sub-agent-decompose-worker-review

Multi-agent task decomposition is automated by a three-agent pipeline that splits complex tasks into ordered subtasks, executes them with a worker agent, and validates outcomes through a reviewer agent. This enables clear handoffs and iterative quality checks.

The best way to structure a multi-agent pipeline is using modular task-decomposer, task-worker, and work-reviewer roles. This pattern separates task breakdown, safe incremental implementation, and output validation to ensure reproducible results.

The fix loop works by having the work-reviewer agent validate task-worker outputs, surface issues, and trigger corrective iterations. The worker agent applies minimal changes again, passing artifacts back through the pipeline until quality criteria are met.

Yes, you can use this multi-agent decomposition approach for data migrations, feature implementations, and research tasks. It applies to any complex workflow requiring clear task breakdown, incremental execution, and iterative quality checks across defined roles.

No external dependencies are required to run the task-decomposer, worker, and reviewer pipeline. The skill operates independently, coordinating artifacts between the decomposition, worker, and reviewer stages without additional framework installations.

You should not use a three-agent decomposition and review pipeline for simple, single-step tasks that lack complex dependencies or acceptance criteria. Overhead from orchestration and iterative quality checks outweighs benefits for trivial workflows.