obra/superpowers@writing-skills

Apply test-driven development to skill documentation by enforcing a RED-GREEN-REFACTOR cycle: establish baseline testing, write minimal skill instructions, and refactor to close loopholes. This ensures AI agent behavior is thoroughly tested before documentation is finalized.

The RED-GREEN-REFACTOR cycle for AI skills is a documentation process where you first watch the agent fail without the skill, write minimal instructions to make it pass, then refactor to eliminate loopholes. It treats skill documentation with code-level testing rigor.

Prevent rationalizations in agent behavior testing by mandating rigorous baseline tests before writing documentation. Counter common excuses for skipping test coverage by requiring a passing GREEN state that proves the skill instructions actually resolve the targeted agent failure.

AI agents fail to follow skill documentation consistently due to loopholes in the instructions. Refactoring skill documentation through a rigorous testing cycle closes these loopholes, preventing rationalizations and ensuring the agent behavior remains predictable and effective.

Yes, you can use TDD principles to maintain existing AI agent skills by applying the RED-GREEN-REFACTOR cycle during updates. Re-establish baseline testing for current agent behavior, modify the skill documentation, and refactor to ensure updates do not introduce new loopholes.

The best way to ensure new AI agent skills are discoverable and effective is to mandate a rigorous test-driven development cycle. By testing baseline agent failures and refactoring minimal skill writing to close loopholes, documentation remains bulletproof and highly functional.