dispatching-parallel-agents

Parallel debugging of independent failures requires grouping distinct bugs into separate problem domains, assigning one self-contained agent per domain, executing concurrently, then reconciling changes and validating with the full test suite.

Use parallel agents for debugging when you have 3 or more independent failing test files or broken subsystems that do not share state or require full-system context. Each agent receives self-contained instructions to prevent inheriting unrelated context during parallel dispatch.

Preventing race conditions with parallel debugging agents requires keeping agent instructions self-contained to avoid inheriting unrelated context. After parallel execution, reconcile changes to avoid conflicts and validate the integrated results with the full test suite.

The best way to orchestrate task investigation across unrelated subsystems is to propose one targeted investigation prompt per independent failing subsystem, run them in parallel, then integrate the results and verify with the full suite to ensure no conflicts were introduced.

You should not use parallel agents if your failing tests share state or require full-system context. This technique specifically requires independent problem domains where each agent can own one scope with self-contained context for clean coordination.

Verifying integration after running parallel debugging agents requires reconciling changes to avoid conflicts and running the full test suite. This integration verification ensures the concurrently generated fixes from each independent problem domain work together correctly.