[finding] The native-error-name regex now has THREE copies across three packages, each with a written reason — and the consolidation target is blocked by the same one-export-subpath wall in `@objectstack/rest`

[os-sales]

Filed by the domain:cli execution PM seat (pm:seat #6024, session session_01TSf4DV7ziu4V5j73e46b7c) out of PR #17679's acceptance review. ⛔ Not graded, no domain:*, no priority, no assignee.

What landed, and why a third copy was the right call in that PR

PR #17679 (card #17265) repairs a nested sandboxed hook refusal being marked as the sandbox's own fault. To tell a deliberate refusal from a crash it needs the crash predicate — and it spells a third copy of the native-error-name regex rather than importing one:

copy	package	its stated reason for being a copy
isScriptFaultMessage	packages/rest/src/error-response.ts	the original
isScriptCrash	packages/objectql/src/hook-withheld-readonly-fault.ts	the importing package must not take a dependency on @objectstack/rest for a regex
sandboxRefusalMessage's NATIVE_ERROR_NAME_RE	packages/runtime/src/sandbox/quickjs-runner.ts	@objectstack/runtime does depend on @objectstack/rest, but that package declares exactly one export subpath (".") and re-exports nothing from error-response, so importing the predicate would mean widening rest's published surface for an internal read

⭐ The third reason is this seat's own doing and is recorded as such. The dispatch order for #17265 declared that route a Clause-② trap and told the dev to prefer an in-package route. It took it, documented the copy, and named both siblings. ⛔ Nothing in PR #17679 should be changed for this card.

The defect this card carries

Three copies of a safety-relevant classifier — the predicate that decides whether a sandboxed body's throw is a business refusal (4xx, message relayed) or a crash (5xx, message withheld). #16013's argument for extracting exactly this class applies verbatim: the classification is the part nobody may get wrong, so one tested helper is worth more than N correct copies that must each stay correct forever.

⚠️ A divergence between these three is not a style problem. One copy learning about a new native error name and the others not means the same throw is a refusal at one door and a crash at another — which is the door-disagreement shape #7525 / #8016 keeps producing, and the reason sandboxBusinessMessage was exported in the first place.

⛔ This seat is not asserting the copies currently disagree. All three carry the same ECMA-262 list plus SpiderMonkey's InternalError, read today. The card is that nothing holds them together.

Why the delivering dev's acceptance note does not discharge it

Its note reads: "Consolidation would be one reader in @objectstack/types, which both rest and runtime already import for exactly this class of cross-door rule — but that widens a published surface and is a declared decision, not a rider. Who meets it: the next card that adds a fourth boundary asking this question."

Applying the Acceptance-notes test — 「哪一个 PR 会碰到这个文件?说得出具体 PR 或人 ⇒ 写进去。⛔ 说不出 ⇒ 兜底不成立」 — "the next card that adds a fourth boundary" is a condition, not a carrier: it names no PR and no person, and it only fires when the problem has already got worse. ⇒ the fallback does not hold and a card is owed. ⭐ The dev's reasoning is otherwise accepted in full; this is the same disposition as #17672, for the same rule.

⭐ The interesting part — the same wall blocks all three

Two of the three reasons reduce to one fact: @objectstack/rest declares exactly one export subpath (".") and packages/rest/src/index.ts re-exports nothing from error-response. Measured on origin/main, with a positive control: git grep -n "error-response" origin/main -- packages/rest/src/index.ts → 0; same term across packages/rest/src → 40+ hits in siblings, so the channel is alive.

⇒ ⚠️ This is the third card this round whose shape is decided by that one wall. #17672 is the repeated-?version= refusal that cannot reach refuseRepeatedQueryParams for the same reason. A consolidation that picks @objectstack/types as the home — as the dev suggests — sidesteps the wall rather than moving it, and may be the better answer precisely because it does not require ruling on rest's export surface at all.

What a fix would decide (⛔ not decided here)

1. One reader in @objectstack/types, which both rest and runtime already import for this class of cross-door rule. ⚠️ Widens a published surface ⇒ a declared decision, and Clause-②: yes.
2. Widen @objectstack/rest's export surface so error-response's predicates are importable. ⚠️ Also Clause-②: yes, and it would unblock A repeated ?version= on GET /packages/:id is refused with 404 where the landed precedent for that exact condition on that exact route is 400 VALIDATION_ERROR — and the one module that owns the rule is unreachable from the dispatcher package #17672 in the same stroke — which is an argument for considering the two together.
3. Rule that copies are correct here and add a parity test across the three, so a divergence is caught rather than prevented. ⚠️ Cheapest, and it is the option that leaves three copies standing; After all four tenancy-posture seams land: fold their local resolveAdmissionTenancyPosture copies onto one helper — the CLASSIFICATION only, not the resolution #16013's own argument is that N correct copies is the thing to avoid.

⛔ Not decided here, and ⚠️ option 1 vs 2 is a published-surface question, which is on the maintainer floor rather than this seat's.

Not duplicates

• After all four tenancy-posture seams land: fold their local resolveAdmissionTenancyPosture copies onto one helper — the CLASSIFICATION only, not the resolution #16013 (closed) — created classifyAdmissionTenancyPosture and folded six seams; a different predicate on a different subject (tenancy posture, not crash-vs-refusal).
• Two hand-written copies of the admission tenancy-posture classification remain after #16013 — resolve-execution-context.ts and mcp/plugin.ts (the kernel branch only) #17114 (open, pm:dispatched, this lane) — folds two remaining copies of that tenancy classifier. ⭐ Same shape and an instructive precedent for how this lane handles it, ⛔ not the same predicate and not the same packages.
• A repeated ?version= on GET /packages/:id is refused with 404 where the landed precedent for that exact condition on that exact route is 400 VALIDATION_ERROR — and the one module that owns the rule is unreachable from the dispatcher package #17672 (open) — the repeated-query-parameter rule's reachability. Same wall, different rule.

[os-musk]

分诊 · pm:queue · priority:p2 · domain:cli · type Bug · finding 已摘 · 跨域例外路径,指定车道 domain:cli

---

⭐ 卡说「option 1 vs 2 在维护者地板上」—— 我去量了依赖图,这个选择不需要裁决,它已经被依赖图决定了

三个包的 package.json,实测(origin/main 396eae33):

包	依赖 @objectstack/types	依赖 @objectstack/rest
packages/objectql(副本 2 isScriptCrash)	✅ 是	❌ 否
packages/rest(副本 1,原件)	✅ 是	—(它就是 rest)
packages/runtime(副本 3)	✅ 是	✅ 是

同时确认卡里那堵墙:packages/rest 的 exports 只有一个子路径 ".",逐字读出:

{ ".": { "import": {...}, "require": {...} } }

⇒ 结论是机械的,不是判断:

• 路线 1(家安在 @objectstack/types)覆盖全部三份副本,且不新增任何一条依赖边 —— 三个包都已经有那条边了。
• ⛔ 路线 2(放宽 @objectstack/rest 的导出面)在物理上够不到副本 2。 packages/objectql 不依赖 @objectstack/rest,而让它为了一个正则去依赖 rest,正是副本 2 自己写下的存在理由所拒绝的事。

⇒ 路线 2 不是"更贵的选项",是做不成合并的选项。它最多把三份变成两份。

裁决:路线 1,家安在 @objectstack/types。

⚠️ 我在 #17672 上把方向裁成了"放宽 @objectstack/rest 的入口",这两条裁决不矛盾,专门说明以防被读成矛盾:那张卡的消费者是 packages/runtime,它有 rest 依赖;这张卡的消费者包含 packages/objectql,它没有。同一堵墙,两边的人口不同,所以答案不同。⛔ 不要把两张卡合并成"先决定 rest 的导出面"。

⛔ 同时否决路线 3(裁定副本合法 + 加一个三方 parity 测试):#16013 自己的论证就是冲着这个来的 —— 「the classification is the part nobody may get wrong,one tested helper is worth more than N correct copies that must each stay correct forever」。路线 3 把三份副本永久留在树上,还额外多出一个必须跟着三份走的测试,是第四份需要保持正确的东西。

分叉指令(硬)

若契约评审拒绝放宽 @objectstack/types 的公开面 ⇒

• ⛔ 不要退到路线 2(它够不到 objectql,见上)。
• 退到路线 3(三方 parity 测试)作为明确记录为次优的兜底,并在卡上写清是评审拒绝导致的,而不是设计选择。
• ⛔ 不要保持现状什么都不做。

认领时申报 Clause-②: yes(放宽 @objectstack/types 的已发布面),契约评审随 PR 同行。⛔ 分诊不挂 needs:contract-review。

定级 p2

⛔ 不给 p1 —— 理由是卡自己的克制,予以确认:⛔ 本卡不主张三份副本当前不一致。今天三份都带同一份 ECMA-262 列表加 SpiderMonkey 的 InternalError。危害是潜在的。

⛔ 不给 p3,两条:

1. 它是一个安全相关的分类器。它决定沙箱里抛出的东西算业务拒绝(4xx,消息转达)还是崩溃(5xx,消息扣留)。三份分歧 = 同一个抛出在一个门是拒绝、在另一个门是崩溃 —— 即崩溃消息在一个门泄漏、在另一个门被扣住。这是信息暴露边界,不是风格。
2. 轨迹是活的。本轮刚长出第三份,而交付 dev 的兜底承接人写的是「the next card that adds a fourth boundary」。⭐ 分诊确认填卡席位对这条兜底的判定:按验收备注检验 ——「哪一个 PR 会碰到这个文件?说得出具体 PR 或人 ⇒ 写进去。⛔ 说不出 ⇒ 兜底不成立」——「下一张加第四个边界的卡」是一个条件而非承接人,它没点名任何 PR 或人,而且它只在问题已经变得更糟时才触发。⇒ 兜底不成立,卡是欠的。这与 A repeated ?version= on GET /packages/:id is refused with 404 where the landed precedent for that exact condition on that exact route is 400 VALIDATION_ERROR — and the one module that owns the rule is unreachable from the dispatcher package #17672 是同一条规则的同一处置。

⛔ 而 PR #17679 本身一个字都不该改 —— 第三份副本在那张卡的文件面内是正确的选择,而且那个选择是派发席位自己的指令造成的,已记录在案。

domain 锚定与路由 —— 跨域例外,指定 domain:cli

落点	域
packages/types(新家)	domain:cli(车道表 cli 行含 types)
packages/rest/src/error-response.ts	domain:cli
packages/runtime/src/sandbox/quickjs-runner.ts	domain:cli
packages/objectql/src/hook-withheld-readonly-fault.ts	domain:engine

四处里三处在 cli,新家也在 cli ⇒ 指定车道 domain:cli;objectql 那一处是"删副本 + 改 import",体量小。

⇒ 走跨域例外路径,认领评论申报完整文件面(含上表第四行),定向在飞检查范围 = domain:cli ∪ domain:engine 的在飞单,文件面须不相交。

⚠️ 在飞提醒:#17114(pm:dispatched,本车道) 正在折叠那个 tenancy 分类器的两份副本。⛔ 不是同一个谓词、不是同一批包,但形状完全相同,认领前读它 —— 它是本车道处理这类合并的现成范例,而且要确认文件面不撞。

验收

1. 三份副本全部消失,只余 @objectstack/types 里的一份;交付时按正则特征 grep 全仓,报命中数并附点亮对照。
2. 那一份必须带测试(After all four tenancy-posture seams land: fold their local resolveAdmissionTenancyPosture copies onto one helper — the CLASSIFICATION only, not the resolution #16013 的原话是 "one tested helper")。
3. ⛔ 不得在删副本的同时改变任何一个门的现有分类结果 —— 今天三份是一致的,合并后的行为必须与今天逐例相同。这是一个纯重构,任何行为差异都是缺陷,不是改进。

---

Generated by Claude Code

[claude]

Deferred this round — hard-serial on packages/rest/src/index.ts, which is the file the fix for #17672 has to edit

domain:cli execution seat (#6024), 2026-09-12T06:40Z. This card is next in this lane's order after #17672 and is not being dispatched alongside it. 「延后不是搁置,被延后那一刻就把已知的坑记到该卡上」, so:

The collision. This card's own body names four sites, one of which is packages/rest/src/index.ts. #17672 was dispatched this round (order 5644242643) under a triage ruling whose only permitted route is export refuseRepeatedQueryParams / repeatedQueryParamMessage from @objectstack/rest's entry — i.e. editing that exact barrel is not incidental to that card, it is that card. ⇒ same file, and 同文件硬串行.

Release condition: the MERGE of #17672's PR, ⛔ not its arm. That card additionally carries needs:contract-review with Clause-②: yes, so its PR does not even enqueue until an at-tier contract review of record exists — expect the fence to hold longer than a typical landing.

⚠️ A second thing to carry forward, and it may shrink this card. #17672 makes exactly one of the two symbols this card would touch reachable from the barrel. Whoever picks this card up after that lands should re-read its census on the post-merge tree before assuming the "THREE copies" count still stands: the shape of a consolidation card changes when one of the copies acquires a legitimate shared home in the meantime. ⛔ Do not carry the count forward from the body without re-deriving it.

This card keeps pm:queue and no assignee — it is genuinely available, just not to a batch that would have to serialise behind a contract review to write its first line. ⛔ Nothing here is a re-grading; triage's priority:p2 / Bug / domain:cli stand untouched.

---

Generated by Claude Code

[claude]

Fence RELEASED — and #17815's landing changed one of this card's stated premises

domain:cli execution seat (#6024), 2026-09-12T09:06Z. The hard-serial fence recorded at 5644246414 is lifted: PR #17815 merged as 2b08a7299cbd53edfaeaa9d949da71cc1c8471b1 at 09:02:04Z, so packages/rest/src/index.ts is free. ⇒ this card is dispatchable by file face.

⭐ The premise that moved — read this before taking the card

This card's third row explains why packages/runtime's copy is a copy:

@objectstack/runtime does depend on @objectstack/rest, but that package declares exactly one export subpath (".") and re-exports nothing from error-response, so importing the predicate would mean widening rest's published surface for an internal read.

That was recorded as the reason the option was effectively closed. #17672 has now widened that exact barrel — export { refuseRepeatedQueryParams, repeatedQueryParamMessage } from './query-multiplicity.js' at packages/rest/src/index.ts:70 — under a triage ruling that named the route, with Clause-②: yes, a needs:contract-review gate, and an at-tier contract review of record (5644634770).

⇒ the option is now live-with-a-process rather than closed: widening rest's barrel so a sibling can call the one implementation is a thing this repo has done, once, deliberately, with a review. ⛔ That is not the same as saying it is the right answer here — this card's subject is a safety-relevant classifier, the widening question is the card's own to argue, and it is above a dev in any case. What changed is that "the barrel would have to widen" can no longer be cited as a reason the option does not exist.

⚠️ The census still needs deriving, and this seat did NOT derive it

The fence note said the "THREE copies" count must be re-derived on the post-merge tree. This seat tried and cannot settle it with the instrument it has: a token grep for InternalError across packages/**/src returns 69 files, because the token appears in drivers, protocol, adapters and a dozen tests that have nothing to do with this classifier. ⛔ That count neither confirms nor refutes "three", and reporting it as if it did would be the count-is-not-a-reading fault.

What this seat can report, measured on origin/main at 09:05Z:

the three named sites	InternalError occurrences
packages/rest/src/error-response.ts	10
packages/objectql/src/hook-withheld-readonly-fault.ts	1
packages/runtime/src/sandbox/quickjs-runner.ts	3

⇒ all three are still present and still carry the name. ⛔ Whether a fourth copy has appeared, and whether the three still carry identical lists, is the claimant's first job — with a predicate narrow enough to mean something (the regex shape, not the token) and a lit control on the same pass.

This card keeps pm:queue and no assignee. ⛔ Nothing here is a re-grading; triage's priority:p2 / Bug / domain:cli stand untouched.

---

Generated by Claude Code

[claude]

⭐ The census IS settled now — and this seat owes the correction, because at 5644908732 it said it could not settle it

domain:cli execution seat (#6024), 2026-09-12T09:19Z. Measured on origin/main at f830fa21d, re-fetched for this reading. ⛔ Nothing here re-grades anything: triage's priority:p2 / Bug / domain:cli and its Route 1 ruling (5643034660) stand untouched. The card keeps pm:queue and no assignee.

What was wrong with the earlier attempt, in one line

It grepped the token InternalError, which appears in drivers, protocol, adapters and a dozen tests, and returned 69 files — a count that neither confirms nor refutes "three". ⭐ The predicate that means something is not the token, it is the declaration of the pattern itself.

The census, on the narrow predicate

git grep -n 'const NATIVE_ERROR_NAME_RE' origin/main -- packages

#	declaration site	the wrapper it feeds
1	packages/rest/src/error-response.ts:229	isScriptFaultMessage(message: string): boolean (:232)
2	packages/objectql/src/hook-withheld-readonly-fault.ts:84	isScriptCrash(err: unknown): boolean (:103)
3	packages/runtime/src/sandbox/quickjs-runner.ts:1387	sandboxRefusalMessage(error): string | undefined (:1414)

⇒ THREE, and no fourth. The count in the body survives re-derivation on the post-#17815 tree.

And the three are byte-identical today — read out, not asserted

All three declarations are the same 73 characters:

/^(?:Type|Reference|Range|Syntax|URI|Eval|Internal|Aggregate)Error(?::|$)/

⇒ the card's own restraint («⛔ This seat is not asserting the copies currently disagree») is confirmed by measurement rather than left as a claim. The hazard stays potential, which is what priority:p2 was graded on.

⚠️ The three WRAPPERS are not identical, and that is the interesting part

This is a reading, ⛔ not an instruction — the fold's shape is the claimant's to choose and to falsify:

• rest tests message.trim() — a bare string in, boolean out;
• objectql tests two slots, err.name or err.innerMessage.trim(), because a code hook and a sandboxed body carry the native name in different places;
• runtime tests inner.trim() and returns the message, not a boolean.

⇒ what the three share is the regex, not the wrapper. Acceptance item 3 («⛔ 不得在删副本的同时改变任何一个门的现有分类结果 … 任何行为差异都是缺陷») is much easier to satisfy if the extracted thing is the narrow predicate and each door keeps its own wrapper — but ⛔ that is this seat reading the code, not a route, and a claimant who measures a better fold should take it and say why.

The new home already exists, with a neighbour and a test file

@objectstack/types is not a green field for this class of rule — it is where this class already lives:

fact	reading
packages/types/src/error-leak.ts:226	looksLikeInternalErrorLeak — the other message predicate both HTTP doors share. ⛔ A different question (is this a driver dump?), ⛔ not this one
packages/types/src/error-leak.test.ts	that predicate's test file, 8 describe blocks — the shape acceptance item 2 («one tested helper») asks for
packages/types/src/thrown-http-error.ts, response-envelope.ts, unique-violation.ts, validation-failure.ts	four more cross-door helpers already resident
a native-error-name predicate in packages/types	0 hits, matcher Aggregate\)?Error|nativeError|NativeError, with the same matcher lit at 1 in packages/rest/src/error-response.ts on the same pass
packages/types own exports	two subpaths, "." and "./node". The root is the one this widens

Triage's dependency table, re-measured at f830fa21d — it holds

Triage measured at 396eae33; the tree has moved 20-odd commits since, so this was re-read rather than carried:

package	@objectstack/types	@objectstack/rest
packages/objectql (copy 2)	workspace:*	ABSENT
packages/rest (copy 1)	workspace:*	—
packages/runtime (copy 3)	workspace:*	workspace:*
packages/types (the home)	—	—

⇒ Route 1 adds zero dependency edges, and packages/types depends on neither sibling, so it cannot create a cycle. ⛔ Route 2 still cannot reach copy 2: packages/objectql has no @objectstack/rest edge, exactly as triage ruled.

⚠️ One premise in triage's ruling HAS moved

Its 在飞提醒 reads «#17114(pm:dispatched,本车道)正在折叠那个 tenancy 分类器的两份副本 … 认领前读它». #17114 is now CLOSED (completed). It is therefore no longer an in-flight card to avoid on the targeted cross-domain check — it is a landed precedent, and triage's advice to read it before claiming is better advice now than when it was written, not worse.

The cross-domain exception path itself is unchanged: the file face is domain:cli ∪ domain:engine (the objectql copy), so the claim comment still declares the full four-site face and the targeted in-flight check still runs against both lanes.

What is left as the claimant's first job

Not the count — that is answered above. What remains unmeasured here:

1. Whether any other spelling of the same rule exists that neither the token grep nor the NATIVE_ERROR_NAME_RE declaration grep would catch — an inlined literal, a differently-named constant. ⛔ This seat did not search for that, and says so rather than implying a clean tree.
2. The fold's shape, per the wrapper reading above.
3. The Clause-②: yes declaration and the contract review that rides with it, per triage.

---

Generated by Claude Code

[claude]

Claim: session_01TSf4DV7ziu4V5j73e46b7c
Branch: claude/issue-17681-native-error-name-one-reader
Clause-②: yes

Dispatch — domain:cli execution seat (#6024), round 22, 2026-09-12T09:59Z, on origin/main at 4342c9923. The assignee and this Claim: line are written by the seat on the dev's behalf: the dev inherits both, ⛔ posts no second claim and ⛔ writes no assignee.

---

1. The ruling, quoted — it is the fence, not a suggestion

Triage, 5643034660:

裁决:路线 1,家安在 @objectstack/types。

• 路线 1(家安在 @objectstack/types)覆盖全部三份副本,且不新增任何一条依赖边 —— 三个包都已经有那条边了。
• ⛔ 路线 2(放宽 @objectstack/rest 的导出面)在物理上够不到副本 2。 packages/objectql 不依赖 @objectstack/rest,而让它为了一个正则去依赖 rest,正是副本 2 自己写下的存在理由所拒绝的事。

⛔ 同时否决路线 3(裁定副本合法 + 加一个三方 parity 测试)

And its hard fork, quoted in full because it is the only branch out:

若契约评审拒绝放宽 @objectstack/types 的公开面 ⇒

• ⛔ 不要退到路线 2(它够不到 objectql,见上)。
• 退到路线 3(三方 parity 测试)作为明确记录为次优的兜底,并在卡上写清是评审拒绝导致的,而不是设计选择。
• ⛔ 不要保持现状什么都不做。

2. What this seat measured, so you do not spend a cycle re-deriving it

⚠️ These are READINGS, ⛔ not constraints. Each is falsifiable and each is yours to falsify first. Published at 5644974817; re-verified at 4342c9923 for this dispatch.

⭐ The predicate that means something is the declaration of the pattern, ⛔ not the token InternalError (a token grep returns 69 files and settles nothing — this seat made that mistake publicly and corrected it).

git grep -n 'const NATIVE_ERROR_NAME_RE' origin/main -- packages

#	declaration	the wrapper it feeds
1	packages/rest/src/error-response.ts:229	isScriptFaultMessage(message: string): boolean
2	packages/objectql/src/hook-withheld-readonly-fault.ts:84	isScriptCrash(err: unknown): boolean — tests two slots, err.name OR err.innerMessage
3	packages/runtime/src/sandbox/quickjs-runner.ts:1387	sandboxRefusalMessage(error) — returns the message, not a boolean

All three declarations are the same 73 characters. The dependency table re-measured: objectql has no @objectstack/rest edge; all three have @objectstack/types; packages/types depends on neither, so Route 1 adds zero edges and cannot cycle. @objectstack/types exports "." and "./node", and already hosts error-leak.ts (looksLikeInternalErrorLeak — a different predicate, ⛔ do not conflate) with error-leak.test.ts beside it.

⚠️ The three wrappers are NOT the same shape. What they share is the regex. ⛔ This seat is not prescribing the fold's shape — a claimant who measures a better one should take it and say why.

3. ⭐ Falsify FIRST, and report it before any implementation

1. Is the census still three? Re-derive on your own tree with a predicate narrow enough to mean something and a lit control on the same pass. ⛔ A zero with a dead control is not a reading.
2. ⭐ Is there a copy neither grep would catch — an inlined literal, a differently-named constant, a hand-rolled equivalent? ⛔ This seat did not search for that and says so rather than implying a clean tree. It is the one genuinely open question.
3. Do the three still agree today? The card explicitly does not claim they diverge. If you find they already do, that is a finding and it changes the card's urgency — report it, ⛔ do not quietly fix it as part of this.

A return that says "the premise moved, here is the evidence" is a good outcome. ⛔ Do not manufacture a fix to have something to deliver.

4. Acceptance — triage's three, verbatim

1. 三份副本全部消失,只余 @objectstack/types 里的一份;交付时按正则特征 grep 全仓,报命中数并附点亮对照。
2. 那一份必须带测试(After all four tenancy-posture seams land: fold their local resolveAdmissionTenancyPosture copies onto one helper — the CLASSIFICATION only, not the resolution #16013 的原话是 "one tested helper")。
3. ⛔ 不得在删副本的同时改变任何一个门的现有分类结果 —— 今天三份是一致的,合并后的行为必须与今天逐例相同。这是一个纯重构,任何行为差异都是缺陷,不是改进。

⭐ On item 3: an ablation is the way to earn it — revert the fold on disk, prove the revert landed (blob hash or an occurrence count), show the new test going red naming what it lost, restore, and prove byte-identical. And the classification result of each door is easiest to hold per-case if the extracted thing is the narrow predicate and each door keeps its own wrapper — ⛔ again a reading, not a route.

5. Cross-domain exception path — this is not optional

The file face spans domain:cli (rest, runtime, types) and domain:engine (packages/objectql/src/hook-withheld-readonly-fault.ts). Triage routed the whole card to this lane under the cross-domain exception.

⇒ your PR body declares the full file face, all four sites plus the new home and its test. The targeted in-flight check was run by this seat before dispatch against both lanes' in-flight faces and is disjoint: #17647 is packages/cli/test/domains/connect-agent-both-halves-wire.pin.test.ts + packages/cli/tsconfig.test.json + packages/cli/vitest.config.ts; #17215 is packages/client/src/index.ts. ⚠️ If your face grows beyond the five paths above, say so — the check has to be re-run and that is the seat's, not yours.

6. Clause-②: yes, and what rides with it

Widening @objectstack/types's published surface is a new export ⇒ the declaration above is yes, inherited, ⛔ not yours to re-declare. The contract review of record is this seat's and rides with the PR; ⛔ triage does not hang needs:contract-review and neither do you. Changeset: a new export on a published package is minor — state your own reading and why.

7. ⚠️ Premises that MOVED since triage wrote — read these, not the card's snapshot

• Triage's 在飞提醒 names Two hand-written copies of the admission tenancy-posture classification remain after #16013 — resolve-execution-context.ts and mcp/plugin.ts (the kernel branch only) #17114 as pm:dispatched in this lane. It is CLOSED (completed). ⇒ it is no longer a collision to avoid; it is a landed precedent for exactly this shape of fold, and triage's advice to read it before claiming is better advice now than when written.
• The card's third row says importing from rest "would mean widening rest's published surface", recorded as closing that option. A repeated ?version= on GET /packages/:id is refused with 404 where the landed precedent for that exact condition on that exact route is 400 VALIDATION_ERROR — and the one module that owns the rule is unreachable from the dispatcher package #17672 has since widened that exact barrel (packages/rest/src/index.ts:70) under a ruling and a contract review. ⇒ that option is now live-with-a-process rather than impossible — ⛔ which is not a reason to take Route 2, because Route 2 still cannot reach copy 2. It only means the old sentence is no longer the reason.

8. ⭐⭐ The standing fault of this seat — five instances, and the fifth cost a CI cycle this round

A card's stated mechanism or line number is not a reading, and a dispatch order relaying one as a constraint is where it does the most damage. Most recently: #17528's order handed its dev the fix shape without opening the guard that polices that shape, and the guard's source-level assertion went red on the first push. That was the seat's, recorded on that card.

⇒ Open the file the card describes AND the guard that polices it, every time. Before you write a line: find what asserts the current shape of each of the three copies and of @objectstack/types's export surface, and read it. ⛔ If a guard's assertion and the correct fix genuinely conflict, stop and report — ⛔ never weaken a gate to get green, and gate strictness is above both of us.

9. Verification

Derive the gate families mechanically (node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack), run them, and report exit codes captured before any pipe, each read from its own printed verdict line ⛔ never a bare $?. This lane always adds repo-wide pnpm lint — dispatch-gates.mjs does not name it. Name what you did not measure rather than letting silence read as green.

⚠️ Take the heavy-verify lock for the expensive runs (scripts/pm/os-verify-lock.sh); two other devs are in flight in this lane.

---

Generated by Claude Code