dispatching-parallel-agents

Dispatching parallel agents assigns one agent per independent failing test to debug concurrently, reducing total investigation time by enforcing a one-problem-per-agent pattern. This prevents cross-domain interference by isolating tasks and defining clear goals for each agent.

The best way to orchestrate parallel workflows for unrelated failures is using a central dispatch to assign and aggregate results. This pattern applies to scenarios with 2+ unrelated failures or tasks that can be solved in isolation, avoiding shared state between agents.

Yes, you can dispatch parallel agents to investigate race conditions and timing issues concurrently. The skill supports running three independent investigations in parallel, assigning separate agents to investigate tool-abort, batch-completion, and tool-approval test failures without shared state.

You should not use parallel agent orchestration when tasks share state or have cross-domain dependencies, as the skill enforces a one-problem-per-agent pattern that relies on isolation. It only applies to scenarios with 2+ unrelated failures or tasks that can be solved independently.