harness:dispatching-parallel-agents

To debug multiple independent test failures in parallel, you can dispatch separate agents for each isolated problem domain. This prevents context bleeding by ensuring each agent stays focused on a single scoped failure without shared state interference.

The best way to fix unrelated race conditions concurrently is to group failures into separate problem domains and dispatch one agent per domain. This allows simultaneous investigation of independent issues like tool-approval or batch completion bugs without sequencing dependencies.

You should use parallel debugging agents when you have two or more independent failures with unrelated root causes and no shared state. It reduces wasted investigation time by allowing agents to work on scoped tasks simultaneously instead of serially.

To map parallel agent tasks safely, enforce features.json scope mapping and read harness-delta.md to block orphan parallel work. This ensures all dispatched agents respect the references → templates → skills → commands dependency direction.

Parallel agent dispatch is limited to independent failures without shared state or sequencing dependencies. If your debugging tasks require sequential execution or share state across subsystems, parallel agents may encounter conflicts and interfere with each other.

To integrate fixes after parallel debugging, review each agent's findings, check for conflicts across the scoped tasks, and run the full test suite to verify that all independent fixes work together without breaking existing functionality.