`i18n-extract.ts` describes the `_views` translation convention as one client-side reader set — the three tails have three different readers, and objectui#7219 retired only `viewDescription`

[os-litant]

Filed by the triage seat while closing #15180, whose premise this replaces.

What #15180 got right, inverted

#15180 concluded that objects.<object>._views.<view>.description was 「declared, accepted, read by nothing」. The key is read — server-side, and the resolved value reaches the screen. What is actually stale is the prose describing the convention, which names a client-side symbol that no longer exists.

⇒ The trap inverted: the declaration is true, the documentation about it is false, and a reader who checks the documentation concludes the declaration is dead. That is what cost #15180 a dispatch round.

The site

packages/cli/src/utils/i18n-extract.ts:363, verbatim on origin/main:

* reads (\viewLabel` / `viewDescription` / `viewEmptyState` in`

naming @object-ui/i18n symbols that objectui#7219 retired — it removed useObjectLabel().viewDescription(), the client-side second resolution.

⚠️ A second candidate site, ⛔ not yet judged

viewDescription appears in exactly three places on origin/main (measured; the count is small enough to enumerate rather than sample):

site	verdict
packages/cli/src/utils/i18n-extract.ts (1)	this card — stale prose
packages/console/CHANGELOG.md (2)	⛔ historical record; leave it
packages/lint/src/validate-translation-references.ts (1)	⚠️ NOT JUDGED — read it before deciding. If it is a live reference the lint rule depends on, ⛔ leave it; if it is the same stale naming, fix it here.

⛔ Do not assume the lint site is stale because the CLI site is. Read it.

What the correct prose must say

The convention is live and resolved by the server: resolveViewDescription (packages/spec/src/system/i18n-resolver.ts:424), called at :908, registered in METADATA_DOCUMENT_TRANSLATORS (:987) from which TRANSLATABLE_METADATA_TYPES is derived (:1009) and read by @objectstack/rest. ⇒ Describe the server-side resolver, ⛔ not a retired browser-side helper.

Lane

packages/cli ⇒ domain:cli.

⚠️ Same file as #17511 (walkScreenFlows is region-blind, domain:cli, p2, graded this shift). ⇒ Different defect, one file face — ⛔ do not dispatch both at once, and whoever takes either should fix both in that worktree if the second is a small consequence, declaring it in the PR.

Refs: #15180 · #17511 · objectui#7219

---

Generated by Claude Code

[claude]

Claim: domain:cli execution seat (#6024), round R73.
Session: session_01TSf4DV7ziu4V5j73e46b7c · account os-sales · claimed 2026-09-11T18:29Z (stamp from date -u in this posting call).
Branch: claude/issue-17546-i18n-extract-view-convention-prose

Clause-②: no

Judged here: the deliverable is a docblock inside packages/cli. No exported symbol, no accepted-input set, no published behaviour moves.

Declared file surface (what a cross-domain in-flight check reads, and the dev's fence):

• packages/cli/src/utils/i18n-extract.ts — the docblock above pushViewEntries.
• ⚠️ packages/lint/src/validate-translation-references.ts — ⛔ NOT authorised. See the fence below.

Premise re-checked on origin/main before claiming. LIVE:

packages/cli/src/utils/i18n-extract.ts   the docblock above `pushViewEntries` still reads, verbatim:
    "the convention the runtime resolver reads (`viewLabel` / `viewDescription` /
     `viewEmptyState` in @object-ui/i18n)"
viewDescription across packages/ + content/ — exactly THREE sites, matching the card's enumeration:
    packages/cli/src/utils/i18n-extract.ts               1   <- this card
    packages/console/CHANGELOG.md                        2   <- historical, ⛔ leave it
    packages/lint/src/validate-translation-references.ts 1   <- :381, a COMMENT, still NOT JUDGED
the replacement content is real:
    packages/spec/src/system/i18n-resolver.ts:424   export function resolveViewDescription(
                                            :908   called here
#17511 landed as d07fc178 (PR #17644) touching this same file ⇒ the card's own
    「⛔ do not dispatch both at once」 fence is DISCHARGED

⭐ The inversion is the point, and it is why this card replaces #15180's premise. #15180 concluded the key was 「declared, accepted, read by nothing」. The key IS read — server-side, and the value reaches the screen. What is stale is the prose, which names a client-side symbol objectui#7219 retired. ⇒ the declaration is true, the documentation about it is false, and a reader who trusts the documentation concludes the declaration is dead. That inversion cost #15180 a whole dispatch round. Your replacement prose must describe the server-side resolver, ⛔ never a retired browser-side helper.

⚠️ The second site: read it, judge it, and ⛔ do not touch it

packages/lint/src/validate-translation-references.ts:381 is a comment reading "viewLabel / viewDescription resolve view tabs, and the i18n walker…" — the same naming style. The card says in as many words: ⛔ do not assume the lint site is stale because the CLI site is. Read it.

⇒ You owe a judgement on it in the PR body, and ⛔ nothing more. Even if it is stale, ⛔ do not edit it: packages/lint's spec-facing surface is routed to domain:spec by the domain-anchoring exception, so a fix there is another seat's to land. Report it and this seat files or routes it.

⛔ THE RED LINE THIS SEAT OMITTED ONCE TODAY, NOW WRITTEN INTO EVERY CLAIM

⛔ Touching packages/spec in ANY way is a stop-and-report, not a judgement call. 「凡触 packages/spec 一律转 domain:spec 座位,不论谁需要它」 — the line sits in SKILL.md:231, references/core-rules.md:62, SKILL.md:287, references/lanes/cli.md:12 (this lane's own charter), references/lanes/spec.md:12, and references/dispatch-runbook.md:158, which adds that the folding rule does not exempt it: 「唯一所有者规则更硬」.

⚠️ Earlier today this seat dispatched #17178 with an order that said only "whichever single package the export lands in". The dev landed a correct, well-argued export in packages/spec — and the whole PR had to be held and re-routed, because this seat may not land a packages/spec change or perform its clause-② review. ⛔ That was the order's fault, not the dev's. ⇒ reading packages/spec is fine and often necessary; writing a single byte into it stops the dispatch.

Hot-file serial queue: CLEAR. #17454 holds none of this surface, and #17511 — the one card the #17546 body names as a same-file collision — landed this round (d07fc178), which discharges that fence. ⚠️ Verify that for yourself on origin/main before you start; if any other open PR now touches i18n-extract.ts, stop and report.

H17 on-hold trigger-file index: no intersection.

Batch independence. Dispatched alongside #17169 (packages/cli/test/build-multi-package-artifact.e2e.test.ts). Zero file-surface intersection.

---

Generated by Claude Code

[claude]

⚠️ Dispatched dev died with the container — claim RETAINED, work preserved, ⛔ nothing verified

domain:cli execution PM seat (#6024), session session_01TSf4DV7ziu4V5j73e46b7c, 2026-09-11T19:01Z.

The container running this session was restarted, which killed the dispatched os-dev mid-flight. ⚠️ Per the standing reading, 子代理消失是正常死法 — a subagent vanishing is an ordinary death, ⛔ not a maintainer abort, which is only established by an explicit signal (原话, or the host reporting stopped by the user). Neither exists here. ⇒ this is a dead-claim reclamation, and the claim stays with this seat.

What survived on the remote — measured, not assumed

git ls-remote origin refs/heads/claude/issue-17546-i18n-extract-view-convention-prose
  -> 72b23c092694                         # the branch IS pushed
  CONTROL: git ls-remote origin refs/heads/main -> resolves (query shape live)
git log --oneline origin/main..<branch>
  -> 72b23c09  chore(changeset): patch @objectstack/cli for the `_views` reader-per-tail docblock
  -> 75b97af6  docs(cli): name the three readers of `_views.<view>.*` separately, per tail
git diff --stat origin/main...<branch>
  -> .changeset/17546-i18n-extract-view-reader-per-tail.md | 52 +++
     packages/cli/src/utils/i18n-extract.ts                | 32 ++-
     2 files changed, 80 insertions(+), 4 deletions(-)

⚠️ Two things the successor must judge rather than inherit, both visible from the commit subjects alone and neither confirmed by any report:

1. ⚠️ It added a patch changeset where the dispatch expected skip-changeset. The order said a comment-only change normally takes skip-changeset and required the choice to be stated with a reason. The reason was never reported. ⇒ the successor either justifies the changeset in the PR body or changes it; ⛔ it is not adopted by default.
2. ⚠️ The subject says "name the three readers … separately, per tail", which is a more differentiated claim than the card's "describe the server-side resolver instead of the retired client-side helper". That may be a better answer or an over-reach; ⛔ it is NOT established either way, and the successor must derive the reader set itself.

⛔ Still owed and, on this record, never delivered: the one-paragraph judgement on packages/lint/src/validate-translation-references.ts:381. That file is ⛔ not authorised for editing — packages/lint's spec-facing surface routes to domain:spec by the domain-anchoring exception — so the successor reads it, judges it, reports it, and this seat files or routes whatever it finds.

⛔ What is NOT established, and this is the load-bearing half

⛔ No report arrived. 报告缺席永不读作成功 — a missing report is never read as success, and a pushed branch is not a substitute for one. Specifically unestablished:

• ⛔ No PR exists, so no CI has ever run on this work — not one check.
• ⛔ No gate union was reported, so the lane's 50-plus derived commands plus pnpm lint are unrun as far as any record goes.
• ⛔ No ablation was reported.
• ⛔ Nothing in the fence, the premise, or the controls was confirmed back.

⇒ what survives is a draft with zero verification on record. ⭐ It is worth preserving rather than discarding: the branch is real remote output and re-deriving it would waste good work. But it is explicitly ⛔ not a deliverable, and ⛔ nothing about it is to be inherited as measured.

Disposition

• pm:dispatched and the assignee stay — the card has not lost its owner, and ⛔ a silent assignee strip is never how a claim is released.
• The claim comment's Branch: line already names this branch, so the half-state patrol can still see it.
• A fresh dev is being dispatched now, fenced identically, and told to start from this branch rather than from main — re-deriving the premise itself, re-running every gate, and reporting what it measures. ⛔ It inherits no reading from the dead run.

⚠️ A reader arriving before that dispatch lands sees pm:dispatched + an assignee + a pushed branch + no PR. That combination is a half-state, which is why this comment exists at all rather than the state being left to speak for itself.

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 17546,
  "status": "done",
  "branch": "claude/issue-17546-i18n-extract-view-convention-prose",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/17721",
  "premise_still_valid": false,
  "summary": "Recovery pass on the dead R73 dev's branch. The card's SITE is real and was fixed; the card's TITLE is partly false and that is the headline finding. It calls viewLabel / viewDescription / viewEmptyState 'symbols objectui#7219 retired' — measured on objectui's origin/main (the repo IS reachable from this container, so this half is MEASURED, not unmeasured), only viewDescription was retired: useObjectLabel().viewLabel is live at packages/i18n/src/useObjectLabel.ts:481 and useObjectLabel().viewEmptyState at :512, and objectui's own comment at :501 states the shared viewSuffixes helper survived with only the 'description' tail dropped. So the deliverable was not 'swap a retired name for the server resolver' but 'name three different readers correctly': label is read on BOTH sides (resolveViewLabel via translateView at the REST metadata boundary, plus useObjectLabel().viewLabel), description SERVER-SIDE ONLY (resolveViewDescription; client member removed), emptyState CLIENT-SIDE ONLY (zero empty-state reader in the spec resolver, with two same-file controls firing). The draft's reader topology was therefore CONFIRMED and KEPT; one omission was CORRECTED — objectui does not merely lack a helper, it PINS this key as 'an inert catalog entry' in a whole describe block, and no bundle in this repo authors a description leaf (nine shipped en.objects.generated.ts: 104 label leaves, 9 emptyState, ZERO description). Left unstated, the next reader greps objectui, reads 'inert', and deletes the emission — this card's defect class with the polarity flipped. The docblock now names both facts, scopes them to the browser path objectui#7219 removed, and cites the reach claim (ObjectView.tsx:2401 relays, ListView.tsx:3404 renders via pickLocalized) instead of asserting it. The draft's patch changeset was REVERSED to skip-changeset. packages/spec was read, never written.",
  "tests": "All exit codes captured BEFORE any pipe (cmd > log 2>&1; EXIT=$?), measured at final commit 5fce933b. GATE UNION: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 48 families (script derived its own change set; no hand-written file list); reconciled via --ran with a per-command exit code: '48 derived famil(ies) accounted for — 46 run, 2 NOT-MEASURED (2 DERIVED from a recorded exit 3)', 0 UNRUN. Re-derived after a fresh git fetch origin main: byte-identical to the union run. 46/48 exit 0. TWO NOT MEASURED, not red, verdict lines read rather than the number: pnpm check:dual-build-cjs-loads :: exit 3 'PREREQUISITE NOT MET — this gate reads built output, and some package has no dist/ ... this is NOT a pass: nothing was measured' (12 packages outside this closure unbuilt; its own --self-test passed, 93 cases); pnpm check:i18n-coverage :: exit 3 'Nothing was compared: 12 config(s) did lint, but a partial round cannot judge the ratchet ... says NOTHING about whether any declared label went untranslated' (examples/app-showcase cannot load its config without a repo-wide build). Both need a whole-repo pnpm build = CI's Build Core. The two i18n gates not needing it, pnpm check:i18n and pnpm check:i18n-walk-parity, are GREEN and are the pair that would move if the emitted key set had changed. LANE BLIND SPOT ADDED: pnpm lint :: exit 0 — FULL repo-wide run (eslint . --no-inline-config), not narrowed, completed in the foreground, so no narrowing had to be argued. BUILD: pnpm --filter '@objectstack/cli^...' build --concurrency=2 :: exit 0, then pnpm --filter @objectstack/cli build :: exit 0 (VERDICT command-exit 0 from os-verify-lock). TYPECHECK: pnpm --filter @objectstack/cli typecheck :: exit 0, including check:test-typecheck OK. TESTS: pnpm --filter @objectstack/cli exec vitest run --project unit --maxWorkers=2 :: exit 0 — 'Test Files 196 passed (196)', 'Tests 2813 passed (2813)'. Only the unit tier is owed locally (diff touches no integration-layer file, no bin/ entry, no spawn helper); integration declared to CI. ABLATION: NONE IS POSSIBLE and this is stated rather than skipped — the diff is one JSDoc block, a mutation to prose moves no gate's verdict, so a red/green pair would be fabricated. ZERO-HIT CONTROLS, each sharing its target's failure mode: (1) 'empty.?state' case-insensitive in i18n-resolver.ts on origin/main = exit 1, zero hits, with two same-file same-shape controls firing — description (89 lines) and camelCase pluralLabel (8 lines); (2) resolveViewEmptyState zero tree-wide under ONE pattern that simultaneously matched resolveViewLabel and resolveViewDescription, so the control is inside the query; (3) zero description leaves under _views across nine bundles, same parser finding 104 label and 9 emptyState leaves in the same blocks; (4) docblock text absent from packages/cli/dist/utils/i18n-extract.d.ts while an exported symbol's JSDoc in the same file is present — pushViewEntries in that .d.ts: zero hits exit 1, control collectExpectedEntries: 3 hits exit 0. WRAPPED-PROSE FAILURE MODE DEMONSTRATED, not assumed: git grep -c -F on the lint file's verbatim quote of i18n-extract.ts exits 1 (zero hits) while the phrase IS present — it wraps mid-phrase behind a ' * ' prefix; a flatten-then-search pass finds it, and a control phrase known present with the same wrapping behaves identically. Every phrase-level claim here was measured that way, never with a bare git grep -F.",
  "mcp_calls": "0 — zero MCP GitHub calls for the whole run. A repo-scoped REST probe (GET /repos/objectstack-ai/objectstack) returned 200, so the card, its two comments, all 11 open PRs' file lists, PR creation and the label write all went through REST; git and the local objectui checkout covered the rest.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: packages/lint/src/validate-translation-references.ts:381 still spells viewDescription and says viewLabel / viewDescription 'resolve view tabs'. JUDGEMENT (this is the judgement the dead run never delivered): LEAVE IT. Its load-bearing clause — that _views.* is a LIST-view convention, so the default form view gets addSections but no _views name — is TRUE and unaffected by objectui#7219, and unlike the CLI site it never says 'in @object-ui/i18n', so it points no reader at a removed member of a named package. viewLabel is live on both sides. Two imprecisions remain and both are cosmetic: viewDescription names no client-side resolver any more (server-side the symbol is resolveViewDescription), and 'resolve view tabs' conflates _views with _tabs, which is resolveTabLabel / translateInterfaceTabs territory. No behaviour, no gate and no reader's conclusion turns on either, so it fails all three filing classes — an imprecision in a comment whose conclusion holds is not a defect. Successor: domain:spec, which owns that file's spec-facing surface; if the seat disagrees, the edit is that seat's. Read, judged, NOT edited.",
    "noted, not filed: i18n-extract.ts:1242 is a THIRD viewLabel site inside the authorised file, and packages/lint quotes it VERBATIM ('form views have no counterpart in the viewLabel / _views.* resolver convention'). Re-measured as ACCURATE, so no repair was owed — and editing it would have dangled the lint file's citation for no gain. Deliberately untouched. Recorded because 'the card named one site' is not 'the file has one site'.",
    "noted, not filed: i18n-extract.ts:451, the pushViewEmptyState docblock naming viewEmptyState in @object-ui/i18n, re-measured as ACCURATE (live at useObjectLabel.ts:512, and client-side-only is exactly right). The draft left it alone and that was correct.",
    "noted, not filed: the card's own title needs correcting if anyone cites it later — it names three retired symbols where only viewDescription was retired. Not filed as a card because the card is being closed by this PR; flagged here so the seat can fix the title or let it close as-is."
  ]
}

---

Generated by Claude Code

[claude]

⭐ The dev falsified this card's TITLE, and this seat verified it independently — retitling before the card closes

domain:cli execution PM seat (#6024), session session_01TSf4DV7ziu4V5j73e46b7c, 2026-09-11T19:51Z. The recovery dev reported premise_still_valid: false with open_questions: []. ⭐ 前提被证伪是好产出 — this is the best outcome the dispatch could have had, and it is the second time today a dispatched dev corrected this seat rather than the reverse.

What the title claims, and what the source says

This card's title calls viewLabel / viewDescription / viewEmptyState "symbols objectui#7219 retired". ⛔ Only one of the three was retired.

⚠️ The dev reported the objectui repo as reachable from this container — contradicting the dispatch order, which told it to record that half as UNFIRED-BECAUSE-UNMEASURED. ⇒ this seat checked that claim rather than taking it, and it is true: /home/user/objectui is a local checkout. Read from objectui's own origin/main (ad80b06), packages/i18n/src/useObjectLabel.ts — the file's presence on that ref confirmed with git ls-tree rather than from a working-tree listing:

viewLabel: (objectName, viewName, fallback) =>
  resolve(viewSuffixes(objectName, viewName, 'label'), fallback),

/* There is deliberately NO `viewDescription` member here … (objectui#7219,
 * maintainer ruling 2026-09-02, option B).
 *
 * `viewSuffixes` is NOT retired with it: `viewLabel` above and
 * `viewEmptyState` below share that helper and keep resolving. Only the
 * `'description'` tail passed to it is gone. */

viewEmptyState: (objectName, viewName, fallback) => { … }

⇒ objectui's own comment says it in as many words. viewLabel and viewEmptyState are live and still resolving; only the 'description' tail is gone. The title is false on two of its three symbols.

Retitled, and why the title rather than a new card

The card closes on PR #17721's merge and its title becomes the durable thing people cite. A closed card carrying a false premise in its title is the same defect class this card is about, one level up — so the title is corrected now rather than left as an archive of the mistake. ⛔ No new card: the dev correctly declined to file one against a card that is closing, and said so.

⚠️ ⛔ The body is not rewritten. 存量不追溯改写 — the body records what the filing seat measured and believed, and edit history is the archive; only the title, which is the citation surface, is corrected.

⭐ The correction the dev made to its own draft, which is the sharper half

The dead run's draft had the reader topology right — label read on both sides, description server-side only, emptyState client-side only. The recovery pass kept that and fixed an omission that matters more than the original defect:

objectui does not merely lack a helper — it PINS this key as "an inert catalog entry" in a whole describe block, and no bundle in this repo authors a description leaf: across nine shipped en.objects.generated.ts, 104 label leaves, 9 emptyState, and zero description.

⇒ left unstated, the next reader greps objectui, reads "inert", and deletes the emission — ⭐ this card's defect class with the polarity flipped. The docblock now names both facts, scopes them to the browser path objectui#7219 removed, and cites the reach claim (ObjectView.tsx:2401 relays, ListView.tsx:3404 renders via pickLocalized) rather than asserting it.

The packages/lint judgement is ACCEPTED as reported — ⛔ no card

The judgement the dead run never delivered: leave it. packages/lint/src/validate-translation-references.ts:381 still spells viewDescription, but its load-bearing clause — that _views.* is a list-view convention, so a default form view gets addSections and no _views name — is true and untouched by objectui#7219; and unlike the CLI site it never says "in @object-ui/i18n", so it points no reader at a removed member of a named package. Two imprecisions remain and both are cosmetic. ⇒ fails all three filing classes: an imprecision in a comment whose conclusion holds is not a defect. ⛔ Read, judged, not edited — exactly the fence.

Also accepted as reported

• The draft's patch changeset was reversed to skip-changeset. ✅ The dispatch required it justified or changed; it was changed.
• packages/spec read, never written, and the PR's file list confirms it: one file, zero packages/spec/ and zero content/docs/releases/ paths.
• ⚠️ No ablation, and the dev said so rather than skipping it: the diff is one JSDoc block, so a prose mutation moves no gate's verdict and a red/green pair would be fabricated. ⭐ Declining to manufacture evidence is the right call and is recorded as such.
• Two further sites inside the authorised file (:1242, :451) were re-measured as accurate and deliberately untouched — and :1242 is quoted verbatim by packages/lint, so editing it would have dangled that citation for no gain. ⭐ "the card named one site" is not "the file has one site" is worth carrying.

Landing follows once CI converges on 5fce933b; the review of record will carry the gate readings.

---

Generated by Claude Code