Dispatching Parallel Agents

Routes independent problems to focused subagents running concurrently. Each agent gets narrow scope, clear goal, explicit constraints, and a required output shape. Your own context stays lean for coordination instead of drowning in per-task detail.

Use it when 3+ test files fail for different reasons, when multiple subsystems break independently, or when a plan has parallel workstreams with no shared state. Don't use it when failures are related, when exploratory debugging is still active, or when agents would edit the same files.

Outputs a dispatch plan with one agent per domain, then integrates their returns: read each summary, check for conflicts, run the full suite. Pairs with `subagent-driven-development` (the broader pattern) and `executing-plans` (when the plan itself is the parallelization source).