native-parallel

To coordinate parallel agents without heavy infrastructure, use native runtime concurrency primitives. This approach applies dependency-aware lane partitioning and dispatches tasks to Claude or Gemini, maximizing throughput while minimizing idle time.

Dependency-aware lane partitioning splits two or more independent tasks into parallel execution lanes. It organizes work so independent tasks run concurrently while respecting dependencies, allowing efficient orchestration across multiple agents.

Dispatch independent tasks concurrently by choosing Claude Task tool or Gemini async-invoke based on runtime optimization. This native dispatch mechanism sends parallel work across multiple agents without requiring external orchestration infrastructure.

Yes, you can use worktrees for isolated parallel execution across multiple agents. The system creates on-demand worktrees when needed, allowing concurrent tasks to operate in isolated environments without conflicting with each other.

Collect and reconcile results from parallel agent tasks by integrating them into a SCRATCHPAD. This collective result handling moves completed parallel work into subsequent stages, ensuring progression and state management.