Add two minimal tools for agent orchestration:
Spawns an agent asynchronously. Returns immediately with run_id.
Input:
{
"task": "string, required - what the agent should do",
"workflow": "string, optional - coder|researcher|reviewer, default coder",
"name": "string, optional - human-readable name"
}
Output:
{
"run_id": "abc123",
"name": "foo-impl",
"status": "running"
}
Implementation: Wraps agentd run internally.
Non-blocking status check on a spawned agent.
Input:
{
"run_id": "abc123"
}
Output:
{
"status": "running|completed|failed",
"result": "...", // if completed
"error": "...", // if failed
"cost_cents": 12.5,
"duration_secs": 45
}
Implementation: Wraps agentd status --json.