# Using a coding agent

Your coding agent drives Introspection through the CLI, which also serves its workflows.

```bash
npm install -g @introspection-ai/cli   # Node.js 24 or later
introspection setup                    # Pi, Recipes, and a discovery skill for each detected agent
introspection skills                   # the workflows the agent reads
introspection login                    # you approve in the browser
```

`setup --skill claude-code` (or `codex`) installs the discovery skill for one
agent only. Mention Introspection in your request so the agent uses it: "Use
Introspection to create a refund agent." An agent without a skill directory reads
`https://docs.introspection.dev/agents.md`. Every page here is also served as
Markdown at its URL plus `.md`, indexed in `/llms.txt`.
`INTROSPECTION_SKILLS_INDEX_URL` points `introspection skills` at another skills
index, and `INTROSPECTION_SKILLS_CACHE_DIR` moves its local cache.

There is no read-only login, so limit the agent with its own permissions. In
`.claude/settings.json`:

```json
{
  "permissions": {
    "allow": ["Bash(introspection skills:*)", "Bash(introspection conversations get:*)", "Bash(introspection events list:*)"],
    "deny": ["Bash(introspection runtimes update:*)", "Bash(introspection runtimes delete:*)", "Bash(introspection api-keys:*)", "Bash(introspection members:*)"],
    "ask": ["Bash(introspection:*)"]
  }
}
```

Using a coding agent

Your coding agent drives Introspection through the CLI, which also serves its workflows.


npm install -g @introspection-ai/cli   # Node.js 24 or later
introspection setup                    # Pi, Recipes, and a discovery skill for each detected agent
introspection skills                   # the workflows the agent reads
introspection login                    # you approve in the browser

setup --skill claude-code (or codex) installs the discovery skill for one agent only. Mention Introspection in your request so the agent uses it: “Use Introspection to create a refund agent.” An agent without a skill directory reads https://docs.introspection.dev/agents.md. Every page here is also served as Markdown at its URL plus .md, indexed in /llms.txt. INTROSPECTION_SKILLS_INDEX_URL points introspection skills at another skills index, and INTROSPECTION_SKILLS_CACHE_DIR moves its local cache.

There is no read-only login, so limit the agent with its own permissions. In .claude/settings.json:


{
  "permissions": {
    "allow": ["Bash(introspection skills:*)", "Bash(introspection conversations get:*)", "Bash(introspection events list:*)"],
    "deny": ["Bash(introspection runtimes update:*)", "Bash(introspection runtimes delete:*)", "Bash(introspection api-keys:*)", "Bash(introspection members:*)"],
    "ask": ["Bash(introspection:*)"]
  }
}