is meaningful, since there is nothing to outvote absent a conflict.

Line and token arithmetic, per file

file	lines before	lines after	ceiling	headroom
.claude/skills/pm-dispatch/SKILL.md	989	989	1005	16 (unchanged)
.claude/skills/pm-dispatch/references/decision-analysis.md	54	54	54	0 (unchanged)
.claude/agents/os-dev.md	469	469	469	0 (unchanged)
skills/objectstack-pm-dispatch/SKILL.md	753	753	not line-ratcheted	—

No line in any edited file exceeds the 120-byte rule; the widest edited lines are SKILL.md L454 (119 B), L904 (118 B) and os-dev.md L418 (120 B). max-table-row-bytes pins are untouched — no table was edited.

Published entry, token ratchet (convention ceil(utf8 bytes / 4)), measured per site:

site	items	before	after	delta
L553	3	77 B	71 B	−6 B
L592-596	3, 4, 5	372 B	335 B	−37 B
L657-662	3, 4	235 B	253 B	+18 B
file		38,832 B = 9,708 tokens	38,807 B = 9,702 tokens	−25 B / −6 tokens

How item 4's added words are paid, inside the same file, without a re-wrap and without a ceiling change: item 4's only cost is and contract accretion at L660-661, +23 B. It is paid by item 4's own other half — deleting the dangling citation through axis ②'s own definition — (38 B, replaced by as , 3 B) at L594 — with item 3's −16 B of count-free rewording on top. The ceiling stays at 9,708 and is not lowered, per the dispatch's explicit no-ceiling-change constraint; the file now sits 6 tokens under it.

Published-surface readings (skills/** diff)

Both readings the published surface requires, lines first, tokens beside them:

reading	before	after	delta
edited file — skills/objectstack-pm-dispatch/SKILL.md	753 lines · 38,832 B · 9,708 tokens	753 lines · 38,807 B · 9,702 tokens	0 lines · −25 B · −6 tokens
whole published package — all 11 skills/*/SKILL.md	6,862 lines · 318,948 B · 79,737 tokens	6,862 lines · 318,923 B · 79,731 tokens	0 lines · −25 B · −6 tokens

No net expansion in either reading: this PR corrects wording in place and the published surface comes out marginally smaller. There is no new capability here and therefore no "small feature, big write-up" to price — the one clause that gains words (item 4's and contract accretion) is paid twice over inside the same file.

skip-changeset

Applied. The diff publishes nothing from any package: .claude/** agent protocol plus the skills/** catalog, which ships through npx skills add, not through a package release. Measured rather than assumed — none of the six most recent commits touching skills/objectstack-pm-dispatch/SKILL.md carried a .changeset/ file (c985ae958, a9b2be0b0, 2bff79ae3, bcc9189e6, 3d0897459, d250754fd), and 2bff79ae3 is this card's own parent PR with the identical face set.

Gates

Every command ran through scripts/pm/os-verify-lock.sh (slot issue-14146), each exit code captured by redirect before any pipe. The union below was re-derived after the final commit with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands and run at dd9f7d39f — the branch head.

gate	verdict line
check:skill-frame-sync	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files · 4 axes: business-need → long-term-soundness → ai-authoring-safety → startup-scope-discipline · binding sentence present in all 4; 4 count mention(s) agree; 40 markdown files scanned for undeclared copies.
check:skill-frame-sync --self-test	✓ check-skill-frame-sync self-test: 12 cases pass, plus 5 dispatch-gates declaration cases.
check:skill-frame-freshness	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:skill-frame-freshness --self-test	✓ check-skill-frame-freshness self-test: 12 cases pass.
check:pm-skill-ratchet	✓ …SKILL.md is 989 lines (ceiling 1005; headroom 16). · ✓ …decision-analysis.md is 54 lines (ceiling 54; headroom 0). · ✓ …os-dev.md is 469 lines (ceiling 469; headroom 0).
check:pm-skill-ratchet --self-test	✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g). · self-test 14 cases pass.
check-skills-token-ratchet	✓ …skills/objectstack-pm-dispatch/SKILL.md is 9702 tokens (ceiling 9708; headroom 6). · ✓ …36 authored bundle file(s) within their ceilings
check-skills-token-ratchet --self-test	✓ check-skills-token-ratchet self-test: 64 cases pass.
@objectstack/spec check:skill-docs	✓ skills/README.md · ✓ content/docs/ai/skills-reference.mdx · ✅ Skill docs in sync
check:nul-bytes	check-nul-bytes: OK (scanned 8052 text file(s) — 8052 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes). · self-test 75 assertions

Full derived union at dd9f7d39f — 20 of 22 green, and the two that are not green are NOT MEASURED, not red:

check-ci-filter-parity · check-cross-package-test-inputs · check-shard-attestation · check-skills-token-ratchet · check:skill-docs · check:agent-model-declared · check:agent-test-spelling · check:corpus-claim-drift · check:cross-package-test-inputs · check:doc-authoring · check:nul-bytes · check:pm-governed-merges · check:pm-governed-prose · check:pm-skill-id-lint · check:pm-skill-ratchet · check:role-word · check:skill-compatibility · check:skill-frame-sync · check:skill-identifier-liveness · check:skill-frame-freshness — all green.

• check-test-completeness — exit 3, NOT MEASURED. Its own text: "That list names this script with NO argument… There is no local log to hand it, so the local reading for this gate is NOT MEASURED. ⛔ It is not a red, and there is nothing here to fix." It consumes a turbo run test log that only CI produces; this diff touches no test.
• check:doc-formula-expressions — exit 3, PREREQUISITE NOT MET (@objectstack/formula is not built in this worktree). Not a finding. Building it needs the shared verify lock, and two bounded acquisitions timed out at the 540 s budget behind a sibling's run (exit 99, place kept — also NOT MEASURED, never a red). Declared narrowing, measured from the gate's own predicate rather than asserted:fencedBlocks() keeps only fences whose language tag is ts, typescript or tsx, and the text scan is FIELD_RULE_TEXT_RE (visibleWhen|readonlyWhen|requiredWhen). Across all four edited files the gate would parse 0 fenced blocks — the fences present are bash, json, text and bare, none of them a parsed language — 0 of the 17 changed lines fall inside a parsed block, and there are 0FIELD_RULE_TEXT_RE occurrences. The gate has nothing to read in this diff, so its verdict cannot move. CI builds the package and runs it for real.

Beyond the derived union I also self-scanned the four edited files for raw control bytes (grep -naP '[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]'): zero hits.

Premise notes

• Card line numbers: all five landing points moved, as the card and the dispatch anticipated. Item 1 L441 to L454; item 4 L703 to L594 and L859 to L660-661. Every position above is re-measured on 224f8ea4a.
• Item 3's "five places" is six. Five are the weighting clauses the card enumerates; the sixth, published SKILL.md L553 (On the other three axes they), is the same count − 1 hard-coding inside axis ①'s worked example. Fixed here rather than filed: same defect class, same file already in this claim's surface, mechanical, no new verification surface — and leaving it would have left the drift class half-open in the copy that ships to third parties.
• Everything else on the card held. Item 5's split really is 3 unconditional / 2 conflict-scoped; the MENTION_PATTERNS blindness to 另N轴 is real (neither pattern can match it, verified against the script); both published copies do sit in the entry after the skills(pm-dispatch): split the entry into SKILL.md + rules/ under #14296 items 1 and 4 (entry 9,708 tokens, package net −21) #14660 split.
• Batch-3 item 4 ([Decision] Skills optimization program — batch 3 (5 items): DATA-F-07 defineHook() vs the platform's own examples · #13597 Phase-2 — X0 quote policy · CLAUDE.md excerpt form · the four-copy decision frame · cross-file line-budget moves #14685) may later retire two of the four copies. This fix is landed on all four and survives either ruling.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01LraLgQVGq8egUwfYZpbYt1

---

Generated by Claude Code

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 33706142983 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

失败的 job（日志抽取，best effort）：

• Test Core (1/6) — 失败步骤: Run this shard's tests

  @objectstack/cli:test: FAIL integration test/run-dev-unbuilt-workspace.e2e.test.ts > the mirror direction: a reader that is never coming back > gives up and exits instead of waiting forever
  ↳ 失败原因: @objectstack/cli:test: AssertionError: expected 'SIGKILL' to be null
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• test/run-dev-unbuilt-workspace.e2e.test.ts — 24h 内已弹出 19 个 PR（fix(plugin-sharing): let system writes materialize sharing rules — drop the isSystem skips in bindRuleHooks (#13533) #14528、fix(cli): let the dev artifact door own stack-declared security metadata on a host config #14631、fix(metadata): drop the leftover boot-time debug probe from MetadataPlugin.init #14649、fix(cli): walk bulkActions, validation messages and datasets in i18n extract, so the coverage ratchet can see them #14651、fix(objectql,metadata): derive a view container's object through the shared helper, so the row's own name is LAST at every SOURCE registrar (#14399) #14665、fix(cli): remove the ghost field types from generate.ts's three vocabularies #14675、test(runtime): measure which declaration copy the security/sharing seeders consume on an artifact boot with an engine #14687、Declare the scripts/ population two Silent gate families walk but never named #14692、test(cli): make the unread-reader ceiling a load-independent constant at RUN_TIMEOUT_MS #14715、test(cli): derive the vitest tier population instead of freezing it in a list #14717、fix(devx): partition os-regen-merge.sh's step-1 conflict message by class #14733、fix(spec): send a top-level flow trigger to the START node config, not to a type rename #14735、fix(deps): take the fix for the four OSV advisories blocking every manifest PR #14736、fix(tooling): a template-literal mode for the vocabulary gate's four shared textual scanners #14739、test(runtime): pin the measured loadMany vs by-name divergence for #14423 (a) #14741、test(plugin-approvals): pin the lock release in the tenant-admin reverse check #14750、fix(lint): consolidate five more private "Did you mean?" copies onto suggestName #14757、feat(dispatch-gates): a checkable whole-tree-population channel, so a repo-wide gate is placed by declaration instead of landing near undetermined #14774、docs(skills): repair five semantic minors in the axis-weighting and batch-presentation encoding #14779），分属 12 条投机栈 ⇒ 12 次独立命中。汇总 issue：Queue-flake anchor: test/run-dev-unbuilt-workspace.e2e.test.ts #14648（已刷新）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 39 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)