Subagents

The main agent can spawn independent subagents for focused tasks. Each subagent gets its own conversation context and tool access, preventing context bloat in the main session.

Single Subagent

The spawn_subagent tool creates one focused worker:

{
  "task": "Read all files in lib/src/auth/ and summarize the authentication flow",
  "model_ref": "anthropic/claude-haiku-4-5"
}

Parallel Subagents

The spawn_parallel_subagents tool runs multiple tasks concurrently:

{
  "tasks": [
    "Analyze the auth module structure",
    "Analyze the payments module structure",
    "Analyze the notifications module structure"
  ],
  "model_ref": "anthropic/claude-haiku-4-5"
}

Safety Rules

• Read-only by default — subagents can only use read_file, list_directory, and grep
• Depth limited — subagents can spawn sub-subagents, but only up to 2 levels deep
• Model override — pass model_ref (e.g. "anthropic/claude-haiku-4-5") to use a cheaper/faster model for exploration; defaults to the active model
• Fresh context — each subagent starts clean, no conversation history inherited

See also

• SubagentTools
• AgentManager