superpowers-subagent

To coordinate parallel task execution without context pollution, dispatch a fresh subagent for each task. This isolates execution, prevents context bleed, and enables focused work for faster iteration across multiple steps.

A two-stage review process validates AI-generated code by first confirming specification alignment, then validating code quality. This iterative feedback loop ensures the implementation meets requirements and maintains standards before final integration.

To implement multi-step features, dispatch a fresh subagent per task to handle planning, implementation, testing, and self-review. Apply a two-stage review to confirm spec compliance and code quality before merging the final code.

Yes, you can handle end-to-end task execution within a single session. By dispatching isolated subagents per task, each agent handles its own planning, implementing, testing, and self-review before final integration.

Context bleed happens in agent orchestration when multiple tasks share the same context window, causing bottlenecks. Avoid it by dispatching a fresh subagent per task to ensure isolated execution and maintain focused iteration.