writing-skills

To enforce consistent agent behavior when writing skill documentation, you apply a test-driven development approach to SKILL.md authoring, using RED-GREEN-REFACTOR cycles to validate compliance before deployment. This prevents agents from bypassing required processes.

Test-driven development for skill authoring is a validation process where you write SKILL.md entries by defining failing baseline tests, observing rationalizations, and iterating until agents comply under maximum pressure. It reduces brittle, ad-hoc skill documentation.

You pressure test AI agents to find process loopholes by running realistic pressure scenarios with subagent tests without the skill applied, surfacing rationalizations, and then iterating on the SKILL.md until agents comply under maximum pressure.

Yes, writing skills requires specific frontmatter formats for SKILL.md files. It satisfies requirements for YAML frontmatter, discovery-optimized descriptions, and progressive disclosure of reference files to keep runtime context efficient.

The best way to prevent agents from bypassing required processes is to apply test-driven development to SKILL.md authoring and verification, using pressure-based subagent tests to surface and close loopholes before deployment.

Your agent bypasses skill instructions under pressure because ad-hoc skill documentation lacks enforcement mechanisms. Applying TDD with pressure-based subagent tests surfaces rationalizations and validates compliance under maximum pressure.