This is TDD for documentation. You write pressure scenarios with subagents to see where they violate a rule (red), write the skill to address those specific violations (green), then iterate to close loopholes (refactor). The core insight is that if you didn't watch an agent fail without the skill, you don't know if it teaches the right thing. Most of the document focuses on making skills discoverable through rich descriptions that list triggering conditions, not workflow summaries, because testing showed Claude will follow a workflow description instead of reading the actual skill content. Requires understanding TDD fundamentals first.
npx skills add https://github.com/davila7/claude-code-templates --skill writing-skills