This stops you from building the wrong thing by forcing a structured conversation before any code gets written. When you get a vague request like "add login feature" or "implement payment," it runs a 100-point scoring rubric across functional clarity, technical specificity, implementation completeness, and business context. It asks focused questions in rounds of two or three until the spec hits 90 points, then generates a full PRD in your docs folder with acceptance criteria, execution phases, and risk assessment. The systematic approach works well for features over two days or anything involving cross-team work, though it can feel heavy for simple bug fixes where you already know exactly what to do.
npx skills add https://github.com/davila7/claude-code-templates --skill requirements-clarity