writing-skills

Test-driven documentation for Claude skills uses RED-GREEN-REFACTOR cycles to draft, validate, and refine skill rules before deployment. This process enforces structured SKILL.md bodies and YAML frontmatter to prevent agent behavior regressions and drift.

Test-driven skill creation is an authoring method that validates new agent skills through baseline stress tests before release. You need it when building reusable skills that must resist rationalization and maintain consistent structure across edits.

Pre-deployment verification validates skill structure by enforcing YAML frontmatter with name and description fields, checking SKILL.md body consistency, and running safety checks for toxicity, dependencies, and components to ensure discoverability.

Writing skills requires no external dependencies or components to start. You only need a minimal skill draft following the RED-GREEN-REFACTOR workflow, which you then validate against baseline stress tests for structural compliance.

Preventing rationalization and drift in agent documentation requires applying test-driven design practices during skill creation. Structured rules, patterns, and references built through RED-GREEN-REFACTOR cycles resist drift and enforce consistent behavior.

Skills fail pre-deployment verification when they lack required YAML frontmatter fields, have an improperly structured SKILL.md body, or trigger safety checks for toxicity, unlisted dependencies, or missing components during validation.