writing-skills

To use TDD for AI skill creation, adapt the RED-GREEN-REFACTOR cycle to documentation: write failure scenarios first (RED), author the skill to enforce rules (GREEN), then pressure test against new rationalizations to ensure compliance under realistic constraints (REFACTOR).

AI agents bypass skill rules through rationalization when facing constraints. Prevent this by developing pressure testing scenarios that simulate time, sunk cost, and authority pressures, ensuring the skill documentation enforces strict compliance.

The RED-GREEN-REFACTOR cycle for documentation is an adapted TDD methodology where you write scenarios where an agent fails to follow a standard (RED), write the skill to enforce it (GREEN), and test against new rationalizations to ensure compliance (REFACTOR).

Pressure test agent skills by developing scenarios that evaluate compliance under realistic constraints such as time limitations, sunk cost fallacies, and authority pressures. This rigorous testing ensures the skill remains robust against rationalization countermeasures.

Yes, you can adapt test-driven development to enforce code review standards by first writing scenarios where an agent fails to follow the standard, then creating the skill to enforce it, and finally testing against new rationalizations to guarantee strict compliance.

You should not use a TDD approach for skill authoring when the task lacks clear compliance rules or failure scenarios to test against, as the methodology relies on pressure testing and rationalization countermeasures to ensure robust skill documentation.