Applies Test-Driven Development principles to create reliable, subagent-tested documentation and skills.
This skill transforms the way Claude Code skills are created by treating documentation as code. It implements a rigorous RED-GREEN-REFACTOR cycle where authors must first prove a skill is necessary by watching a subagent fail a pressure scenario. By documenting the agent's rationalizations and iteratively refining the skill until loopholes are closed, it ensures that every skill added to the library is functional, concise, and effectively steers AI behavior. It provides strict guidelines on structure, naming, and token efficiency to maximize the effectiveness of custom Claude workflows.
主な機能
01RED-GREEN-REFACTOR workflow for process documentation
02Pressure-testing with subagents to verify agent compliance
0316 GitHub stars
04Claude Search Optimization (CSO) naming and description patterns
05Strict token efficiency guidelines for skill files
06Standardized directory and SKILL.md structure templates
ユースケース
01Refactoring existing documentation to close behavior loopholes and rationalizations
02Verifying agent adherence to complex process rules before library deployment
03Writing new Claude Code skills for custom domain-specific workflows
