What problem does it solve?
It solves the problem of slow, single-threaded assistants by letting you spawn background agents that work independently and in parallel.
Core Features & Use Cases
- Parallel subagent execution: Run independent workers concurrently for research, planning, or implementation.
- Scoped tool access via roles: Use
researcher, coder, and planner roles to restrict tool permissions to what each task needs.
- Mid-run parent updates: Use
notify_parent so the main conversation can react to important progress without waiting.
- Fork mode for context reuse: Spawn subagents that inherit the parent’s full messages and memory for near-free context carryover.
Use Case: You need to research an unfamiliar library and simultaneously start an implementation plan; spawn a researcher and a planner in parallel, then read the results when they finish.
Quick Start
Spawn a background research agent labeled "investigate-auth" with the objective "Find how this project secures authentication and summarize the recommended approach," and let it notify the parent while it runs.