@object-ui/fields declares a `./style.css` export but nothing builds `dist/index.css` for it — the subpath quick-start teaches looks unresolvable

[yinlianghui]

Out-of-scope finding from the #3883 / #3884 docs sweep (PR pending). Not fixed there — that PR's file surface is content/docs/guide/*.md only.

What was checked, on origin/main @ 11c1e71

packages/fields/package.json declares the subpath:

"exports": {
  ".": { "types": "./dist/index.d.ts", "import": "./dist/index.js", "require": "./dist/index.cjs" },
  "./style.css": "./dist/index.css"
}

but nothing in that package appears to produce dist/index.css:

1. packages/fields/package.json scripts.build is tsc && vite build — no CSS step. The only package with one is @object-ui/components: "build": "vite build && node scripts/build-css.mjs", and scripts/build-css.mjs exists only there (grep -rln build-css --include=package.json packages/ returns one file).
2. packages/fields/src contains no .css file at all, and no module under it imports one (grep -rn 'import.*\.css' packages/fields/src is empty). Vite's library build emits a stylesheet only when one is in the entry graph.
3. packages/fields/vite.config.ts has no CSS entry or cssCodeSplit handling.
4. Nothing inside this repo imports the subpath, so no build here would have noticed: repo-wide, @object-ui/fields/style.css occurs only in content/docs/guide/quick-start.md. The three real in-repo consumers all import the components one (packages/plugin-gantt/demo/main.tsx, packages/plugin-grid/demo/main.tsx, packages/plugin-grid/demo/bulk-actions.tsx).

Why it matters

content/docs/guide/quick-start.md:57 teaches consumers to write @import "@object-ui/fields/style.css"; as step 3 of the first-run setup. If the file is not in the published tarball, a bundler resolving that @import fails the consumer's build outright — a harder failure than the missing-utilities class #3883 / #3884 are about, and on the same page.

What I did NOT verify, and how to settle it

I did not run the build: this was a docs-only worktree with no install, and the #3883 dispatch's measurement basis explicitly scoped me to cheap spot-verification. So the evidence above is a source read, not a measurement. The deciding runs are:

• pnpm --filter @object-ui/fields build && ls packages/fields/dist/index.css
• pnpm --filter @object-ui/fields exec npm pack --dry-run and check whether dist/index.css is in the file list.

If it is genuinely absent, the two ends are (a) give fields a CSS entry plus a build step mirroring components/scripts/build-css.mjs, or (b) drop the ./style.css export and the quick-start.md:57 line that teaches it. Which one is right depends on whether the fields widgets carry styling of their own beyond the components palette — not a call this finding makes.

Dedup

Searched open issues for fields style.css, "@object-ui/fields" css, exports dist/index.css build, subpath export missing file. Hits were only #3883 / #3884 (the sweep this came from), #3890 (CLI temp-app alias table) and #4043 (spec symbol derivation) — none covers this export.

---

Generated by Claude Code

[yinlianghui]

Measurement upgrade (objectui seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3, 2026-08-10): the filer flagged this as a source read pending the deciding build. The published-artifact reading settles it without a build — inspected the npm tarballs directly:

• @object-ui/fields@17.3.0 — https://registry.npmjs.org/@object-ui/fields/-/fields-17.3.0.tgz contains zero .css files (grep over the full tar -tzf listing). The "./style.css": "./dist/index.css" export is a phantom: a consumer's @import '@object-ui/fields/style.css' fails to resolve against the shipped package.
• Control: @object-ui/components@17.3.0 tarball ships package/dist/index.css as expected.

So this is a real, user-visible defect in the published package, not a hypothesis. Consequence already acted on: PR #4060 (the #3883/#3884 docs sweep) was sent back for a targeted rework to stop teaching the fields import on all three guide pages — the docs half of this defect dies there; what remains for this card is the package half (either build real CSS for fields or remove the phantom export — and if fields' class shapes are not a subset of components' published utilities, removal alone under-styles fields, so the disposition needs that one measurement first).

---

Generated by Claude Code

[yinlianghui]

Triage + release board (objectui seat PM — this repo's board producer): pm:queue + target:v17, class ② — the published @object-ui/fields@17.3.0 package declares a "./style.css" export that resolves to nothing (zero .css files in the tarball, measured above): a shipped contract surface that lies. The docs half died in PR #4060; this card is the package half.

Claim: PM loop round 2
Session: session_017Qqyix2QcnpUC9XeYVDzx3
Branch: claude/issue-4059-fields-style-export
Worktree: objectui-issue-4059
Domain: repo:objectui (whole-repo seat)
File surface: packages/fields/package.json (exports), possibly packages/fields build config, + tests/gates pinning the disposition. ⛔ Region-exclusive vs the in-flight #3569 branch: do not touch GridField*/source component files in packages/fields. Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: opus
Serial constraints cleared: #3569's pushed branch touches packages/fields source (GridField.tsx), disjoint from packaging/exports; no other claim or PR on packages/fields packaging.

Delegated ruling with a falsifiable premise (maintainer authorized seat rulings from the long-term-stability standpoint, 2026-08-10; veto window open):

• Ruled route: retire the phantom export (remove "./style.css" from packages/fields' exports + a gate asserting every declared subpath export resolves to a file that ships), premised on: fields' class shapes are a subset of the utilities already published in @object-ui/components' dist/index.css. The premise must be measured first (the quick-start 教的两条 @source node_modules 行是冗余的:实测多产 100 kB CSS、14 条没人用的选择器,且补不回它看似要补的主题 utility #3884-style read: extract class-shaped tokens from fields' built dist/*.js, compare against components' published dist/index.css rule set).
• If the premise is falsified (fields has class shapes components' sheet does not carry): STOP — do not remove the export and do not silently switch to building CSS; report the fork with the measurement (the alternative is a real CSS build for fields, which is a product-shape call).
• Rationale: an export that has never resolved cannot have a working consumer, so removing it breaks no one and ends the lie; building CSS for a package with no CSS sources only makes sense if the subset premise fails.

---

Generated by Claude Code

[yinlianghui]

Ruling premise measured — FALSIFIED. No PR opened; the disposition is back with the PM.

The delegated ruling (retire the phantom export) was premised on: fields' class shapes are a subset of the utilities already published in @object-ui/components' dist/index.css. Measured first, per the instruction. It is not a subset.

1. The phantom export, confirmed three independent ways (on origin/main @ 0a09793f2)

• Local build: pnpm --filter @object-ui/fields build -> exit 0, find packages/fields/dist -name '*.css' | wc -l = 0.
• Published tarball @object-ui/fields@17.3.0: 274 entries, zero .css. Its package.json still declares "./style.css": "./dist/index.css".
• Control @object-ui/components@17.3.0: ships package/dist/index.css (165 871 bytes).

2. Premise measurement (#3884's method)

Compiled fields' built dist/index.js + dist/index.cjs — the exact surface a node_modules glob can see — against components' own @theme (its src/index.css, with only the @source line repointed), then compared the produced rule set against components' published dist/index.css (downloaded tarball, not a local build).

set	classes	selectors
fields' dist, scanned + compiled	1510	1557
@object-ui/components@17.3.0 published dist/index.css	1358	1411
in fields, NOT in components' published sheet	152	150

Method control (the same pipeline pointed at components' own dist, reproducing #3884's reading): 0 classes / 0 selectors missing — 1358 vs 1358, i.e. components' pre-built CSS really is a strict superset of its own scan surface, exactly as #3884 measured. The pipeline is therefore validated, and the 152 are real.

Spot-traced to fields source, so these are not bundling artifacts:

class	origin
cursor-crosshair	packages/fields/src/widgets/SignatureField.tsx:113
group/email	packages/fields/src/index.tsx:1165
text-indigo-700	packages/fields/src/index.tsx:922 (tag colour map)
min-w-[8ch]	packages/fields/src/widgets/TagsField.tsx:79
hover:fill-yellow-500	packages/fields/src/widgets/RatingField.tsx:83

3. The split that decides the fork

Of the 152, 17 depend on the ObjectUI @theme block, which lives in packages/components/src/index.css and is not published (files ships dist only). No consumer-side @source can regenerate them — #3780 measured that grid already:

bg-background/60, bg-destructive/5, bg-muted/60, bg-primary/20, border-background, border-border/40, border-muted-foreground/25, border-muted-foreground/30, focus-visible:ring-ring/60, focus:ring-ring/60, hover:bg-accent/30, hover:bg-muted/80, hover:bg-primary/10, hover:bg-primary/50, hover:border-primary/50, ring-border/60, ring-destructive/50

The remaining 135 are plain Tailwind built-ins (palette, sizing, positioning) and are recoverable by an @source over node_modules/@object-ui/fields/dist.

4. Corroboration: no in-repo host relies on a pre-built fields sheet

Every host that renders fields reaches into fields' source instead — apps/console/src/index.css:17, examples/console-starter/src/index.css:9, examples/byo-backend-console/src/index.css:29, all @source '.../packages/fields/src/**/*.{ts,tsx}'. That is a workspace-relative path no external consumer can write. So the repo has never exercised the ./style.css export, which is why the phantom survived — and it also means in-repo rendering proves nothing about consumer rendering.

5. Why removal alone is no longer a no-op

The ruling's rationale was "an export that has never resolved cannot have a working consumer, so removing it breaks no one". That still holds for the export. But it now collides with a second measured fact: after PR #4060, content/docs/guide/quick-start.md:60 teaches

That is the whole styling setup: you do not add @source lines for the ObjectUI packages, and pointing Tailwind at them inside node_modules only regenerates utilities the import already gave you.

That sentence is true for components (#3884 measured exactly that) and false for fields: it generalizes a components-only reading across the package boundary. A consumer following the current quick-start today gets none of the 152. Removing the export makes that the permanent, documented state with no supported remedy — which is a product-shape call, not a cleanup.

Reporting needs_decision with the A/B fork per the ruling's own escape clause. No files changed, no PR, worktree torn down.

---

Generated by Claude Code

[yinlianghui]

Delegated ruling #2 (objectui seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3, 2026-08-10): the premise of ruling #1 is falsified by the measurement above — ruled route is now B, narrow variant. Authorization: maintainer's 2026-08-10 delegation (decisions may be taken by this seat from the long-term-stability standpoint). Veto window open.

The dev's stop was correct and is the escape clause working as designed: 152 fields-only class shapes, 17 of them unrecoverable by any consumer-side @source (their @theme tokens are unpublished), falsify "removal breaks no one and costs nothing" — removal would ratify a permanently under-styled fields with no supported remedy.

Ruling: B — give @object-ui/fields a real CSS build, narrow variant (compile against @object-ui/components' theme, emit only fields' own utilities; do NOT re-emit the shared @theme — components' sheet owns it and consumers already import it first). Three axes:

1. 实际业务需求 — measured, not asserted: the 152 utilities style nine named shipped widgets (tag colours, signature crosshair, rating hover, tags sizing); fields' whole job is rendering inputs. The 17 theme-dependent ones have exactly one possible producer: the package itself.
2. 长远合理性/稳定性 — B is the existing packages/components/scripts/build-css.mjs pattern applied consistently, not a second mechanism; it is the only fork under which the published manifest tells the truth (same shape as the maintainer's #5175 ruling: when the capability and the declaration both exist, enforce — make the implementation catch up — rather than remove). A/A' both make "fields is styled" a property of consumer Tailwind configuration — the consumer-side-tolerance shape contract-first rejects.
3. 防 AI 犯错 — with B, one @import is correct by construction and the durable gate can land green asserting a true thing. A' fails as slightly-wrong colours even when transcribed perfectly — the least detectable failure mode available.

Startup-focus counterweight acknowledged (zero external reports since publication = weak pull); the cost side is one mirrored build script, which is why it does not tip the ruling.

Implementation scope (same PR, per the dev's own note):

• fields CSS build (narrow), export stays and becomes true; state in the PR how the coupling to components' unpublished src/index.css is expressed (build-time path is acceptable inside the monorepo — it mirrors how components builds itself);
• the durable gate lands in the SAME PR (extend scripts/__tests__/package-files-exist.test.ts with the producer test + files-containment test described in the report) — never before it, it would be red on main;
• docs alignment in the same PR: re-teach the fields style.css import where docs(guide): teach the Tailwind 4 CSS-first setup on theming / troubleshooting / quick-start #4060 removed it, and correct the over-generalized sentence at quick-start.md:60 — with B the measurement that justified the removal is superseded; docs must describe the package as it will ship from this commit;
• the changeset must say plainly that versions ≤17.3.0 never shipped this file (so the import only works from the release that carries this change).

---

Generated by Claude Code