Dispatch Parallel Agents for Independent Debugging

This skill is a decision framework and template for dispatching multiple AI agents in parallel when facing two or more independent problems, such as several unrelated test failures across different files or broken subsystems. Its core principle is to dispatch one agent per independent problem domain and let them work concurrently rather than investigating everything sequentially. The pattern has four steps: identify independent domains by grouping failures by what's actually broken; create focused agent tasks, each with a specific scope, a clear goal, explicit constraints such as "don't change other code", and an expected output format; dispatch the agents in parallel, shown with Task(...) calls running concurrently in a Claude Code or AI environment; and, once agents return, review each summary, verify the fixes don't conflict, run the full test suite, and integrate the changes.

Use it when 3+ test files are failing with different root causes, multiple subsystems are broken independently, each problem can be understood without context from the others, and there is no shared state between the investigations. Do not use it when the failures are related (fixing one might fix the others), when understanding requires seeing the full system state, when you are still in exploratory debugging and don't know what's broken yet, or when the agents would interfere with each other by editing the same files or resources.

A good agent prompt is focused on one problem domain, self-contained with all the context needed (paste the actual error messages and test names rather than describing the bug vaguely), and specific about what the agent should return, typically a summary of the root cause and the changes made. The skill lists common mistakes to avoid: a prompt that is too broad ("fix all the tests"), one with no context, one with no constraints on scope, or one with a vague output request. A worked example dispatches three agents against three separate failing test files, each instructed not to just increase timeouts but to find the real timing or logic issue.

Built for AI coding environments that support concurrent task dispatch, such as Claude Code's Task(...) mechanism for running multiple agents at once.

Developers or AI agents coordinating debugging or fix-up work across multiple unrelated failures, who want a repeatable way to split the work into independent, parallel-safe agent tasks and then verify the results integrate cleanly - illustrated by a real session where 6 failures across 3 files were fixed by 3 parallel agents with zero conflicts.