For agents and tools
A coding assistant that does not know the library makes up props. So that it does not have to guess, the package describes itself in machine-readable files, generated on every build from the real API.
What ships in the package
| File | How it is imported | Size | What it is |
|---|---|---|---|
| llms.txt | @amset/ui/llms.txt | 59 KB | The index for a language model: the rules and one line per component with its props. |
| llms-full.txt | @amset/ui/llms-full.txt | 917 KB | The same with every prop, its type and its description. |
| AGENTS.md | @amset/ui/AGENTS.md | 59 KB | The instructions for a coding assistant: rules and a summary of each component. |
| api.json | @amset/ui/api.json | 961 KB | The real props of each component read from the compiler, the icons and the utilities. |
| docs.json | @amset/ui/docs.json | 1 MB | The text of each component page: what it is, when to use it and when not to, keyboard and accessibility. |
| props.en.json | @amset/ui/props.en.json | 499 KB | The English description of each prop. |
| knobs.json | @amset/ui/knobs.json | 36 KB | The CSS variables that can be adjusted on each component. |
| examples.json | @amset/ui/examples.json | 149 KB | A real usage example for each story in the test bench (Storybook). |
| locales/en.json | @amset/ui/locales/en.json | 123 KB | Every text key in the library: the template for translating it into another language. |
llms.txt and llms-full.txt are also served by this site at its root, exactly as they come out of the package for version 0.35.0:
AGENTS.md and the skill
AGENTS.md starts with the rules that are not up for discussion (do not invent components or props, import from the index, do not write colours by hand) and continues with a summary of each component. It is enough to tell the assistant where it is:
Before writing code with @amset/ui, read node_modules/@amset/ui/dist/AGENTS.md.If a component or a prop is not in node_modules/@amset/ui/dist/api.json, it does not exist.The package also includes a skill for Claude Code, skills/amset-ui/SKILL.md, with the same rules and when to use it: copy it to your project's .claude/skills/.
The MCP server
@amset/mcp gives your assistant tools to search components, read their real props, see examples, compile a snippet against the installed library and generate a theme.
Tokens in Figma
The library's repository includes a Figma plugin that adds the tokens as Variables: a colour collection with light and dark modes (switching mode on a frame switches the whole theme, like data-theme in code) and another with typography, spacing, radii, control heights, durations and easing curves. Run it again when they change: it updates the values, it does not duplicate them.
npm run export:figma # in the amset-ui repository: writes dist/figma-tokens.jsonnpm run figma # tokens and components together, in figma-export/- Colours come out in hexadecimal (Figma does not understand OKLCH) and sizes in pixels.
- It works on any Figma plan: writing variables through the Figma API requires the Enterprise plan, a plugin does not.