writing-skills

To apply TDD to agent skill documentation, you map RED-GREEN-REFACTOR to skill-authoring by capturing a failing baseline agent rationalization, writing minimal SKILL.md changes to force compliance, and re-testing to close loopholes. This validates instructions under pressure scenarios rather than assuming clarity.

AI follows incorrect documentation under pressure because agents rationalize excuses involving time, sunk-cost, authority, or exhaustion constraints. Pressure-scenario testing captures these exact rationalizations verbatim, allowing you to add explicit negations and red flags until compliance holds.

The best way to bulletproof SKILL.md frontmatter against agent loopholes is iterative loophole closure. You capture baseline rationalizations, add explicit negations and rationalization tables to the trigger-focused YAML description, then re-verify until behavior stays consistent under stress.

Yes, writing-skills requires failing-test baselines for compliance testing. You must run a baseline pressure scenario without the skill to capture verbatim rationalizations, ensuring the documentation changes are verified against baseline agent behavior rather than assumed clarity.

Yes, you can use progressive disclosure to optimize agent skills for discovery. Writing-skills enforces a concise, trigger-focused YAML description and structured skill body, ensuring findability while maintaining compliance-proof documentation that resists rationalization.

You should not use TDD for skill documentation when updating trivial metadata that does not influence agent behavior under pressure. This approach is designed for creating, editing, and deploying agent skills that must resist rationalization under time, sunk-cost, authority, and exhaustion pressures.