skill-authoring

To validate agent behavior documentation, apply a RED-GREEN-REFACTOR workflow that proves compliance against pressure scenarios, checking that baseline failure tests pass and loopholes are closed before shipping.

TDD for prompt engineering converts skill writing into a RED-GREEN-REFACTOR cycle using baseline behavior checks, ensuring agents actively follow documented process rules rather than just interpreting them.

Write a Claude search-optimized description using a "When to Use" format that avoids tempting the agent to shortcut full instructions, ensuring the skill is reliably triggered during relevant tasks.

Agents bypass process rules when documentation lacks compliance proof against pressure scenarios. Iterative loophole closure through refactoring ensures rationalizations are eliminated and pass behavior is reliable.

Yes, a TDD-mapped baseline failure test is required for SKILL.md entries to verify that the documented technique, pattern, or process rule actively enforces intended agent behavior.

The best way to ensure quality assurance for reusable agent techniques is running a pressure scenario without the skill, writing the skill to fix specific failure modes, and refactoring until no rationalizations remain.