skills-sweep ⑧: objectstack-i18n (807 lines, 3 files) — behavioral-claim verification, content-class execution-first

[huangyiirene]

Member card of program #13658 (published-skills factual sweep). Flight ⑧ target: skills/objectstack-i18n/** — 3 files, 807 lines at origin/main this hour: SKILL.md (727) · evals/README.md (45) · references/_index.md (35, generator-owned — verify sync, never hand-edit).

Scope: verify every behavioral claim against the implementation per the program method as BOUND by #13658 comments 5474435934 / 5474464789 / 5475032239 + calibrations 5476328676 / 5476875577 / 5477729758 / 5479170359 (content-class first; both error directions; cross-file contradiction + md-vs-generated oracle first; "silently / ignored / dropped" seed; eval rubrics FULL weight; spec-.describe() twins filed as out-of-scope findings). Likely implementation homes: packages/spec/src/system/** (i18n/book schemas), i18n services under packages/services/**, the kernel i18n fallback (packages/core), objectui @object-ui/core label resolution for cross-repo claims (probe-read only from a detached worktree if needed — zero objectui edits).

Constraints: published governed surface — draft PR, human merge; ceilings unchanged; net line budget ≤ 0; NOT MEASURABLE recorded. Clause ② CONTENT limb (批 #12): needs:contract-review to PR AND this card same-stroke.

Part of #13658; closes with the flight's PR (Fixes).

Generated by Claude Code

[huangyiirene]

Claim: PM loop round 1
Session: session_01EnE7G31tqbxN1rqpQmzurT
Branch: claude/issue-13815-skills-sweep-i18n
Worktree: objectstack-issue-13815
Domain: domain:skills
File surface: skills/objectstack-i18n/** in objectstack ONLY (stop on breach); objectui is probe-read territory via a detached worktree if cross-repo claims need it — zero edits there.
Container & model: M (claim verification with executed probes), mode:subagent, model: opus — no path-derived mandate on skills/** per this morning's dispatch-gates.mjs --tier posture.
Clause-②: yes — 批 #12 CONTENT-limb criterion; needs:contract-review to PR AND this card same-stroke; this seat reviews at CONTRACT_REVIEW_TIER at ACCEPT.
Serial constraints cleared: zero open PRs touch skills/objectstack-i18n/** (listings read this hour); sibling flight ⑦ (#13814, skills/objectstack-api/**) dispatches this same wave — disjoint faces, no read-coupling.

---

Generated by Claude Code

[huangyiirene]

os-dev-report

{
 "issue": 13815,
 "status": "done",
 "branch": "claude/issue-13815-skills-sweep-i18n",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/13833",
 "premise_still_valid": true,
 "summary": "Flight 8 swept skills/objectstack-i18n (3 files, 807 lines; face matched the PM's assumption exactly: SKILL.md 727 / evals/README.md 45 / references/_index.md 35). Roughly 200 behavioral claims inventoried; 17 distinct FALSE facts across 20 landing sites corrected (8.5% distinct / 10% sites — ABOVE the 1.5-6% working range, the highest density measured in the programme so far), about 14 recorded NOT MEASURABLE. Net -8 lines, -2 tokens: SKILL.md 6337/6338 and evals/README.md 410/411, both shrink, no ceiling raised. references/_index.md is generator-owned and was NOT hand-edited; check:skill-refs green. The dominant pattern is sharper than the seed the programme handed me: almost every falsehood is an ENUMERATION THAT STOPPED GROWING WHEN THE SCHEMA DID — the object sub-key table missed _tabs (#5377) and _actions.description (#7367), the top-level group list missed pages (#6080) and flows (#7646), the II18nService method list missed setSupportedLocales (#7679), the optional-method set missed getFieldLabels (#4127), os i18n check's reported-kind list named 5 of 14 live source kinds, and os i18n extract's flag list named 4 of 12. Every one hides shipped capability — the deader-than-real direction, 11 of the 17. Two independent falsehoods: objects.OBJ.label documented as required when the schema makes it optional by design, and an authored item's unresolvable-locale skip called 'a silent skip' when the sync first falls back to a BCP-47-looking item name and then warns naming the row. Two false-example facts: the CRM Blueprint described a layout examples/app-crm does not have (it is one crm.translation.ts with en+zh-CN, no index.ts, no ja-JP/es-ES), and 'the shipped example apps' layout is examples/app-todo alone. Both eval rubric rows were carriers: 'secondary o.* format' and 'ICU is experimental' would each have taught a grader to fail a correct answer. PR is DRAFT for human merge (governed published surface); needs:contract-review attached to the PR AND to this card in the same stroke as PR creation, never self-cleared; skip-changeset attached and confirmed by delayed read-back. My face is untouched on origin/main since the branch point (verified by diff, not by absence of branches); no open PR appeared on it mid-flight. No objectui probe was needed — this package makes no cross-repo renderer claim, so no probe worktree was created.",
 "tests": "All gates run at final commit 5d16fd477 (git rev-parse --short HEAD, quoted from that run). Builds first, under scripts/pm/os-verify-lock.sh: @objectstack/service-i18n... / @objectstack/lint... / @objectstack/platform-objects... / client+client-react, VERDICT command-exit 0 each — so every read is against rebuilt dist, never a stale .d.ts. GATE UNION (21 commands; families derived by 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack', exit 0, 14 families, re-derived byte-identical on the final commit after the push): check-ci-filter-parity 0, check-cross-package-test-inputs 0, check-shard-attestation 0, check-skills-token-ratchet 0, check:doc-formula-expressions 0, check:agent-test-spelling 0, check:corpus-claim-drift 0, check:cross-package-test-inputs 0, check:doc-authoring 0, check:pm-governed-merges 0, check:role-word 0, check:skill-compatibility 0, check:skill-frame-sync 0, check:nul-bytes 0, check:pm-skill-ratchet 0, check:pm-skill-id-lint 0, check:skill-frame-freshness 0, spec check:skill-examples 0, spec check:skill-refs 0, spec check:skill-docs 0. Exit codes captured by redirect-then-capture, never after a pipe. check-test-completeness exited 3 = NOT MEASURED, quoting the gate's own printed verdict: 'It is not a red, and there is nothing here to fix' — it grades a saved turbo test log and none exists locally. Ratchet judgement line: 'check-skills-token-ratchet: skills/objectstack-i18n/SKILL.md is 6337 tokens (ceiling 6338; headroom 1)' and evals/README.md 410/411. check:skill-examples judgement line: '260 prose examples type-check across 3 surface(s) — every marked block parsed' — population verified NON-EMPTY (this package contributes 4 os:check fences), not an empty-population skip; its first run refused with a PREREQUISITE (client-react dist unbuilt), which was resolved by building, not by reading it as red. ESLINT: run WHOLE, not narrowed — 'pnpm lint' (eslint . --no-inline-config) exit 0 in 59s under the lock; separately measured that both changed files report 'File ignored because no matching configuration was supplied', i.e. markdown is outside the linted population (no files block in eslint.config.mjs names .md). EXECUTED PROBES (two temporary scripts, deleted before commit — the diff is 2 markdown files, git status clean, verified): (a) schema/adapter probe from packages/services/service-i18n — TranslationDataSchema keys came back ['objects','apps','messages','globalActions','dashboards','pages','flows','settings','metadataForms','settingsCommon'] (10, doc said 8); ObjectTranslationDataSchema keys ['label','pluralLabel','description','fields','_views','_actions','_sections','_tabs']; _actions keys ['label','description','confirmText','successMessage','params','resultDialog']; parsing {objects:{task:{pluralLabel:'Tasks'}}} with no label -> success=true, disproving 'label is required'; all 10 LEGACY_OBJECT_FIRST_KEYS rejected at the ITEM door AND 'o' rejected at the BUNDLE door, with defineTranslationBundle({en:{o:...}}) THROWING — disproving 'files registered in that shape resolve to nothing'; interpolation double '{{userName}}' -> 'Welcome, Alice!', single '{userName}' passed through unchanged, an ICU string echoed its key; setSupportedLocales present on BOTH shipped adapters (file=true memory=true) while getFieldLabels/getCoverage/suggestTranslations are absent from both. (b) CLI probe via tsx in packages/cli, driving collectExpectedEntries and computeI18nCoverage on a real fixture — 15 distinct ExpectedEntry.source kinds emitted; 12 distinct kinds reached os i18n check's issue list from one small fixture (action, app, dashboard, field, globalAction, metadataForm, navigation, object, option, page, view, widget) against the doc's 5; objects.task._tabs.urgent.label harvested, proving _tabs is a live gated surface; --strict flipped 783 warnings to 783 errors, confirming the promotion claim. Every measurement above is a RUN, not a read: the non-vacuity control is satisfied by facts 1, 2, 3, 5, 8, 9, 10, 13 and 14 independently. NOT MEASURABLE (about 14, recorded not skipped): the shipped 'os' binary end-to-end against a real project (flag definitions and the underlying walker/detector were driven directly instead); boot-time and publish-time loading of authored translation items (hook wiring read, not booted — the programme's recurring cross-package-runtime class); the metadata-door HTTP rejection path (verified at the schema door); I18nServicePlugin route self-registration at kernel:ready (read, not booted); and the ADR-0053 currency fallback end-to-end render (the manifest contract and ExecutionContext assembly were verified; the render path is cross-package runtime).",
 "mcp_calls": "11 — one issue_read get_comments (#13658 tail beyond the payload channel's 15 items), one create_pull_request, two search_issues (positive control FIRST, which hit, then the dedupe query), one list_issues ordered by CREATED_AT for minute-fresh duplicates, two issue_write creates (the two findings), two issue_write label updates (union-write on PR #13833 and on this card), one pull_request_read get for the authoritative body/draft/label read-back, plus this comment. Everything else went zero-quota: git for all history and branch checks, and the public-repo page payload for card bodies, comments and both label read-backs. REST was probed once and is CLOSED for this session (403 'GitHub access is not enabled for this session' on a repo-scoped read), so the label writes took the declared fallback: read current set, union, whole-group write, compare read-back, plus a delayed second read-back — both sets confirmed intact with nothing stripped.",
 "open_questions": [],
 "out_of_scope_findings": [
 "filed as #13835: lint/validate-translation-references has no _tabs leg — an authored key naming a filter-preset tab that does not exist warns nobody, while _views/_sections/_actions orphans do. Exactly the shape of #11608 (the flows leg), which was accepted and has since landed — one group over. os i18n check already DEMANDS a tab translation, so the surface is gated on the demand side and unguarded on the orphan side.",
 "filed as #13837: 'os i18n check --help' under-counts what it reports by 9 of 14 key kinds — the CLI-side twin of correction #10 in PR #13833 (the #13809 pattern: fixing only the skill leaves the CLI's own description teaching the same falsehood at the moment a user runs the command). metadataForm alone was 769 of 783 issues on the measured fixture.",
 "NOT filed, handed to PM as a programme-level observation: 15 of this flight's 20 landing sites are enumerations that fell behind a schema addition. #13678 (named-identifier liveness gate) covers the PHANTOM direction — a skill naming an identifier that does not exist. This is its INVERSE — a schema member with no row — and it is just as mechanical: compare a doc's enumeration against the shape's key set. Suggest it as a sub-issue of #13678 rather than a new root card, since one gate could reasonably carry both directions. Deliberately not opened as a third card without PM triage."
 ]
}

---

Generated by Claude Code