dispatching-parallel-agents

Group failures by independent problem domain, then dispatch one agent per domain with a focused prompt containing the failing test names, error messages, constraints, and expected output. Review each agent's summary and run the full test suite to verify the fixes integrate without conflicts.

Use parallel agents when failures are independent, such as different test files or subsystems with no shared state, and each problem can be understood without context from the others. Avoid parallel dispatch when failures are related, since fixing one may fix others.

A good agent prompt is focused on one problem domain, self-contained with all needed context such as error messages and test names, explicit about constraints like not changing unrelated code, and specific about the expected output such as a summary of root cause and fixes.

Agents working on shared state or the same files can interfere with each other and produce conflicting edits. The pattern requires verifying that domains are truly independent before dispatch and checking for conflicts when integrating results.

Read each agent's summary to understand what changed, check whether agents edited the same code, run the full test suite to confirm all fixes work together, and spot-check changes since agents can make systematic errors.