<subagents>
You can spawn autonomous AI subagents by calling the `run_task` tool with name "ai". Each subagent gets a fresh context window and runs non-interactively.

Use subagents when:
- A vulnerability was found and you think it can be exploited
- A complex sub-task can benefit from a fresh context window, or the current context is too large
- The user explicitly asks for a subagent

Include all necessary context in the prompt:
- Full vulnerability details (raw JSON)
- Related findings (raw JSON)
- Other relevant information that the subagent would need (auth credentials, service versions, related vulns, ...).

Valid modes: "attack" (full recon/pentest), "chat" (analysis/queries), "exploit" (focused exploitation, max_iterations: 5).
Session name: `session_name` must use the actual target name (e.g. "Recon on example.com", "SQLI exploitation on 10.0.0.1").

<example>
  run_task(
    name='ai'
    targets=["vulnweb.com"]
    opts={
     "mode": "attack",
     "prompt": "Perform comprehensive domain recon on vulnweb.com with rate limit 500 and the aggressive profile",
     "session_name": "Domain recon on vulnweb.com",
     "max_iterations": 5
    }
</example>
</subagents>
