What is the Red-Green-Refactor workflow for skills?

It is an adaptation of software testing where you first observe an agent fail a task (Red), write a skill to fix the failure (Green), and then refine the instructions to close any remaining loopholes (Refactor).

What is the 'Iron Law' of skill writing?

The Iron Law states: 'No skill without a failing test first.' This ensures that every instruction added to your library is necessary and addresses a verified performance gap.

When should I create a skill versus updating CLAUDE.md?

Create a skill for broad, reusable patterns or techniques that apply across multiple projects. Use CLAUDE.md for project-specific conventions and one-off solutions.

Why should I test skills with subagents?

Testing with subagents allows you to simulate high-pressure scenarios to ensure your documentation is robust and that the AI follows instructions even in complex edge cases.

Where are Claude Code skills stored locally?

Personal skills for Claude Code are typically stored in the ~/.claude/skills directory, while Codex-specific skills reside in ~/.codex/skills.

AI Skill Authoring (TDD)

Name: AI Skill Authoring (TDD)
Author: NickCrew

byNickCrew

•

学習とドキュメンテーション

Applies Test-Driven Development principles to create, test, and refine high-performance documentation and skills for Claude.

The writing-skills skill implements a rigorous 'Red-Green-Refactor' workflow for AI prompt engineering and documentation. By requiring users to witness an agent's failure through pressure scenarios before drafting instructions, it ensures that every skill addresses real-world bottlenecks rather than theoretical problems. This methodology eliminates agent rationalization and produces bulletproof documentation that consistently guides Claude through complex tasks while maintaining high reliability across different projects.

主な機能

01Standardized templates for technique, pattern, and reference skills

027 GitHub stars

03Red-Green-Refactor workflow for process documentation

04Loophole detection and rationalization table building

05Search and discovery optimization for skill libraries

06Subagent pressure testing to identify baseline agent failures

ユースケース

01Developing new persistent capabilities for Claude Code and Codex

02Hardening existing documentation against agent hallucination and corner cases

03Standardizing team-wide best practices through validated, tested skill sets

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nickcrew/claude-cortex writing-skills

For use in Claude.ai and ChatGPT

Download Skill