> ## Documentation Index
> Fetch the complete documentation index at: https://docs.userplane.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Subagents

> Four isolated subagents that power the plugin's slash commands — each with scoped tools and a single job

Each [slash command](/integrations/claude-code-commands) delegates to exactly one subagent. Subagents run in isolated context windows with their own tool permissions — no shared state, no cross-calling.

## Architecture

| Command                | Subagent        | Tools                               | Can write files |
| ---------------------- | --------------- | ----------------------------------- | --------------- |
| `/userplane:integrate` | integrate-agent | Read, Glob, Grep, Bash, Edit, Write | Yes             |
| `/userplane:audit`     | audit-agent     | Read, Glob, Grep, Bash (read-only)  | No              |
| `/userplane:debug`     | debug-agent     | Workspace MCP tools, Read, Grep     | No              |
| `/userplane:privacy`   | privacy-agent   | Read, Glob, Grep                    | No              |

Only the integrate-agent can write files. Only the debug-agent can access recording data.

<CardGroup cols={2}>
  <Card title="integrate-agent" icon="plus" href="/integrations/claude-code-integrate-agent">
    Writes the full Userplane install for your framework
  </Card>

  <Card title="audit-agent" icon="clipboard-check" href="/integrations/claude-code-audit-agent">
    Read-only verification against the framework skill
  </Card>

  <Card title="debug-agent" icon="bug" href="/integrations/claude-code-debug-agent">
    Fetches and correlates recording session data via MCP
  </Card>

  <Card title="privacy-agent" icon="shield-halved" href="/integrations/claude-code-privacy-agent">
    Scans the repo for PII exposure in recordings
  </Card>
</CardGroup>

## Related articles

* [Commands](/integrations/claude-code-commands) — the slash commands that invoke these subagents.
* [Skills](/integrations/claude-code-skills) — the knowledge layer subagents load for each task.
* [MCP Servers](/integrations/claude-code-mcp) — the workspace server the debug-agent connects to.
* [Cursor Workflows](/integrations/cursor-workflows) — Cursor commands that use the same workflow model.
* [Codex Workflows](/integrations/codex-workflows) — Codex workflow skills for the same jobs.