spawn_run
Start a new external coding-agent run with specified backend and role, returning immediately to allow parallel task execution.
Instructions
Start a new external coding-agent run and return immediately.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| backend | Yes | Backend to execute the run. Supported values are "codex", "claude_code", and "remote_a2a". | |
| role | Yes | Supervisor role for this run: planner, worker, or reviewer. | |
| prompt | No | Primary instruction for the coding agent run. | |
| input_message | No | Structured input message for multi-part or A2A-compatible runs. | |
| cwd | Yes | Absolute working directory where the agent should run and where artifacts are stored. | |
| session_mode | Yes | Use "new" to create a fresh session or "resume" to continue an existing one. | |
| session_id | No | ||
| profile | No | Optional path to a profile/persona/job-description file. Leave blank unless explicitly instructed to use a profile. | |
| output_schema | No | Optional JSON Schema for structured final output from the run. | |
| metadata | No | Optional orchestration metadata for task/step correlation. It is stored but not interpreted by the MCP server. | |
| backend_config | No | Optional backend-specific configuration, such as remote_a2a agent_url and auth headers. |