dispatching-parallel-agents

Dispatching parallel agents allows each investigation to happen concurrently in isolation. This approach applies when you have 3+ failures across test files or subsystems that are unrelated and can be explored independently, preventing sequential bottlenecks.

You should use parallel agents for debugging when you have 3 or more independent failures across different test files or subsystems. It is designed for scenarios where these unrelated bugs can be safely explored in isolation before a unified integration plan is applied.

You dispatch parallel agents by issuing a command with each independent problem domain as an argument, such as passing multiple test file names. This triggers focused task scopes for each agent, executing them in parallel and generating a concise summary per domain.

Yes, dispatching parallel agents works for investigating race conditions if they are isolated within separate subsystems. As long as the failures are unrelated and can be explored in isolation, agents can concurrently debug them and provide a unified integration plan.

The limitation of using parallel agents for debugging is that failures must be completely unrelated and able to be explored in isolation. If your test failures or subsystem bugs are interdependent, dispatching agents concurrently will not work effectively before integration.