This is TDD for documentation. You write pressure scenarios with subagents to watch them fail without the skill, document rules to make them pass, then iterate to close loopholes. The approach treats skills as onboarding guides for specialized knowledge: techniques, patterns, references, or tool integrations. It maps the RED-GREEN-REFACTOR cycle directly to skill creation, which means you always baseline agent behavior before writing instructions. The structure guidance is solid, frontmatter matters for discovery, and the flat namespace keeps things searchable. If you're writing skills that agents keep finding ways to rationalize around, the systematic pressure testing here will tighten them up.
npx -y skills add neolabhq/context-engineering-kit --skill create-skill --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
anthropics/skills
obra/superpowers