# Agent Skills

> Teach your coding agent to write commands, build programs and drive Fireargs CLIs.

Fireargs ships [Agent Skills](https://agentskills.io/) that tell coding agents such as Claude Code, Cursor and Codex how a Fireargs CLI works. Once installed, your agent knows how zod fields become arguments and options, how to wire a subcommand tree and how to drive a CLI through `--json` and `--llms`. You don't have to explain any of it in the prompt.

| Skill                 | Your agent uses it when                                                        |
| --------------------- | ------------------------------------------------------------------------------ |
| `fireargs-commands`   | writing a command: input and output schemas, arguments, options and its config |
| `fireargs-programs`   | building a subcommand tree and sharing parameters and commands across it       |
| `fireargs-automation` | driving a CLI through `--json` and `--llms`, testing it and handling errors    |

## Install from the website

fireargs.dev publishes its skills under `/.well-known/agent-skills/`, so the [`skills` CLI](https://github.com/vercel-labs/skills) can install them from the site URL. Run this at your project root:

```bash
npx skills add https://fireargs.dev
```

It asks which skills and agents to install for. To skip the prompts, pass them as flags:

```bash
npx skills add https://fireargs.dev --skill fireargs-commands --agent claude-code
```

Add `--global` to install for your user instead of the project, and `--list` to see what's available without installing anything. Run `npx skills update` later to pull newer versions.

## Install from the package

The skills also ship inside the npm package. Installing them from there keeps them matched to the Fireargs version in your lockfile, while the website always serves the latest release:

```bash
npx skills add ./node_modules/fireargs/skills
```

## Use them

You don't need to invoke a skill yourself. Agents read each skill's description and load it when a task matches. For example, asking for a new `deploy` command loads `fireargs-commands`, and asking to group commands under a `db` subcommand loads `fireargs-programs`. If your agent supports invoking skills by name, you can also call one directly, as with `/fireargs-automation` in Claude Code.

## See also

- [LLMs mode](/llms-mode/) — the `--llms` manifest an agent reads at runtime, rather than while writing code.