writing-skills

Writing AI skills with TDD involves creating failing baseline tests for non-compliance, authoring minimal skills to pass them, and refactoring to close loopholes using pressure scenarios and rationalization countermeasures.

Rationalization countermeasures in AI skill development are techniques used to bulletproof skills against AI workarounds. They test skill compliance under realistic constraints, preventing the AI from skipping instructions or using excuses like "I'll review it later."

Yes, you can apply test-driven development to all AI skill types, including discipline-enforcing, technique, pattern, and reference skills. The methodology adapts the Red-Green-Refactor cycle to ensure discoverability and effectiveness across any documentation format.

The best way to pressure test AI skills for compliance is to develop realistic scenarios where the agent might skip instructions, such as under time pressure. You then evaluate the skill against these constraints and apply rationalization countermeasures to close loopholes.

AI agents skip instructions when skill documentation contains rationalization loopholes. Prevent this by using a TDD methodology to define failure scenarios, write minimal enforcing constraints, and test against new rationalizations.