writing-skills

Test-driven documentation for agent skills is written by following a structured RED-GREEN-REFACTOR cycle to guide development, capture rationalizations, and iteratively refine instructions. This framework ensures clarity and robustness against agent loopholes.

The RED-GREEN-REFACTOR cycle for skill creation is a systematic testing process where you design pressure scenarios to reveal agent rationalizations, verify compliance, and iteratively refactor documentation to close loopholes. It produces bulletproof skills suitable for deployment.

You test agent compliance by designing pressure scenario testing methods that force agents to reveal rationalizations and loopholes under stress. This exposes weaknesses in the documentation, which you then systematically refactor to enforce desired behavior reliably.

Yes, this framework works for developing discipline-enforcing skills like TDD or verification requirements. It provides structured skill development and iterative improvement to ensure strict agent compliance under stress, making it ideal for AI teams.

You need scenario testing for documentation skills when your instructions must reliably enforce specific behavior under stress. It is required to identify and close documentation loopholes that allow agents to bypass intended constraints or verification requirements.

The best way to refactor agent skills to close loopholes is through an iterative improvement process using the RED-GREEN-REFACTOR cycle. You systematically refine the documentation after pressure scenario testing until it reliably enforces the desired behavior.