writing-skills

Authoring reliable AI skills requires creating SKILL.md files with valid YAML frontmatter, concise guidance, and progressive disclosure to ensure agents discover and follow instructions accurately under pressure.

Evaluation-driven development tests AI skills using baseline scenarios, pressure testing, and rationalization analysis to verify agent compliance and ensure operational reliability before deployment.

Structure SKILL.md frontmatter with required YAML name and description fields targeting discovery scenarios, ensuring prompt engineering workflows remain discoverable by AI agents.

AI agents ignore skill documentation under pressure when it lacks explicit safeguards against common rationalizations and targeted evaluation scenarios to verify compliance.

Yes, you can apply evaluation scenarios to existing SKILL.md files by running pressure testing and rationalization analysis to verify agent compliance and iteratively refine documentation.