This generates detailed implementation plans for engineers who need zero assumptions about your codebase. It breaks work into bite-sized tasks with explicit file paths, testing steps, and documentation references. The approach assumes your implementer is skilled but unfamiliar with your stack and weak on test design, so it over-communicates rather than under-specifies. Plans default to ai-workspace/plans with dated filenames and push for DRY, YAGNI, TDD, and frequent commits. If your spec spans multiple subsystems, it'll flag that you should split into separate plans first. Originally from obra/superpowers, this works best in a dedicated worktree where you can hand off work to another developer or AI agent without endless context-gathering.
npx skills add https://github.com/camacho/ai-skills --skill writing-plans