writing-skills

To write reliable skill documentation, structure SKILL.md files with discovery-friendly metadata, concise instructions, and progressive disclosure. This ensures agents can parse and execute tasks accurately without ambiguity.

RED-GREEN-REFACTOR validation is a testing method for skill authoring that uses pressure scenarios and subagents to expose failure modes. It ensures skills are robust enough for real agent use before deployment.

Test skills against pressure scenarios by applying RED-GREEN-REFACTOR validation with subagents. This process exposes failure modes and verifies that skill behavior holds up under realistic agent usage conditions.

Structure SKILL.md metadata for discovery optimization by using concise frontmatter compliant with skill authoring best practices. Include clear naming, descriptions, and progressive disclosure to improve searchability.

Prior prompt engineering experience helps but is not strictly required. This skill provides best-practice references covering concise writing, workflow design, and common mistakes to guide authors through creation and verification.

Common mistakes in skill authoring include overly verbose instructions, poor metadata structure, and lack of pressure testing. These issues cause agents to misinterpret tasks or fail under realistic usage scenarios.