docs(agents): the Vite rewrite claim does not reproduce anywhere (objectui#9191)

[claude]

Fixes #9191

Measurement card. The bullet asserted a prescription (one spelling only) and a factual claim with a named mechanism (Vite rewrites the two-argument form to an http://localhost:3000/@fs/… URL, so fileURLToPath throws ERR_INVALID_URL_SCHEME under both cwds and the suite does not load). The prescription was never in question. The mechanism claim was measured, and it does not reproduce.

Governed surface — this PR opens DRAFT and stays DRAFT

node scripts/check-governed-queue-guard.mjs --test AGENTS.md (exit 3 = governed):

⛔ GOVERNED — 1 of 1 path(s) are on a governed surface:
   AGENTS.md x1 — the repo-root agent instruction file
     - AGENTS.md

   One governed path governs the WHOLE pull request — proportion is not a question.
   ⛔ Do not flip it ready, enqueue it, or arm auto-merge. Park it as a DRAFT and leave the merge
      to the maintainer; a human merge IS the review record for a governed surface.

Not flipped ready, not enqueued, no auto-merge, no approving review. The draft state is the finished state.

Premise readings (P1–P5), all on 541ce4e0, 2026-09-14T10:06Z–10:20Z

#	premise	verdict
P1	content of the bullet	holds. Anchored by content, measured at lines 236–238 for the sentence and 239 for the warning; the bullet opens at 231.
P2	population re-measured on the tip	holds, and it grew — see the table below.
P3	where the mechanism would live	located. vitest.config.mts declares four root projects — unit (node), dom and dom-heavy (happy-dom), dist (happy-dom, behind OBJECTUI_DIST_PINS) — plus the apps/console project, which merges that app's own vite.config.ts and its full plugin set. The root config registers no Vite plugin at all and declares no browser mode anywhere in the repo (control: the same grep finds the environment: keys it should). Nothing in it touches import.meta.url.
P4	the objectui#9190 gate declares this form silent	holds. Its header: "⚠️ This gate does NOT flag the two-argument new URL(rel, import.meta.url) form, and the reason is a measurement rather than a preference." Its self-test pins the case named the two-argument new URL form, which is self-rooted to the verdict 'silent'.
P5	objectui has no .claude/agents/os-dev.md	holds. The definition used is objectstack's.

P2 — the population, re-measured on 541ce4e0

Same predicate as the card's §2: a new URL( whose second argument is import.meta.url and whose first is a quoted string beginning with neither / nor a scheme, over *.ts,*.tsx,*.mjs,*.js,*.cjs; lines whose code part begins with *, // or /* separated out as comments.

reading	card, at c5b0cecf67	here, at 541ce4e0
all new URL(…, import.meta.url)	66	78
relative first argument — raw hits	27 (21 files)	31 (25 files)
— comment prose	4	4
— live call sites	23 (17 files)	27 (21 files)
— — in test files	21 (15 files)	23 (17 files)
— — outside the test transform	2	4

Controls, both with a known direction:

control	expected	got
a nonsense token in place of new URL(	0	0
bare import.meta.url, repo-wide	hundreds	741

Two notes on the non-test remainder. Three are new URL('.', import.meta.url) in scripts/*.mjs — the module's own directory under plain node, outside the stated mechanism twice over, exactly as the card says. The fourth is the objectui#9190 gate's own self-test fixture, which is the P4 pin rather than a call site.

The finding that drives everything below: all 23 live test-file sites are .test.ts, so every one of them runs in the unit/node project. The population the card weighed therefore never exercised happy-dom at all — which is precisely where the three prose comments locate the hazard. That is the gap this PR's fixture closes.

The reproduction attempt

A temporary fixture (never committed; deleted, with git status --porcelain empty and zero residual files afterwards) printed import.meta.url from inside a test and then did fileURLToPath(new URL(relative, import.meta.url)) plus a real readFileSync. Placed at packages/fields/src/widgets/ — the same directory as the comment that says "in this very file it evaluates to http://localhost:3000/@fs/…" — as both a .test.ts (so it lands in unit) and a .test.tsx (so it lands in dom), and again under apps/console/src/.

fixture	project / environment	cwd	import.meta.url	two-arg new URL	result
.test.ts	unit / node	repo root	file:///…	file:	read OK, 13581 bytes
.test.tsx	dom / happy-dom	repo root	file:///…	file:	read OK, 13581 bytes
.test.ts	unit / node	packages/fields (--root ../..)	file:///…	file:	read OK, 13581 bytes
.test.tsx	dom / happy-dom	packages/fields (--root ../..)	file:///…	file:	read OK, 13581 bytes
.test.tsx	apps/console / happy-dom + full plugin set	repo root	file:///…	file:	read OK, 19741 bytes

The bare-repo-root form new URL('../../../..', import.meta.url) resolved to file:///…/objectui-issue-9191/ in every row. No http: URL, no @fs, no ERR_INVALID_URL_SCHEME anywhere, under either cwd, in either environment.

dom-heavy and dist are not separately tabulated because neither adds a variable: dom-heavy is happy-dom with vitest.setup.dom.tsx, which is the setup the apps/console row already ran; dist is happy-dom with the light setup, which is the dom row's combination.

The existing population, run for real

Seven files — the four that carry the prose comments plus three others drawn from the 23 — under both invocations the sentence names.

invocation	files	tests	result
repo root, all seven at once	7	78	passed, exit 0
packages/plugin-view, --root ../..	1	6	passed, exit 0
packages/fields, --root ../..	1	7	passed, exit 0
packages/plugin-ai, --root ../..	1	14	passed, exit 0
packages/types, --root ../..	1	13	passed, exit 0
packages/app-shell, --root ../..	1	13	passed, exit 0
packages/react, --root ../..	1	5	passed, exit 0

Disposition: the card's 1 — the sentence is wrong about the mechanism

The edit keeps the ⛔ and the prescription, and replaces the mechanism with what was measured. Net +6/−4 inside the claimed region, one hunk, no line moved, no new section, the file's ⛔/⚠️ markers and its ASCII-punctuation register preserved (the file carries zero fullwidth semicolons or parentheses; the first draft of this hunk introduced some and was redone).

Why the other two are excluded by the readings, not by preference:

• Disposition 2 (a narrower condition exists) would need a condition that reproduces. Every project the config declares was probed, in both environments and under both cwds, including the one project that carries real Vite plugins. None produced the rewrite. There is no condition left to name, and naming one unmeasured would be the defect this card was filed about.
• Disposition 3 (latent hazard, wants a gate) requires the rewrite to be reachable. The @fs shape is produced by Vite's dev-server/browser transform, and this repo declares no vitest browser mode; I could not reach it by any invocation available. Asserting reachability I did not reproduce is exactly what this PR is removing, so I do not assert it — and per the dispatch a gate is a different card and is not built here.

The warning line is corrected rather than deleted: the process.cwd() class it declared ungated is gated today by scripts/check-test-path-roots.mjs (wired into lint.yml, green here), and that gate deliberately does not flag this spelling. That asymmetry is now what the line says.

Gates

Every command run in the foreground in the worktree, exit code captured before any pipe, at bef04584.

command	exit	evidence
node scripts/check-governed-queue-guard.mjs --test AGENTS.md	3	governed classification, quoted above
pnpm check:governed-queue-guard	0	OK check-governed-queue-guard self-test: 185 cases pass …
pnpm check:control-bytes	0	✅ check-control-bytes: OK (scanned 7658 tracked text file(s); skipped 85 binary).
pnpm docs:check-links	0	the one gate that reads AGENTS.md off disk ({ path: 'AGENTS.md', rule: 'disk' })
pnpm check:new-line-citations	0	the differential cross-file line-citation gate; this hunk adds none
pnpm check:test-path-roots	0	✅ check-test-path-roots: OK (1975 filesystem call(s) in 412 of 3150 test file(s) …)
node scripts/check-changeset-presence.mjs	0	No source or published contract of a released package changed in this range, so no changeset is owed.
pnpm changeset:check	0	✅ No changeset declares a major bump.
pnpm check:changeset-claims	0	—
the four instrument suites (doc-links, governed-queue-guard, test-path-roots, changeset-presence)	0	Test Files 4 passed (4) · Tests 212 passed (212)

Declared narrowing on lint and type-check. Both are turbo run tasks across every package; the repo-wide farm run belongs to CI. The narrowing here is measured, not assumed, in three pieces: (1) the population under test is read from ESLint's own config, whose files globs are **/*.{ts,tsx} and test-file variants of the same — never .md; (2) the file count is read from ESLint's --format json output, which for AGENTS.md returns one entry, errorCount: 0, and the message File ignored because no matching configuration was supplied.; (3) the invariance claim: this diff contains only markdown, so no ESLint-governed or tsc-program file changed, and no type-aware rule can see a byte of it. CI runs both in full.

Which gates these are, and why not git grep -l 'AGENTS.md' over tests. That grep returns 371 test files — effectively the whole suite — because they mention the filename in prose. Filtering to files that actually read it off disk returns zero test files; the control confirms the filter works (the same predicate does find scripts/check-doc-links.mjs's { path: 'AGENTS.md', rule: 'disk' } row and check-governed-queue-guard.mjs's exact: 'AGENTS.md' surface row). So the real instrument set is the two scripts plus their suites, all run above.

Acceptance notes

Observations for a successor, filed as nothing and fixing nothing here — this PR edits no test file, no script and no config.

Card §3 suspected at least one of the prose comments states something untrue of this tree. Three of the four do. All three are authors recording the hazard as their reason for avoiding the form, and the fixture above contradicts each on the same paths:

• packages/plugin-view/src/__tests__/ViewSwitcher.test.tsx, in the block beginning "NOT new URL('../x', import.meta.url)" — says Vite rewrites import.meta.url to a server-root-relative path yielding an absolute path missing the repo root, ENOENT for the whole suite. Measured in the dom project: a full file: URL, read succeeds. Note the failure mode it describes (ENOENT on a rootless path) is not the one AGENTS.md described (ERR_INVALID_URL_SCHEME on an http: URL) — two different stories about one form.
• packages/fields/src/widgets/CapabilityMultiSelectField.specParity-6285.test.tsx, at the bullet beginning "new URL('../../../..', import.meta.url) is a form VITE REWRITES" — says that in that very file it evaluates to http://localhost:3000/@fs/… under both cwds. The fixture sat in that same directory as a .test.tsx, so same project and same depth: file: URL under both cwds. This comment is the apparent source of the AGENTS.md sentence.
• packages/plugin-ai/src/registrationInputs-8178.test.ts, at "under the happy-dom project import.meta.url is not a file: URL" — measured directly in the dom/happy-dom project: it is a file: URL. This is the premise the card notes objectui#7800 already falsified by six readings; it is still in the tree, now with a seventh reading against it.

The fourth, scripts/__tests__/vite-maplibre-worker.test.ts, is correct and out of scope: it describes how maplibre-gl resolves its own worker as a sibling of its script URL, a third-party bundling fact, not a claim about this repo's test transform.

Successor: whoever next touches those three files. Each comment's repair is a one-line edit in a test file this PR is forbidden to open, and each is opportunistic-repair territory rather than a sweep. None of the three misleads anyone into a defect — they talk authors out of a form that is merely redundant — so the cost of leaving them is a reader believing a mechanism that is not there, which is the same cost this PR just paid down on the governed surface.

Clause-②: yes

---

Generated by Claude Code

[claude]

Contract review

Served-tier: 2573/2584, then CONTRACT_REVIEW_TIER
Head-sha: bef04584498e1d42b5b214e98e0271536b7d857a

In-seat record on a GOVERNED surface (objectui AGENTS.md, rules tier — the record is one of the four pieces; the landing waits for an authorized approval under ruling C), domain:skills execution PM seat, 2026-09-14T10:25Z. Spelled from node scripts/pm/check-clause2-carriers.mjs --template on objectstack origin/main. The dev's os-dev-report is comment 5662497171 on objectui#9191; the judgments below are the seat's own reading of the diff and of the tree.

① Derived judgments

• One hunk, inside the claim. AGENTS.md +6/−4 at :233–:242 (head) within the region claimed at :232–:239 on 541ce4e0; .changeset/issue-9191-new-url-import-meta-url-measured.md with empty frontmatter (no release — the file declares why). No test file, script or config moved; the sibling regions (objectui#9198's :383–:410, PR docs(agents): record the merge_group leg's SECOND refusal predicate (the contract-review carrier) #9466's :519) are untouched.
• The prescription survives the edit. 「⛔ 不引入第三种」 stays and the two-argument new URL(相对路径, import.meta.url) form is still 「别写」; what changes is its stated reason — from a Vite rewrite that the dev could not reproduce in any declared project or cwd to spelling uniqueness. The dev's fixture table (five rows, node and happy-dom, repo-root and package-directory cwds, all file: URLs with successful reads) and the seven-file population run (78 tests green at the root, six per-package runs green) are the readings the new sentence rests on; the seat did not re-run the fixture and adopts those rows as the dev's measurement, named as such.
• The corrected ⚠️ line states the tree. scripts/check-test-path-roots.mjs exists on objectui origin/main and is wired in .github/workflows/lint.yml and package.json (seat's own ls-tree and grep); its header at :121 reads 「This gate does NOT flag the two-argument new URL(rel, import.meta.url)」, which is exactly the asymmetry the new line names.
• Population control. The seat's simpler grep for relative two-argument sites on 541ce4e0 reads 26 raw hits; the dev's stricter predicate reads 31 raw / 27 live / 23 in test files with its own two controls (nonsense token 0; bare import.meta.url 741). Same order, consistent direction (the population grew since the card's 27).
• Register and mechanics. The hunk keeps the file's ASCII punctuation (the dev reverted a fullwidth first draft and proved the revert); the PR body carries no angle-bracket placeholder and no model identifier; Clause-②: yes stands on its own line; the changeset rule for a non-package change is followed in PR docs(agents): cite by content, not by line address (#8875 clause 1) #9301's shape.
• Gates as reported. check-governed-queue-guard --test AGENTS.md exit 3 (the governed classification, quoted); self-test 185 cases; check:control-bytes, docs:check-links, check:new-line-citations, check:test-path-roots, the changeset trio and the four instrument suites (212 tests) all exit 0; lint/type-check declared narrowed with three-piece evidence (markdown-only diff) and left to CI. CI on the head at the seat's read: 24 success, 3 skipped, 7 in progress — the landing step re-reads it green before anything moves.

② Semver level

None. AGENTS.md and a no-release changeset only; no published package source changes.

③ Boundary flags

• Governed rules-layer surface: objectui AGENTS.md. Four-piece applied: ACCEPT on the card, draft kept, needs-user-decision on this PR, reviewers os-zhuang + hotlong requested. ⛔ This seat never approves; ⛔ no ready / enqueue / auto-merge before an authorized APPROVED review (os-zhuang / hotlong, latest-decisive, not dismissed); after one exists this seat lands per ruling C.
• Clause ② judged, not told: declared yes because a ⛔ mechanism claim is softened; measured, the accepted set is unchanged — the form the sentence forbids is still forbidden, and no gate, contract or public surface moves. Not a widening.
• Region-parallel on the same governed file: objectui#9198's PR (this seat, :383–:410) and PR docs(agents): record the merge_group leg's SECOND refusal predicate (the contract-review carrier) #9466 (another seat, :519) hold disjoint hunks; whichever lands second merges origin/main first and the seat re-issues its record on the merged head. mergeable_state read behind at review time (main advanced), which is that clause, not a conflict.
• Observed, not filed: three test-file comments still state the absent mechanism (named in the PR's ## Acceptance notes); a fourth (vite-maplibre-worker.test.ts) is correct and out of scope. Successor: the files' next author; no card.

Implemented-by: claude/issue-9191-agents-new-url-sentence
Reviewed-by: session_01DAcomhvR9kKizeYgg89Vo8

VERDICT: PASS

---

Generated by Claude Code

[claude]

维护者速读

这个 PR 改的是 objectui AGENTS.md 里一句话的「理由」,不改规则本身。 那条测试路径规则说:别写 new URL(相对路径, import.meta.url) 这种第三种拼法。原文给的理由是「本仓测试里 Vite 会把它重写成 http://localhost:3000/@fs/…,fileURLToPath 会抛错,整个套件加载不起来」。卡 #9191 发现树里有 20 多处正在用这种拼法而且测试全绿,理由和事实对不上。

dev 实测了一遍:按 vitest.config.mts 声明的每个 project(node、happy-dom、apps/console 带完整插件)在仓根和包目录两种 cwd 下各跑一遍,import.meta.url 全是 file: URL,读文件都成功;重写不发生。于是:规则不动(还是「别写」),理由改成「拼法唯一」,并把下面那行「今天没有任何门拦得住」改准 —— process.cwd() 那一类现在有 pnpm check:test-path-roots 拦,这个拼法它按设计不拦。一处改动,+6/−4,不碰别的段落。

为什么要你点:AGENTS.md 是受管面,按裁决 C 需要 os-zhuang 或 hotlong 一次 approve,之后由本席落地。CI 目前 24 绿 7 在跑,落地前会再读一次全绿。

要不要批准这一句理由的改写? 是 / 否。

---

Generated by Claude Code

[claude]

Gate order corrected — domain:skills seat, 2026-09-14T10:26Z. The claim on objectui#9191 declared Clause-②: yes but the seat hung no needs:contract-review on either carrier, so check-clause2-carriers --pair 9486 read exit 4 (C3: gate never hung) after the record 5662529777 was already posted. The seat has now hung the gate on the card and on this PR and clears it in the next act, because the review that clears it — the ## Contract review record above, **VERDICT: PASS** on head bef04584 — already exists. Recorded rather than smoothed: the record preceded the gate by minutes; nothing else about the four-piece changes, and the PR stays draft until an authorized approval.

---

Generated by Claude Code

[claude]

Contract review

Served-tier: 10/10, then CONTRACT_REVIEW_TIER
Head-sha: bef04584498e1d42b5b214e98e0271536b7d857a

In-seat record on a GOVERNED surface (objectui AGENTS.md, rules tier — the record is one of the four pieces; the landing waits for an authorized approval under ruling C), domain:skills execution PM seat, 2026-09-14T10:27Z. Spelled from node scripts/pm/check-clause2-carriers.mjs --template on objectstack origin/main. Re-issued: the first record (5662529777) carried a whole-session stamp control that was not total (11 turns served off-tier during a fallback window on 2026-09-13, none since); the stamp above is the reviewing round's own count, every turn since the dev report at 2026-09-14T10:22Z at tier. Judgments unchanged. The dev's os-dev-report is comment 5662497171 on objectui#9191; the judgments below are the seat's own reading of the diff and of the tree.

① Derived judgments

• One hunk, inside the claim. AGENTS.md +6/−4 at :233–:242 (head) within the region claimed at :232–:239 on 541ce4e0; .changeset/issue-9191-new-url-import-meta-url-measured.md with empty frontmatter (no release — the file declares why). No test file, script or config moved; the sibling regions (objectui#9198's :383–:410, PR docs(agents): record the merge_group leg's SECOND refusal predicate (the contract-review carrier) #9466's :519) are untouched.
• The prescription survives the edit. 「⛔ 不引入第三种」 stays and the two-argument new URL(相对路径, import.meta.url) form is still 「别写」; what changes is its stated reason — from a Vite rewrite that the dev could not reproduce in any declared project or cwd to spelling uniqueness. The dev's fixture table (five rows, node and happy-dom, repo-root and package-directory cwds, all file: URLs with successful reads) and the seven-file population run (78 tests green at the root, six per-package runs green) are the readings the new sentence rests on; the seat did not re-run the fixture and adopts those rows as the dev's measurement, named as such.
• The corrected ⚠️ line states the tree. scripts/check-test-path-roots.mjs exists on objectui origin/main and is wired in .github/workflows/lint.yml and package.json (seat's own ls-tree and grep); its header at :121 reads 「This gate does NOT flag the two-argument new URL(rel, import.meta.url)」, which is exactly the asymmetry the new line names.
• Population control. The seat's simpler grep for relative two-argument sites on 541ce4e0 reads 26 raw hits; the dev's stricter predicate reads 31 raw / 27 live / 23 in test files with its own two controls (nonsense token 0; bare import.meta.url 741). Same order, consistent direction (the population grew since the card's 27).
• Register and mechanics. The hunk keeps the file's ASCII punctuation (the dev reverted a fullwidth first draft and proved the revert); the PR body carries no angle-bracket placeholder and no model identifier; Clause-②: yes stands on its own line; the changeset rule for a non-package change is followed in PR docs(agents): cite by content, not by line address (#8875 clause 1) #9301's shape.
• Gates as reported. check-governed-queue-guard --test AGENTS.md exit 3 (the governed classification, quoted); self-test 185 cases; check:control-bytes, docs:check-links, check:new-line-citations, check:test-path-roots, the changeset trio and the four instrument suites (212 tests) all exit 0; lint/type-check declared narrowed with three-piece evidence (markdown-only diff) and left to CI. CI on the head at the seat's read: 24 success, 3 skipped, 7 in progress — the landing step re-reads it green before anything moves.

② Semver level

None. AGENTS.md and a no-release changeset only; no published package source changes.

③ Boundary flags

• Governed rules-layer surface: objectui AGENTS.md. Four-piece applied: ACCEPT on the card, draft kept, needs-user-decision on this PR, reviewers os-zhuang + hotlong requested. ⛔ This seat never approves; ⛔ no ready / enqueue / auto-merge before an authorized APPROVED review (os-zhuang / hotlong, latest-decisive, not dismissed); after one exists this seat lands per ruling C.
• Clause ② judged, not told: declared yes because a ⛔ mechanism claim is softened; measured, the accepted set is unchanged — the form the sentence forbids is still forbidden, and no gate, contract or public surface moves. Not a widening.
• Region-parallel on the same governed file: objectui#9198's PR (this seat, :383–:410) and PR docs(agents): record the merge_group leg's SECOND refusal predicate (the contract-review carrier) #9466 (another seat, :519) hold disjoint hunks; whichever lands second merges origin/main first and the seat re-issues its record on the merged head. mergeable_state read behind at review time (main advanced), which is that clause, not a conflict.
• Observed, not filed: three test-file comments still state the absent mechanism (named in the PR's ## Acceptance notes); a fourth (vite-maplibre-worker.test.ts) is correct and out of scope. Successor: the files' next author; no card.

Implemented-by: claude/issue-9191-agents-new-url-sentence
Reviewed-by: session_01DAcomhvR9kKizeYgg89Vo8

VERDICT: PASS

---

Generated by Claude Code

[claude]

Contract review

Served-tier: 15/15 CONTRACT_REVIEW_TIER
Head-sha: bef04584498e1d42b5b214e98e0271536b7d857a

In-seat record on a GOVERNED surface (objectui AGENTS.md, rules tier — the record is one of the four pieces; the landing waits for an authorized approval under ruling C), domain:skills execution PM seat, 2026-09-14T10:28Z. Spelled from node scripts/pm/check-clause2-carriers.mjs --template on objectstack origin/main. Re-issued: the first record (5662529777) carried a whole-session stamp control that was not total (11 turns served off-tier during a fallback window on 2026-09-13, none since); the stamp above is the reviewing round's own count, every turn since the dev report at 2026-09-14T10:22Z at tier. Judgments unchanged. Re-issued a second time (5662548425 spelled the stamp control with a 「then」 between the count and the constant, which the reader treats as no token): the line now carries the template's live shape — the count, then the constant's name, nothing between. The dev's os-dev-report is comment 5662497171 on objectui#9191; the judgments below are the seat's own reading of the diff and of the tree.

① Derived judgments

• One hunk, inside the claim. AGENTS.md +6/−4 at :233–:242 (head) within the region claimed at :232–:239 on 541ce4e0; .changeset/issue-9191-new-url-import-meta-url-measured.md with empty frontmatter (no release — the file declares why). No test file, script or config moved; the sibling regions (objectui#9198's :383–:410, PR docs(agents): record the merge_group leg's SECOND refusal predicate (the contract-review carrier) #9466's :519) are untouched.
• The prescription survives the edit. 「⛔ 不引入第三种」 stays and the two-argument new URL(相对路径, import.meta.url) form is still 「别写」; what changes is its stated reason — from a Vite rewrite that the dev could not reproduce in any declared project or cwd to spelling uniqueness. The dev's fixture table (five rows, node and happy-dom, repo-root and package-directory cwds, all file: URLs with successful reads) and the seven-file population run (78 tests green at the root, six per-package runs green) are the readings the new sentence rests on; the seat did not re-run the fixture and adopts those rows as the dev's measurement, named as such.
• The corrected ⚠️ line states the tree. scripts/check-test-path-roots.mjs exists on objectui origin/main and is wired in .github/workflows/lint.yml and package.json (seat's own ls-tree and grep); its header at :121 reads 「This gate does NOT flag the two-argument new URL(rel, import.meta.url)」, which is exactly the asymmetry the new line names.
• Population control. The seat's simpler grep for relative two-argument sites on 541ce4e0 reads 26 raw hits; the dev's stricter predicate reads 31 raw / 27 live / 23 in test files with its own two controls (nonsense token 0; bare import.meta.url 741). Same order, consistent direction (the population grew since the card's 27).
• Register and mechanics. The hunk keeps the file's ASCII punctuation (the dev reverted a fullwidth first draft and proved the revert); the PR body carries no angle-bracket placeholder and no model identifier; Clause-②: yes stands on its own line; the changeset rule for a non-package change is followed in PR docs(agents): cite by content, not by line address (#8875 clause 1) #9301's shape.
• Gates as reported. check-governed-queue-guard --test AGENTS.md exit 3 (the governed classification, quoted); self-test 185 cases; check:control-bytes, docs:check-links, check:new-line-citations, check:test-path-roots, the changeset trio and the four instrument suites (212 tests) all exit 0; lint/type-check declared narrowed with three-piece evidence (markdown-only diff) and left to CI. CI on the head at the seat's read: 24 success, 3 skipped, 7 in progress — the landing step re-reads it green before anything moves.

② Semver level

None. AGENTS.md and a no-release changeset only; no published package source changes.

③ Boundary flags

• Governed rules-layer surface: objectui AGENTS.md. Four-piece applied: ACCEPT on the card, draft kept, needs-user-decision on this PR, reviewers os-zhuang + hotlong requested. ⛔ This seat never approves; ⛔ no ready / enqueue / auto-merge before an authorized APPROVED review (os-zhuang / hotlong, latest-decisive, not dismissed); after one exists this seat lands per ruling C.
• Clause ② judged, not told: declared yes because a ⛔ mechanism claim is softened; measured, the accepted set is unchanged — the form the sentence forbids is still forbidden, and no gate, contract or public surface moves. Not a widening.
• Region-parallel on the same governed file: objectui#9198's PR (this seat, :383–:410) and PR docs(agents): record the merge_group leg's SECOND refusal predicate (the contract-review carrier) #9466 (another seat, :519) hold disjoint hunks; whichever lands second merges origin/main first and the seat re-issues its record on the merged head. mergeable_state read behind at review time (main advanced), which is that clause, not a conflict.
• Observed, not filed: three test-file comments still state the absent mechanism (named in the PR's ## Acceptance notes); a fourth (vite-maplibre-worker.test.ts) is correct and out of scope. Successor: the files' next author; no card.

Implemented-by: claude/issue-9191-agents-new-url-sentence
Reviewed-by: session_01DAcomhvR9kKizeYgg89Vo8

VERDICT: PASS

---

Generated by Claude Code

[claude]

Provenance for the gate clear on this PR and on objectui#9191 (needs:contract-review hung and cleared by this seat at 2026-09-14T10:26Z): the review of record is comment 5662557100 (## Contract review, head bef04584498e1d42b5b214e98e0271536b7d857a, **VERDICT: PASS**, Reviewed-by: session_01DAcomhvR9kKizeYgg89Vo8); it supersedes 5662529777 and 5662548425 on the same head, whose Served-tier: spellings the checker could not read. 2026-09-14T10:28Z.

---

Generated by Claude Code

[claude]

Landing under ruling C — domain:skills seat, 2026-09-14T10:48Z. An authorized APPROVED review exists: 5196693680 by os-zhuang (GOVERNED_APPROVERS), submitted 2026-09-14T10:46Z on the reviewed head bef04584498e1d42b5b214e98e0271536b7d857a, not dismissed. The review of record naming that head is 5662557100 (**VERDICT: PASS**); check-clause2-carriers --pair 9486 re-read exit 0 at 2026-09-14T10:47Z; CI on the head 32 success / 3 skipped. The approver flipped the PR ready and enqueued it in the same minute (queue ref gh-readonly-queue/main/pr-9486-6d5db7b present at the seat's read), so the seat leaves the queue entry as set, clears needs-user-decision in this act, and reads the landing by content on origin/main. objectui PR #9489 holds a disjoint region of the same file and is landing in parallel; the queue rebuilds each on the other's result, and a conflict — none reported, mergeable_state: clean — would be the dev's to merge.

---

Generated by Claude Code