dispatching-parallel-agents

Debug multiple independent failures simultaneously by dispatching dedicated parallel agents. Each agent is assigned to a separate problem domain, such as a specific subsystem or test file, to investigate issues concurrently without cross-domain interference.

The best way to handle parallel debugging across subsystems is dispatching one agent per problem domain. This structured workflow isolates independent failures and eliminates shared state, ensuring each subsystem is investigated concurrently and safely.

Dispatch parallel agents for task investigation when you have two or more independent problems with no shared state. This approach works best when each problem domain can be fully isolated and investigated concurrently without interfering with other tasks.

Define prompts and expected outputs for parallel agents by specifying structured inputs, goals, and expected results for each task. This ensures each agent operates self-contained and reports concise findings for consistent reviews across all dispatched investigations.

Parallel task dispatch is not suitable for bugs that share application state. This approach requires completely independent problems with no shared state, as each agent must operate in isolation to avoid cross-domain interference during concurrent debugging.