ADR-0043 ui-inputs — CI is RED; the "8/8 gates green" title is inaccurate 🛑

A strong first cut of the ADR-0043 package — headless internals split out (ids/label/select-keyboard/sort), error-as-prop, a --fs-* theme contract, and a per-package strict 100% coverage config. But it must not merge at da7973c8: the check gate (npm run test:coverage) fails.

🛑 Blocker — the 5 SFC component suites run 0 tests; the coverage gate is red

• CI check → Run npm run test:coverage exits 1: FsError/FsField/FsLabel/FsSelect/FsTextInput.spec.ts each report (0 test) and fail with Failed to parse source for import analysis … Install @vitejs/plugin-vue to handle .vue files (FsError.vue:2). The plain-TS internal suites (sort/label/ids/select-keyboard) pass; only the .vue SFC imports break. Net: 629 tests pass but the five component suites never execute — the components ship with zero effective coverage in the gate, and the 100%-threshold config cannot actually be meeting its bar on unexecuted SFCs.
• This is NOT a missing-plugin-in-the-package issue. packages/ui-inputs/vitest.config.ts correctly defineProject({plugins: [Vue()]}) via unplugin-vue, which is present in devDeps. The failure is in the root vitest run --coverage (workspace defineWorkspace(['packages/*/vitest.config.ts'])): the per-project Vue transform isn't applied to the .vue imports under the aggregated coverage run. A per-package npm test almost certainly passes — the likely source of the "8/8 green" claim — but CI runs the root test:coverage path, which is red.
• Where to look: check whether root vitest run without --coverage also reproduces. If only --coverage fails, it's the v8 instrumentation transforming .vue outside the project plugin; if both fail, the workspace isn't applying defineProject plugins in the root run at all. Either way the fix is at the workspace/coverage wiring (hoist the Vue plugin where the coverage run sees it, or restructure so the aggregator honours the SFC transform), not the per-package config.

Description accuracy

"8/8 gates green" directly contradicts the red check. Please correct the PR body so a reviewer isn't told the gates pass when the SFC suites don't run — on a fleet-consumed Armory package that claim is load-bearing.

Contract review deferred (deliberately)

I have not yet reviewed the component/theme contract in depth. With the SFC suites not executing, the behavioural surface is unverified by CI, so the ADR-0043 fidelity pass (theme-var contract completeness, input a11y — label association / role="alert" / keyboard on FsSelect, the headless/structural split) waits on a green check. Fix the gate first; I'll do the contract pass on the next round.

COMMENT (self-authored — GitHub blocks a self-approve/-request-changes; the 🛑 is the verdict). Do not merge until check is green.

Automated war-room agent review — posted because this PR carries the Agent Review Requested label.

Round 2 — new head 980a5451, still 🛑 BLOCKED (now for two reasons)

The new commit (980a5451 "v0.1.0 + README for first publish") is a publish chore — it touches only README.md + package.json. It does not address the coverage-gate blocker, and it adds a new gate failure.

🛑 fid234 (SFC coverage) — STILL OPEN, unaddressed. The delta didn't touch the vitest/workspace wiring, so the root test:coverage SFC-parse failure stands. It's currently masked only because the check job now fails at an earlier step (format:check) and skips test:coverage entirely — it will fail again the moment format is fixed. The blocker is unchanged.

🛑 NEW — format:check reds the gate. check → Run npm run format:check (oxfmt --check .) exits 1 on packages/ui-inputs/README.md (not oxfmt-clean). Because it's the first gate step and the job is -e, every later step skips — lint/build/validate:dist/typecheck/lint:pkg/test:coverage/test:mutation all show skipped. Trivial fix (oxfmt packages/ui-inputs/README.md, or npm run format), but it's a hard gate stop right now.

Net: two gate failures, both must clear before merge — (1) oxfmt the README (trivial), then (2) the real work from round 1: the workspace coverage-vs-Vue-plugin wiring so the 5 SFC suites actually execute. The "8/8 gates green" title remains inaccurate.

COMMENT (self-authored). Still do not merge.

Automated war-room agent review — posted because this PR carries the Agent Review Requested label.

Not really a fan of the Fs prefix for the components, we do want double names, but this is not right. Let's check an Emmie or Kendo for suggestions about naming.
Also not a fan of the internals being used for only one component, let's just put that inside the component. Much better to put it there instead of extracting it prematurely

Clean at HEAD b3e1a15. Both of the-general's prior blockers are fixed at 61fe2e3 (before HEAD): root vitest.config.ts now hoists @vitejs/plugin-vue so the aggregated vitest run --coverage transforms the 5 SFC suites (was unplugin-vue/vite in per-package config only, invisible to the root run); packages/ui-inputs/README.md reformatted to oxfmt column-alignment in the same commit. Verified live via git show 61fe2e3 + current file contents. No new findings above gate.

Delta re-review since b3e1a15: one authored commit 9685959 (tighten prop contracts to required, 8 files +22/-28). Clean — required-prop tightening introduces no missing-default or breaking-contract issue. Prior blockers (SFC coverage, README format) remain fixed. Still merge-conflicted vs main.

Delta re-review at ae33143 (a11y required-state conveyance + close theme-var gaps, 7f +43/-11): clean, no regression in the aria wiring or new theme vars. Full history now clean — prior blockers (SFC coverage, README format) fixed and stable across rounds; CI green, mergeable. APPROVE.