feature-implementation

Feature-implementation executes multi-phase feature plans from ~/.ai/plans by maintaining persistent state tracking through implementation_state.md, using a Plan agent to explore the codebase and plan details, then implementing each phase progressively with validated outcomes. This disciplined approach prevents context overload and lets you resume work after interruptions.

Phased feature implementation breaks a feature plan into clearly defined phases and milestones, executing each stage with explicit dependencies and codebase exploration before moving forward. Use this approach when you have complex features requiring multiple steps, team coordination, or need to maintain a decision trace across development sessions.

Implementation_state.md records phase completion, decisions made, and current progress status throughout execution. This persistent state file lets you resume from where you left off, avoids repeating work, and provides a trace of what was explored and why each phase decision was made.

Yes. The Skill uses a Plan agent to assess existing codebase state and patterns before proposing implementation steps, ensuring new phases validate against established conventions and dependencies already in place, maintaining consistency across your feature and existing code.

You need a feature plan file in ~/.ai/plans/<feature-name>/ that defines phases, dependencies, and codebase areas to explore. The Skill reads this plan, initializes implementation_state.md if needed, and begins phase-by-phase execution with continuous progress updates.

The Plan agent explores both the plan details and your codebase to understand requirements and existing state before implementation begins, reducing guesswork and ensuring proposed steps fit your actual architecture and dependencies.