| name | produced-skill |
| description | Test fixture — produced SKILL.md ships a complete Setup (root-entry-file + Companion CSS) but Hard rules has zero lines matching shell vocabulary AND a token shape. SHELL_INVARIANTS must FAIL with the count assertion message naming the three pre-seeded slugs. |
Setup
Source: <reference-project> @ src/app/layout.tsx
import { ThemeProvider, BaseSurface } from 'ds-react'
import './globals.css'
export default function RootLayout({ children }) {
return (
<html lang="en">
<body>
<ThemeProvider>
<BaseSurface style={{ backgroundColor: 'var(--ds-surface-default)' }}>
{children}
</BaseSurface>
</ThemeProvider>
</body>
</html>
)
}
Companion CSS — src/app/globals.css
@import "ds-themes/light.css";
body {
background-color: var(--ds-surface-default);
color: var(--ds-text-default);
}
When to Load References
| Trigger | Files to load | Notes |
|---|
| button asks | references/components/button.md | per-component file |
Hard rules
- Any prop, variant, token, or asset the agent cannot ground in source gets a literal
[VERIFY] marker inline.
- Mark unverifiable facts
[VERIFY]. Report blockers instead of guessing.
Final checks
After generating UI: cite each component used to its source file; list any [VERIFY] markers; name the screen-level prompt built.
In scope: tokens, assets, component descriptions, component APIs.