This is TDD for documentation. You write pressure scenarios with subagents, watch them fail without the skill, write the documentation, then verify agents actually comply. The core insight is that if you didn't see an agent violate the rule first, you don't know if your skill teaches anything useful. It maps the full RED-GREEN-REFACTOR cycle to skill creation: baseline behavior is your failing test, the skill document is production code, and you refactor by finding new loopholes agents exploit. Includes practical guidance on frontmatter structure, when descriptions should trigger discovery versus when they accidentally replace the content, and why mechanical constraints belong in validation not documentation.
npx skills add https://github.com/unclecatvn/agent-skills --skill writing-skills