superpowers-writing-skills

TDD for skill writing uses a RED-GREEN-REFACTOR cycle to validate agent instructions. You draft a minimal skill, run it against a baseline RED test to capture failures, then iteratively refine and regression test until the agent consistently complies.

Rationalization tables document baseline failures and explicit counters to prevent agents from taking shortcuts during skill execution. They surface rationalizations, enforcing disciplined skill creation by requiring structured reviews and test-first validation.

Yes, subagents coordinate the testing and documentation of skills by running RED-GREEN-REFACTOR cycles. They surface rationalizations and provide real-time feedback during iterative refinement, ensuring deployment-ready documentation.

Progressive disclosure organizes large reference content within skills for efficient discovery and loading. This approach supports cross-skill references, linking to other skills and external guidance to improve reuse and discovery.

To prevent shortcuts, use red-flag rationalization capture with baseline failure documentation and explicit counters. This enforces disciplined skill creation by requiring test-first validation and structured reviews before deployment.