[objectstack-fleet]

Requested by Steve · project thread

Fixes #26759

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:skills/objectstack-i18n/SKILL.md 的分组表补上两把 18.2 已接受、运行时已下发但技能没教的键:fields.{field_name} 行加 inlineTitle(主从字段在父表单上的内嵌子网格标题,键挂在拥有该字段的子对象下;未翻译时保留作者原文,不回退到子对象复数标签);reports 那句加 reports.{report}.blocks.{block_name}.{label,description}(联合报表各区块标题,按区块必填的 name 取键)。表后各给一行示例,照搬 app-crm / app-showcase 已发货的 zh-CN 包写法。手写作者页 content/docs/ui/translations.mdx 同补两处。eval 文件加第 8 题覆盖两把键;其 token 棘轮行先削重复再按实测钉在 1642;@objectstack/skills 补丁 changeset。
• 为什么改:PR feat(spec): a joined report's block headings and a master-detail field's inline grid heading are translatable #26757 把这两把键加进了协议,卡 skills/objectstack-i18n: the group table misses fields.{field}.inlineTitle and reports.{report}.blocks.{block}.{label,description} (follow-up owed by PR #26757) #26759 是它欠下的 Skills-docs: 后续——AI 作者照技能写翻译包时,这两处标题会一直停在作者语言。
• 风险与回滚:只改技能文本、一页文档、一个 eval JSON、一行棘轮上限和一个 changeset;不动 schema、运行时或任何键。回滚即 revert 本 PR。
• 先看:skills/objectstack-i18n/SKILL.md(L162 字段行、L175–177 reports 句、L182–188 示例)、skills/objectstack-i18n/evals/bundles-locales-coverage.json(第 6 题删重复、第 8 题新增)、scripts/check-skills-token-ratchet.mjs(棘轮行 1451 → 1642,算式在注释里)。
• 两分钟验证:node scripts/check-skills-token-ratchet.mjs,预期 objectstack-i18n/SKILL.md ✓ 余量 877、bundles-locales-coverage.json ✓ 余量 0,总行 81 authored bundle file(s) within their ceilings。
• 需要你做的:批准本 PR(skills/** 受管面,Tier H)。

What changes

• skills/objectstack-i18n/SKILL.md (the group table a bundle author picks keys from):
  • the fields.{field_name} row names inlineTitle — the heading of the inline child grid a master_detail field draws on its parent's form; the field lives on the child object, so the key is the child's; untranslated it keeps its authored text and never falls back to the child's plural label (wording read off FieldTranslationSchema's docblock, packages/spec/src/system/translation.zod.ts L133–147 on origin/main 0ea90c7);
  • the sentence that said reports, positions and permissions hold label + description by name now continues with reports.{report}.blocks.{block_name}.{label,description}, keyed by the block's required name (JoinedReportBlockSchema.name, packages/spec/src/ui/report.zod.ts; the translation slot at translation.zod.ts L1199);
  • one example row each, after the os i18n extract --no-objects-only sentence, spelled exactly as the shipped bundles spell them for zh-CN: objects.crm_opportunity_line_item.fields.opportunity.inlineTitle (examples/app-crm/src/translations/crm.translation.ts L99) and reports.showcase_task_overview.blocks.open_block.label (examples/app-showcase/src/system/translations/index.ts L1362). The skill's own example bundle mirrors examples/app-todo, which has neither a master-detail field nor a joined report, so the rows sit in a two-line fragment fence instead of being invented into the todo bundle.
• content/docs/ui/translations.mdx (the card's hand-written-twin item, named in the claim's file surface): a row for the inline child-grid heading under the field row, and the report row gains the reports.NAME.blocks.BLOCK.label / description spelling. Not governed; it rides this PR because the claim's surface names it and the card thread has one PR.
• skills/objectstack-i18n/evals/bundles-locales-coverage.json: case 8 asks for the Products grid heading on the crm_opportunity form and the Open Tasks block of the joined showcase_task_overview report, and expects the two keys above; must_contain inlineTitle / crm_opportunity_line_item / blocks / open_block, must_not_contain fields.opportunity_line_item (the key on the parent) and components.open_block (a page-component spelling). Paid down first: case 6's os i18n extract --no-objects-only sentence and its --no-objects-only assertion duplicated case 5's pin of the same flag and were cut (77 bytes). Every name in the prompt and assertions is a shipped example's or the Zod source's; none is invented.
• scripts/check-skills-token-ratchet.mjs: the eval row 1451 → 1642, pinned AT the measurement with the arithmetic and the standing ruling quoted verbatim beside the sibling row: 「A:先削重复,余额按实测提高」. The SKILL.md row does not move (5456 of 6333).
• Changeset .changeset/26759-i18n-skill-inline-title-report-blocks.md: @objectstack/skills patch, both declaration lines at line start.
• packages/lint/src/validate-translation-references.ts (round 2, claim amendment 6449881440 — the card's "Stale header line" item): the Reports header comment now lists blocks.<block_name>.label / .description, keyed by the block's name, among a report's leaf keys. Comment only; the validator reads the Zod shape.

Skills readings (skills/** touched; before = origin/main 0ea90c7, after = head; tokens = the ratchet's ceil(bytes/4), stamp line excluded):

Reading	Before	After
skills/objectstack-i18n/SKILL.md lines / tokens	512 / 5282 (ceiling 6333, headroom 1051)	520 / 5456 (ceiling 6333, headroom 877)
skills/objectstack-i18n/evals/bundles-locales-coverage.json lines / tokens	75 / 1451 (ceiling 1451, headroom 0)	85 / 1642 (ceiling 1642, headroom 0)
all SKILL.md lines / tokens (whole package)	4603 / 56079	4611 / 56253

Why

An AI author who writes a translation bundle from this skill picks keys off the group table; the two keys PR #26757 added were missing from it, so a localized app kept a master-detail grid's heading and a joined report's block headings in English even though the runtime serves both translated (the business case on the card).

Tests

• node scripts/check-skills-token-ratchet.mjs at head: 81 authored bundle file(s) within their ceilings; 10 generator-owned file(s) measured, not ratcheted, exit 0; objectstack-i18n/SKILL.md ✓ (5456 / 6333), bundles-locales-coverage.json ✓ (1642 / 1642, headroom 0).
• node scripts/check-skills-token-ratchet.mjs --self-test: 79 cases pass, exit 0.
• The eval file parses, and no case has a must_contain string in its must_not_contain.
• No test file edited; no forced test edit.

Acceptance notes

• The eval is not run by any gate (the repo has no eval runner; check:skill-docs / check:skill-refs / check:skill-examples read SKILL.md, references/ and the marked code fences). The new example fence is a fragment and carries no os:check marker on purpose.

Verification

• node scripts/pm/dispatch-gates.mjs --tier --repo objectstack-ai/objectstack: MANDATORY — skills/**, clause ①, no one-line exit; changed lines 51 vs the 3000 threshold.
• node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at head 4f490d8e (the final commit) derived 87 commands. All 87 ran in the worktree after that commit, each exit code captured before any pipe: 87 of 87 exit 0 (70 source-reading commands in one serial runner, the 17 dist-reading ones in a second after the build) — among them node scripts/check-skills-token-ratchet.mjs and its --self-test, pnpm --filter @objectstack/spec run check:skill-docs / check:skill-examples / check:docs, pnpm check:pm-governed-merges, pnpm check:pm-dispatch-gates, pnpm check:published-files, node scripts/check-empty-changeset.mjs --base origin/main, node scripts/check-adr-0087-registration.mjs --base origin/main, node scripts/check-changeset-no-major.mjs --base origin/main, pnpm check:skill-identifier-liveness, pnpm check:skill-frame-sync, pnpm check:skill-compatibility, pnpm check:doc-authoring, pnpm check:docs-spec-enumerations, pnpm check:ratchet-remedy-authority, pnpm check:nul-bytes.
• dispatch-gates --ran: 87 derived famil(ies) accounted for — 87 run, 0 NOT-MEASURED (a DERIVED zero — all 87 recorded an exit code and none of them is 3), exit 0; no UNRUN family named on stderr.
• Build closure before the dist-reading gates: turbo run build --filter='./packages/**' --concurrency=2 under scripts/pm/os-verify-lock.sh, VERDICT command-exit 0 (held 8m20s).
• The edited gate script's own test suite is its --self-test (79 cases); no package source is touched, so no package test / typecheck is owed; pnpm lint is CI's run.

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic. (No schema touched.)
• 2. Every name I added uses the casing for its kind.
• 3. No formatter ran; pnpm lint passes (the repo-wide run is CI's; no source file edited).
• 4. One issue, one branch off an up-to-date main, one pull request.
• 5. A change that publishes carries a changeset, and both declaration lines are filled in here and in the changeset.
• 6. packages/spec touched: n/a, not touched.
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English.
• The Before Submitting list in CONTRIBUTING.md is done.

🤖 Generated with Claude Code

https://claude.ai/code/session_01ADHkmPqouEb6yELnSqBwRY

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

⚠️ 1 changed file(s) yielded no anchor (packages/lint/src/validate-translation-references.ts), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files. Nothing else in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 1 changed package(s)).

What this run could not see

• 1 changed file(s) yielded no anchor (packages/lint/src/validate-translation-references.ts) — pages documenting those are invisible to this run
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 4 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 8eef9a8db402b6d46b15cc2375f8acddbb0d7b7e → packageMentionDocs.

[objectstack-fleet]

席位核对

ACCEPT · domain:skills#1 · session_01ADHkmSrRBdu9MdC6fc33Uu · head 73e5a8f1 · 2026-10-27T18:36Z

1. 形态 — Tier H (skills/**): skills/objectstack-i18n/SKILL.md teaches fields.{field}.inlineTitle (keyed on the child object that owns the master-detail field; untranslated keeps its authored text) and reports.{report}.blocks.{block_name}.{label,description}, with one example row each from the shipped bundles; evals/bundles-locales-coverage.json case 8 for both keys; the same two rows on content/docs/ui/translations.mdx (the card's hand-written-twin item); the validate-translation-references.ts Reports header comment (round 2, claim amendment 6449881440); eval ratchet row 1451 → 1642 at the measurement; patch changeset for @objectstack/skills.
2. 前提实测 — PR feat(spec): a joined report's block headings and a master-detail field's inline grid heading are translatable #26757 merged; the examples are real (examples/app-crm/src/translations/crm.translation.ts:99, examples/app-showcase/src/system/translations/index.ts:1362); case 6's cut --no-objects-only pin stays pinned by case 5 (must_contain L50), so no fact left the file.
3. 门禁 check-runs — dev ran the 108 derived commands on the head (87 + 21 from the lint path), all exit 0 (--ran 108, 0 NOT-MEASURED); @objectstack/lint typecheck and 7430/7430 tests green; build closure VERDICT 0. CI on the head is read again before landing.
4. 接受的偏差 — the example rows sit in a fragment fence after the table rather than in the todo-mirroring bundle (open question A). The seat patched the PR body (L163 now in this PR; the "not filed" note removed). Payment of the eval row follows 「A:先削重复,余额按实测提高」: 77 B trimmed first, then raised by the measured remainder.
5. 范围外去向 — none. Contract review at CONTRACT_REVIEW_TIER: requested from the director seat.
6. 落地条款 — governed, Tier H: stays draft until a GOVERNED_APPROVERS approval; then it lands through the queue. skills(objectstack-ui, objectstack-i18n): the actions rule and the i18n skill do not teach an action's disabledReason (follow-up owed by #26619) #26790 (same SKILL.md) is dispatched after it lands.

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 73e5a8f1d60b4332b2da22a580cbffaff3f6aeeb
Local-runs: none
Scope: skills/objectstack-i18n/SKILL.md + skills/objectstack-i18n/evals/bundles-locales-coverage.json, scripts/check-skills-token-ratchet.mjs, content/docs/ui/translations.mdx, packages/lint/src/validate-translation-references.ts, .changeset/26759-i18n-skill-inline-title-report-blocks.md, packages/spec/src/system/translation.zod.ts, packages/spec/src/ui/report.zod.ts, examples/app-crm/src/translations/crm.translation.ts, examples/app-showcase/src/system/translations/index.ts

Director seat (summon #52, ledger objectstack#12708), summoned by the domain:skills#1 seat. Read-only at the head above: the diff, card #26759 (filing gate ①, the Skills-docs: follow-up owed by #26757), its thread (claim 6449271046, amendment 6449881440, both dev reports), the seat's ACCEPT 6450504495, the eval file as stored at this head and on origin/main 0ea90c7a, the spec docblocks and the two shipped bundles on main, the head's check runs. skills/** is a governed-register row and publishes (the @objectstack/skills bundle) ⇒ the rules face, CONTRACT_REVIEW_TIER; Tier H, lands on a GOVERNED_APPROVERS approval, not on this record. Served at CONTRACT_REVIEW_TIER. The seat records; it does not approve, ready, queue or merge.

① Derived judgments

• Both taught keys are the spec's, with the spec's own rule. FieldTranslationSchema.inlineTitle (translation.zod.ts:134–147 on main): the heading of the inline child grid drawn on the PARENT's form, addressed on the child (objects.<child>.fields.<field>.inlineTitle), overlaid by translateObject, "No fallback to the child object's translated plural label: an authored inlineTitle with no entry here keeps its authored text". The skill row and the docs row say exactly that. reports.<name>.blocks.<block_name>.{label,description} (translation.zod.ts:1199, a strictObject keyed "by the block's name (JoinedReportBlockSchema.name, required)"; report.zod.ts:280). The skill sentence, the docs row and the lint header comment all spell it that way.
• The example rows are the shipped bundles', byte for byte. crm.translation.ts:97–99: crm_opportunity_line_item: { fields: { opportunity: { inlineTitle: '产品明细' } } }, under the comment "the field lives on the line item, so the key is the line item's". app-showcase … translations/index.ts:1361–1363: showcase_task_overview … blocks: { open_block: { label: '未完成任务' } … }. The fragment fence invents nothing; the skill's own todo-mirroring bundle has neither source, so a fence is the right home (dev open question A — agreed).
• Case 8 grades its own ideal answer right. Reading the stored head file: every must_contain (inlineTitle, crm_opportunity_line_item, blocks, open_block) present; neither must_not_contain (fields.opportunity_line_item, the key on the parent; components.open_block, a page-component spelling) present. The refused spellings are the two mistakes an author would actually make.
• The cut is a duplicate, as the seat read. Case 6 loses its os i18n extract --no-objects-only sentence and the --no-objects-only assertion; case 5 on main and at the head still pins the flag in must_contain. 77 bytes (56 from expected_output, 21 from the assertion list), matching the row comment.
• Ratchet arithmetic at the head. Eval file stored 5802 → 6567 bytes (+765 = the new case − 77); ceil(5802/4) = 1451, ceil(6567/4) = 1642, pinned at the measurement, headroom 0, the tap quoted verbatim. SKILL.md 21866 − 20 (stamp) = 21846 ⇒ 5462 of 6333, row unmoved. The head's own ratchet row reads 1642.
• The lint edit is a comment. validate-translation-references.ts:163–165 header only; the validator reads the Zod shape, which already accepts blocks. No behaviour, no export.

② Semver level

@objectstack/skills takes a patch changeset, both declaration lines present at line start and equal to the PR body's. Skill text, an eval and a docs page; patch is right. The lint comment edit publishes nothing new.

③ Boundary flags

• Governed path (skills/**, Tier H): no review yet; draft. This seat's account is not a GOVERNED_APPROVERS member; this record is not an approval and does not lift Tier H.
• Should-fix, in this PR's own hunk (same class as skills(objectstack-automation): eval fixtures 5, 7 and 8 of flows-triggers-approvals.json contradict their own assertions — expected_output spells a must_not_contain string or lacks a must_contain one, so the ideal answer fails its own check #26515): case 6 — the case this PR edits — still spells its own must_not_contain string inside its ideal answer: expected_output says "never permissionSets" while must_not_contain lists permissionSets, so the ideal answer is graded wrong. Pre-existing on main, but the PR has the case open and the dev's "no must_contain string sits in a must_not_contain" check does not cover expected_output against must_not_contain. A same-byte rewording (for example "not the camel-case alias") keeps the file at its 1642 pin; a longer one is paid under the same ruling. Fix here rather than carry it to a third card.
• Check runs on 73e5a8f1 at this record: 42 success, 9 expected skips, nothing in progress, nothing red — including Lint · repo & tooling gates (the token ratchet), Check Changeset, both Lint · docs & code guards and the @objectstack/lint typecheck/test shards.
• origin/main has not moved on any of the six files since the branch's base; no merge-in owed.
• No gate runs the eval; the green proves parse, ratchet, changeset and the doc gates, not that an agent passes case 8 (named, as on docs(skills): objectstack-ui gains an eval for a report's dateRange and exportOptions #26774 and docs(skills): objectstack-formula evals exercise label(record.kind), and refuse it in a predicate #26810).
• Note, not a finding: the spec docblock says the inline grid is drawn by "a master_detail / lookup field with inlineEdit"; the skill says master_detail. field.form.ts:294 shows the key only when type == 'master_detail' and field.zod.ts:1675 describes it as the master-detail grid title, so the skill follows the authoring surface. Fine as written.
• Dev deviations: the examples in a fragment fence (judged above); the docs page rides this Tier H PR (the card allowed either; nothing on it is blocked by waiting); the seat patched the body after the round-2 amendment so the L163 bullet now matches the diff.

④ Design judgment

• skills/objectstack-i18n/SKILL.md against the spec: the inlineTitle row teaches the one non-obvious fact (the key is the child's, and silence keeps authored text) in the row where an author picks field sub-keys, and the report sentence puts blocks.{block_name} beside the label + description it extends, keyed the way pages.{page}.components.{id} already taught. Case 8's prompt gives real object, field, report and block names from shipped bundles, so a candidate cannot pass by paraphrase, and its refusals are the two plausible wrong addresses. The docs twin carries the same two rows with the same rule. Holds; nothing to break.

Implemented-by: claude/project-thread-jdbl6z
Reviewed-by: session_01ADHkmQr732toX135tyo5zs

VERDICT: PASS

---

Generated by Claude Code

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 57406c6ff6f1e8037aa9e2846f483637d74a8afb
Local-runs: none
Scope: skills/objectstack-i18n/SKILL.md + skills/objectstack-i18n/evals/bundles-locales-coverage.json, scripts/check-skills-token-ratchet.mjs, content/docs/ui/translations.mdx, packages/lint/src/validate-translation-references.ts, .changeset/26759-i18n-skill-inline-title-report-blocks.md, packages/spec/src/system/translation.zod.ts, packages/spec/src/ui/report.zod.ts, examples/app-crm/src/translations/crm.translation.ts, examples/app-showcase/src/system/translations/index.ts

Director seat (summon #52), re-stamping record 6450590107 (head 73e5a8f1, PASS) for round 3. Read-only: the one-file diff 73e5a8f1..57406c6f, the eval file as stored at this head, the ratchet row, the head's check runs. Served at CONTRACT_REVIEW_TIER. The seat records; it does not approve, ready, queue or merge.

① Derived judgments

• The round-3 diff is the should-fix and nothing else. One line of bundles-locales-coverage.json: case 6's ideal answer now says "not the camelCase name" where it said "never permissionSets". Same byte length — the file is still 6567 bytes, ceil(6567/4) = 1642, the ratchet row unchanged at 1642, headroom 0.
• Every case now grades its own ideal answer right. Reading the stored head file: cases 1–8 each contain every must_contain and no must_not_contain string. The skills(objectstack-automation): eval fixtures 5, 7 and 8 of flows-triggers-approvals.json contradict their own assertions — expected_output spells a must_not_contain string or lacks a must_contain one, so the ideal answer fails its own check #26515-class contradiction named in 6450590107 ③ is gone; nothing the earlier record judged on SKILL.md, the docs twin, the lint comment or the changeset has moved.

② Semver level

Unchanged: @objectstack/skills patch, both declaration lines as before.

③ Boundary flags

• Governed path (skills/**, Tier H): draft, no review yet; this record is not an approval.
• Check runs on 57406c6f at this record: 18 success, 7 skips, 19 still in_progress (Test Core 1–8/8, Build & E2E UI, Type Check UI, Build Docs, Build Core among them). The one-line JSON edit touches no gate input but the ratchet, whose job (Lint · repo & tooling gates) is among the ones to land; a red there re-opens this record.
• Should-fix from 6450590107: resolved at this head. No new flags.

④ Design judgment

Unchanged from 6450590107 ④; the rewording keeps case 6's teaching (the stack's permissions[] spelling, not the alias) without spelling the refused token.

Implemented-by: claude/project-thread-jdbl6z
Reviewed-by: session_01ADHkmQr732toX135tyo5zs

VERDICT: PASS

---

Generated by Claude Code