# Agent skill

URL: https://docs.router.ezat.io/agents/skill

> Install the ezat-router agent skill so Claude Code and other agents can set up and debug EZAT Router on their own.

## When to use this page

Use this page to install the `ezat-router` skill, which teaches an AI agent how to set up and debug tools, SDKs and code that call EZAT Router. It also has prompts you can paste into any agent instead.

## What the skill is

The skill is a folder in the [Agent Skills](https://agentskills.io) format: a `SKILL.md` file with a name and a description, plus short documents the agent loads only when it needs them. Claude Code reads skills from `~/.claude/skills/` and from a project's `.claude/skills/`, and other agents that support `SKILL.md` can use the same folder.

The description tells the agent when to use the skill: setting up or debugging anything that calls AI models through EZAT Router. When a task matches, Claude Code loads the skill on its own; you can also call it with `/ezat-router`.

| File | What it covers |
| --- | --- |
| `SKILL.md` | The fixed facts (base URLs, key format, headers, dashboard), the safety rules, and which file to load when |
| `quickstart.md` | Checking a key and making a first request |
| `claude-code.md` | Pointing Claude Code at EZAT Router |
| `openai.md` | OpenAI SDKs and tools that take an OpenAI base URL |
| `sdks.md` | The Anthropic SDKs for Python and TypeScript, and curl |
| `models.md` | Choosing a model, `GET /v1/models` and aliases |
| `billing.md` | Balance, credit, how requests are charged, interrupted streams |
| `errors.md` | Each error type, whether to retry, rate limits and `request-id` |

The skill holds no prices and no model list. It tells the agent to read them live, from `GET /v1/models` and from the [Models](https://docs.router.ezat.io/models.md) and [Pricing](https://docs.router.ezat.io/pricing.md) pages, so an installed copy doesn't go stale.

## Install it in Claude Code

For every project on your machine (`~/.claude/skills/`):

```bash title="Install for your user"
curl -fsSL https://docs.router.ezat.io/skill/ezat-router.zip -o /tmp/ezat-router.zip
mkdir -p ~/.claude/skills
unzip -o /tmp/ezat-router.zip -d ~/.claude/skills
```

For one project only, run this in the project's root folder (`.claude/skills/`):

```bash title="Install for this project"
curl -fsSL https://docs.router.ezat.io/skill/ezat-router.zip -o /tmp/ezat-router.zip
mkdir -p .claude/skills
unzip -o /tmp/ezat-router.zip -d .claude/skills
```

Both create an `ezat-router/` folder with `SKILL.md` in it. The project copy can be committed so your team gets it: the skill contains no keys or other secrets.

The files are also served one by one, starting at `https://docs.router.ezat.io/skill/ezat-router/SKILL.md`, for agents that read a skill from a URL.

### Check that it loaded

1. Start Claude Code (or keep your session open: Claude Code picks up new skills without a restart). If `~/.claude/skills/` or `.claude/skills/` didn't exist when the session started, run `/reload-skills`.
2. Run `/skills`. `ezat-router` should be in the list.
3. To use it explicitly, type `/ezat-router` followed by what you want, for example `/ezat-router set up Claude Code with the key in $EZAT_API_KEY`.

### Update it

An installed skill is a copy and never updates itself. Run the same commands again to replace it with the latest version. The version is in `SKILL.md`, under `metadata.version`, and changes are listed in the [Changelog](https://docs.router.ezat.io/changelog.md).

## Use any agent without installing anything

Paste a prompt like one of these into your agent. Export your key first (`export EZAT_API_KEY="ezat_sk_…"`) in the terminal the agent uses, so the agent never sees it in the chat.

```text title="Set up Claude Code"
Read https://docs.router.ezat.io/llms.txt and set up Claude Code to use EZAT Router with the key in $EZAT_API_KEY. Don't print the key. Check it with GET /v1/models first.
```

```text title="Set up a project"
Read https://docs.router.ezat.io/llms.txt, then change this project to call AI models through EZAT Router. Read the API key from the EZAT_API_KEY environment variable, keep it out of git and out of frontend code, and pick the model from GET /v1/models.
```

```text title="Debug an error"
Read https://docs.router.ezat.io/troubleshooting.md and https://docs.router.ezat.io/reference/errors.md, then find out why my EZAT Router requests fail. The key is in $EZAT_API_KEY; don't print it.
```

## For AI agents

1. The skill is at `https://docs.router.ezat.io/skill/ezat-router.zip` (one folder, `ezat-router/`). Its entry point is `https://docs.router.ezat.io/skill/ezat-router/SKILL.md`.
2. Install it only where the user asks: `~/.claude/skills/` for all their projects, or `.claude/skills/` in one project. Don't overwrite a different skill with the same folder name without asking.
3. Without the skill, read `https://docs.router.ezat.io/llms.txt` and then only the pages you need, as Markdown (`https://docs.router.ezat.io/<page>.md`).