Agent Skill

npx skills add josemarluedke/frontile

Once installed, ask your agent for UI and it builds with Frontile instead of guessing at an API it half-remembers:

"Add a settings form with a name field, an email field, and a save button."
"Put this destructive action behind a confirmation modal."
"Make this table sortable and paginated."
"Restyle these buttons to the danger color."
"This Select should let people type to filter."

What a skill is

A skill is a set of instructions an agent loads when it recognises that a task involves Frontile. It sits in your project, is plain Markdown you can read and edit, and needs no server, account, or running process. The skills CLI installs it and supports 35+ agents including Claude Code, Cursor, Codex, and Windsurf.

An MCP server answers questions over a live connection. A skill is loaded text. For a library whose facts are already on disk in your node_modules, text is enough, which is why Frontile ships a skill and no server.

What it carries

The skill holds judgment and routes to facts. It does not restate arguments, types, or defaults.

An installed skill is a copy of a file, frozen at install time, while your frontile dependency moves independently. Any argument list written into it would eventually describe a version nobody has. So it states where to look instead: first node_modules/frontile/declarations/**/*.d.ts, which is exact for the version you installed, then the Markdown mirrors for prose and examples.

What it does carry:

The lookup order, so the agent checks your installed types before inventing an argument.
Component selection for the genuinely ambiguous choices: Modal versus Drawer versus Popover, Select versus NativeSelect versus Autocomplete, Table versus SimpleTable.
The semantic color system: categories with named levels, and why bg-primary-500 does not exist.
.gts conventions, including the imports that are not where an agent expects them and the Tailwind @source lines that, when missing, leave every component unstyled.
The pre-0.18 argument names, for upgrades and for the older examples still circulating.

Reference files load on demand, so a question about colors does not pull in the migration guidance:

skills/frontile/
├── SKILL.md                          # Lookup order, common traps, routing table
└── references/
    ├── api-naming.md                 # Current vs pre-0.18 argument names
    ├── colors.md                     # Semantic categories and levels
    ├── component-selection.md        # Which component for which job
    └── gts-conventions.md            # Template authoring and setup

Using it

Most agents load the skill on their own once they recognise the task involves Frontile. In agents that support invoking a skill directly, /frontile does it explicitly.

Installing

Project-level by default, which is usually what you want: the skill is committed with the project, so everyone working on it gets the same guidance. Use --global when only some of your projects use Frontile and you would rather not carry it in every agent's context.

npx skills add josemarluedke/frontile          # this project
npx skills add josemarluedke/frontile --global # every project on this machine

The CLI writes into whichever directory each detected agent reads from. For Claude Code that is .claude/skills/frontile, and the files are plain Markdown you can open, edit, or commit.

Target one agent rather than every detected one:

npx skills add josemarluedke/frontile --agent claude-code
npx skills add josemarluedke/frontile --agent cursor

Installing offers a second skill, frontile-contributor-docs. That one is for work inside the Frontile repository itself, writing the component documentation that lives beside the source, and is of no use in an app that consumes Frontile. To skip the prompt:

npx skills add josemarluedke/frontile --skill frontile

Keeping it current

Updates are manual. The CLI pulls; nothing is pushed to you:

npx skills update

Worth running after upgrading Frontile, particularly across a minor version. An installed skill can sit arbitrarily far behind the version in your package.json, or ahead of it if you installed while tracking the development branch.

A stale skill costs you guidance rather than correctness: the facts that decide whether generated code compiles come from your own node_modules.

Without installing anything

Everything the skill points at is fetchable directly, and /llms.txt states the same lookup order in its preamble, so an agent pointed at the index alone arrives at the same place. See LLMs.txt .

Found a problem?

If the skill steers an agent wrong, that is a bug in the skill. Open an issue.