subagent-driven-development

Subagent-driven-development is a core skill within the Superpowers methodology, designed to automate complex software development workflows by delegating granular tasks to specialized agents. By creating a fresh subagent for every individual task, the system ensures that each unit of work operates within a clean, isolated context, preventing history pollution and maintaining focus. This approach is intended for software engineers and coding agents who have already finalized an implementation plan and need to iterate rapidly through independent coding tasks within a single session.

The system enforces a rigorous two-stage review process for every completed task. First, a dedicated spec reviewer evaluates whether the implementation satisfies the initial requirements and architectural constraints. Second, a code quality reviewer assesses the implementation for maintainability, DRY principles, YAGNI compliance, and general best practices. By separating these concerns, the framework achieves high-speed iteration without compromising on stability or adherence to the project's original design documents.

• Dispatches specialized, ephemeral subagents to handle distinct coding tasks with zero context carryover.

• Implements a mandatory dual-layer validation workflow: spec compliance verification followed by code quality assessment.

• Operates on the principle of task independence, ensuring that tasks are tightly scoped to specific file paths or functions.

• Dynamically selects models based on task complexity, using smaller models for mechanical implementation and more capable models for complex review and architecture tasks.

• Integrates seamlessly with existing Superpowers skills such as test-driven-development and finishing-a-development-branch.

• Ensure you have a clear, approved implementation plan before triggering this skill, as it relies on predefined task scopes.

• Monitor implementer status carefully; outcomes such as DONE or DONE_WITH_CONCERNS trigger specific logic branches for fixes or progress.

• Best suited for tasks where individual implementation steps are well-defined and do not require heavy cross-task context switching.

• If tasks are highly coupled or require global architectural changes, consider alternative workflows to avoid synchronization issues between subagents.

• Use in conjunction with TodoWrite tracking to maintain visibility over the progress of the overall implementation plan.