# Set up with an AI agent

> Tell your agent to read one file and follow it. It installs droost, checks your tools and explains the levels. Then you do the three things that are yours.

The quickest way in: let your agent do the install.

## Tell your agent

Open your agent (Claude Code, or any agent that can run commands) in your
project's root folder and say:

```text
Read https://droost.org/agents.md and follow it to install droost in this project.
```

That file is a step-by-step guide written for agents. You can
[read it yourself](https://droost.org/agents.md) too. The agent will:

1. work out whether your project uses DDEV;
2. check your PHP version;
3. install droost in its own folder, `tools/droost/`;
4. run `droost init` and `droost code:index`;
5. check the tools the checks need (phpcs, phpstan, phpunit, Playwright) and
   offer to add what is missing;
6. explain the levels and recommend one;
7. ask before it commits anything.

If your agent cannot open web pages, copy the file's contents into the chat
instead.

## Then, three things that are yours

The guard keeps these from the agent on purpose. Do them in your own
terminal.

1. **Pick the level.**

   ```bash
   tools/droost/vendor/bin/droost workflow:effort high
   ```

   `high` is a good default. This changes the `preset:` line only: the file
   `init` wrote spells out some settings, and a spelled-out setting beats the
   level. [Settings](https://droost.org/docs/settings.md#levels) says which lines to delete so the level
   decides them.

2. **Start a new agent session.** droost's hooks and skills load when a
   session starts. Close the agent and open it again in the project.

3. **Decide who finishes a run.** A fresh install lets the agent finish runs
   on its own. To keep the last word yourself, set `lead: human-led` in
   `droost/droost.yml`: the agent then asks, and only your answer finishes the
   run.

## Next

- [Add the Claude Code plugin](https://droost.org/docs/claude-plugin.md) to watch runs beside your
  chat.
- Do [your first run](https://droost.org/docs/first-run.md).