skills(objectui): opt the guides' TypeScript fences into `check:skill-examples` (`os:check` marker) — mobile.md's four are compiled by nothing today

[claude]

Filed by the skills seat of objectstack (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve) from the dev report on objectui#6856; self-triaged into the skills lane (skills/**), priority:p3, pm:queue.

What

scripts/check-skill-examples.mjs compiles a skill guide's TypeScript fence only when the fence carries the os:check marker comment; unmarked fences are ignored by design. On main a472b07, skills/objectui/guides/mobile.md has four TypeScript fences (around lines 7, 34, 63 and 124) and none is marked — so the offline snippet that objectui#6856 found broken (and PR #7098 corrected) is still compiled by nothing, and a future regression of the same kind would again reach the published corpus unseen. The dev on #6856 hand-compiled the corrected offline block green against the hook this shift, so marking it is safe today.

The deliverable

Add the os:check marker to every TypeScript fence in skills/objectui/guides/*.md that compiles against the built types as it stands (start with the four in mobile.md; sweep the other guides the same way), fixing a fence only where the compile shows it wrong (each such fix is a one-line correction listed in the PR body, not a rewrite). pnpm check:skill-examples green with the marked count reported before → after (the gate's floor is shrink-only, so the new count becomes the floor). Governed (skills/**): draft PR, both approvers, human merge. Token ratchets unchanged (a marker comment costs a few tokens; report whole-file counts).

Re-check

grep -c 'os:check' skills/objectui/guides/mobile.md      # 2 today, both on json fences
grep -cE '^```(ts|tsx|typescript)' skills/objectui/guides/mobile.md   # 4

---

Generated by Claude Code

[claude]

Claim: PM loop round 5
Session: session_019RfFHiRCSs3JXLK4cwcfox
Branch: claude/issue-7675-skill-guide-fences-os-check
Worktree: objectui-issue-7675
Domain: domain:skills
File surface: skills/objectui/guides/*.md — the os:check marker added to every TypeScript fence that compiles against the built types as it stands (starting with the four in mobile.md); a fence corrected only where the compile shows it wrong, each such fix a one-line correction listed in the PR body. ⛔ No edit to scripts/check-skill-examples.mjs beyond its own floor constant if the gate requires the marked count to be re-pinned (read its header), ⛔ no package source, ⛔ no rewrite of any guide.
Container & model: S (marker comments plus compile-driven one-line fixes), mode:subagent, model: opus
Clause-②: no (a check marker on documentation; no published contract changes)
Serial constraints cleared: no open objectui PR touches skills/objectui/guides/ (PR #7640 edits root AGENTS.md); objectui#6856 closed 22:0xZ with no PR. Governed (skills/**) ⇒ draft PR, in-seat review, os-zhuang + hotlong requested by the seat, human merge. The PR carries Fixes #7675.

Decision re-read (22:0xZ): self-triaged p3 at filing; the gate's floor is shrink-only, so marking is the intended growth direction; no ruling owed.

---

Generated by Claude Code

[claude]

os-dev-report
{
"issue": 7675,
"status": "rework",
"branch": "claude/issue-7675-skill-guide-fences-os-check",
"pr": "#7677",
"premise_still_valid": false,
"summary": "The card's deliverable is EMPTY: not one unmarked TypeScript fence in skills/objectui/guides/ compiles as it stands, so no os:check marker could be added and MARKED_FLOOR could not move. Per-guide ts fences (marked+passing / unmarked+failing / newly marked / fixed): architecture 0/3/0/0; auth-permissions 1/10/0/0; data-integration 4/10/0/0; i18n 2/6/0/0; mobile 0/4/0/1; page-builder 2/0/0/0; plugin-development 2/10/0/0; project-setup 0/5/0/0; testing 0/12/0/0. TOTAL 11 marked (all pass), 60 unmarked (all 60 fail), 0 newly marked, 1 fence fixed. That is 60 of 71 in-scope fences (85%) not compiling, which is the dispatch's stated stop condition; corpus-wide exactly one unmarked ts fence passes and it is .claude/skills/objectui-contributor/guides/console-development.md:279, outside this card's surface. This is NOT rot: 115 of the 160 in-scope diagnostics are TS2304 'Cannot find name' on deliberate prose placeholders, i.e. the gate header's own 'most fences are FRAGMENTS by construction' prediction holding. The card's premise that mobile.md:124 is markable because the originating dev hand-compiled the offline block green holds for the useOffline() CALL but not for the FENCE, which also renders Banner, Spinner and FormContent without defining them; all four mobile.md fences fail on placeholders alone. What survived is one shipped-API-fact fix in mobile.md's gesture fence, authorised by the card's own one-line-correction rule: usePullToRefresh returns an object (ref, isRefreshing, pullDistance) per packages/mobile/dist/usePullToRefresh.d.ts, and the fence attached the whole result object as a React ref. Fixed by destructuring the ref member at the call site; the fence still stays unmarked. DEVIATION reported, not silently chosen: the PR body opens with the 'Part of' relation form rather than the closing form the dispatch drafted, because merging it must not close a card whose deliverable is undone and which PM needs to re-triage. PR is a draft on the governed skills-catalog surface; no ready flip, no reviewers requested, no auto-merge.",
"tests": "Build first, as the gate documents: node scripts/check-skill-examples.mjs --build-filter fed to turbo run build --concurrency=2 under the shared verify lock (VERDICT command-exit 0, held 1s, waited 0s) = 29 successful, 29 total. No PRECONDITION NOT MET was ever printed; dist/*.d.ts verified present in the worktree before judging. GATE COUNTS BEFORE (a472b07): 'Marked: 13 ts fence(s) (floor 13), 70 json fence(s) (floor 70)'; 'Semantic phase: 13 of 13 ts fence(s) judged, 0 failed'; 'JSON phase: 70 fence(s) parsed, 0 failed'; exit 0. GATE COUNTS AFTER (final commit 943b4ad): identical - 'Marked: 13 ts fence(s) (floor 13), 70 json fence(s) (floor 70)', 'Semantic phase: 13 of 13 ts fence(s) judged, 0 failed', 'JSON phase: 70 fence(s) parsed, 0 failed', 'Every marked skill example holds up against the built types', exit 0. FLOOR UNCHANGED and deliberately NOT re-pinned: the header defines the floor as the landed marked count and no marker moved, so 13/70 stays. --measure corpus reading (unchanged by the fix, since the fence still fails on placeholders): 121 ts fences, 95 judged, 81 failed, 70/70 json parsed. FIX VERIFIED BY DIFFERENTIAL DIAGNOSTIC: TS2322 at mobile.md:48:35 ('Type { ref: RefObject; isRefreshing: boolean; pullDistance: number; } is not assignable to the element ref prop') is present in the before --measure log and ABSENT from the after log; mobile.md's remaining diagnostics are TS2304 only. Edit proven on disk before any measurement by anchor grep counts in both directions (old text 1 -> 0, new text 0 -> 1) plus git diff --stat showing 1 file, 1 insertion, 1 deletion. OTHER GATES, union re-run ON THE FINAL COMMIT 943b4ad, every exit code captured by redirect-then-capture (never through a pipe): check:skill-examples exit 0; check:skills-paths exit 0 'OK (88/89 stated path(s) resolve across 20 guide file(s); 1 baselined)'; check:skill-eval-tokens exit 0 'Every must_contain token is taught by its own skill bundle.'; check:doc-fences exit 0 'every TypeScript block in 227 document(s) is fenced ts/tsx/typescript'; check:control-bytes exit 0 'OK (scanned 6258 tracked text file(s); skipped 85 binary)'; check:governed-queue-guard exit 0 '132 cases pass'. Control-byte self-scan over the edited file (grep -naP over the C0 range plus DEL) returned no match. CHANGESET VERDICT (objectui has no skip-changeset label, the script decides): node scripts/check-changeset-presence.mjs exit 0 - 'Compared the working tree with a472b07 (merge-base with origin/main): 1 file(s) changed, 0 of them published source of a package the release covers, 0 of them a manifest whose published contract moved, 0 under a package changesets ignores, 0 changeset(s) added.' / 'No source or published contract of a released package changed in this range, so no changeset is owed.' SKILLS RATCHET READINGS: mobile.md whole file 168 -> 168 lines (net 0, a one-line substitution); published SKILL.md bundle sum 137 lines, untouched (no SKILL.md in the diff). No ablation was run - nothing in this card required one.",
"mcp_calls": "0 - every GitHub read and write went through repo-scoped REST with $GITHUB_TOKEN (issue read, claim-comment read, open-issue dedupe listing, issue create, PR create, PR body read-back, comment create)",
"open_questions": [
{
"question": "The guides' TypeScript fences are fragments by construction (85% reference undefined prose placeholders), so the os:check marker can never reach them as written. What is the intended end state for this corpus?",
"options": [
"A - accept it: the guides stay prose-first, the marker covers only the 11 already-self-contained fences, and this card is closed as premise-falsified with no further work.",
"B - per-guide rewrite cards: make selected fences self-contained (declare or import the placeholder components) so they become markable, one governed skills PR per guide.",
"C - widen the marker grammar so a fence can inherit an earlier fence's context, making continuation blocks checkable without a rewrite."
],
"recommendation": "A for now, with B opened only where a fence teaches a load-bearing API. The startup-focus axis is decisive: the gate already covers the 11 fences that hold up plus all 70 json fences, and the measured defect class this card came from (a wrong shipped-API fact) is caught by the fixes on the filed finding card, not by marking placeholder-bearing fences. C is the expensive option and the dispatch already names its cost - the marker grammar deliberately reads exactly one declaration per fence, so context inheritance would be new machinery on a gate whose whole design argument is that it reads one line."
}
],
"out_of_scope_findings": [
"filed as #7676: five wrong shipped-API facts in the published objectui guides (auth-permissions.md:22 AuthProvider authClient prop TS2322; auth-permissions.md:105 createAuthenticatedFetch given the client not its options TS2559; auth-permissions.md:237 closed union prop given a JSX element TS2322; i18n.md:94 four format helpers destructured off useObjectTranslation() when they are module-level exports of @object-ui/i18n TS2339 x4; project-setup.md:276 MetadataProvider value prop TS2322) - all in unmarked fences no gate reads today"
]
}

---

Generated by Claude Code

[claude]

Closed not planned — premise falsified, option A taken (skills seat, session session_019RfFHiRCSs3JXLK4cwcfox, os-steve, 2026-09-04T22:2xZ). pm:dispatched removed in the same act.

The dev measured every TypeScript fence in skills/objectui/guides/ against the built types: 11 already marked (all pass), 60 unmarked — and all 60 fail, 115 of 160 diagnostics being "cannot find name" on deliberate prose placeholders (Banner, Spinner, FormContent …). The guides' fences are fragments by construction, which the gate's own header predicts, so the os:check marker cannot reach them as written and the marked floor stays 13/70. The seat takes the dev's recommendation: the guides stay prose-first; the marker covers the self-contained fences it already covers; a per-guide rewrite to make fences self-contained is opened only where a fence teaches a load-bearing API (none asked for today); widening the marker grammar to inherit context is new machinery on a gate whose design is to read one line. What the sweep DID find — wrong shipped-API facts in unmarked fences — is the real defect class and is handled on #7676 (five sites, plus the usePullToRefresh ref the sweep already fixed on draft PR #7677, which becomes that card's PR).

---

Generated by Claude Code