dispatching-parallel-agents

To fix independent test failures concurrently, dispatch focused agents to investigate each failing test file in parallel using self-contained prompts, then review and integrate the results to verify the full test suite.

The best way to debug unrelated test files is dispatching isolated agents with self-contained instructions for each subsystem, preventing session-state leakage, followed by a review step to detect conflicts before merging.

Use parallel agents for triaging test failures when multiple unrelated subsystems or test files can be diagnosed without shared state or sequential dependencies, such as when a refactor causes three or more unrelated failures.

To prevent session-state leakage when dispatching concurrent agents, provide each agent only the exact context and goals it needs through self-contained prompts, ensuring isolated contexts per agent during parallel investigations.

To integrate fixes after parallel agent dispatch, collect concise summaries from each agent, verify there are no conflicting changes across isolated contexts, and run the full test suite before merging the proposed fixes.

Dispatching parallel agents is not suitable when test failures share state or have sequential dependencies, because isolated contexts per agent prevent detecting shared state issues during concurrent investigations.