sp-dispatching-parallel-agents

To run parallel agent investigations, you dispatch one focused sub-agent per independent problem domain using parallel sessions_spawn, enforcing clear per-agent scope and constraints to prevent interdependence, then aggregate their findings. This resolves multiple failures concurrently without shared state.

You should use concurrent agents for workflow investigation when 2 or more independent tasks or test file failures occur across subsystems that can run without shared state or sequential dependencies, and fast resolution is critical. Sequential execution is better for interdependent problems.

The best way to coordinate sub-agents working on unrelated test failures is dispatching one agent per independent root cause domain in parallel, ensuring each agent has a clear scope and strict constraints to prevent interdependence, then synthesizing the aggregated outcomes.

Parallel agent dispatch does not work for tasks that share state or have sequential dependencies. It is designed exclusively for independent-domain dispatch where 2 or more tasks can run concurrently without interdependence, ensuring isolated execution and structured result review.