What is the 'Iron Law' of writing skills?

The Iron Law states: 'No skill without a failing test first.' You must observe an agent failing a task without the skill before writing the documentation to fix it.

What is the difference between a skill and a project convention?

Skills are reusable techniques or patterns that apply broadly across projects, while project-specific conventions belong in a CLAUDE.md file.

Why is refactoring necessary in skill writing?

Refactoring allows you to close loopholes and address agent rationalizations identified during testing without breaking the core compliance established in the 'Green' phase.

How do you test a discipline-enforcing skill?

These are tested using pressure scenarios where subagents are subjected to time constraints, sunk costs, and exhaustion to see if they maintain compliance with the rules.

AI Skill Authoring TDD

Name: AI Skill Authoring TDD
Author: Moonsong-Labs

byMoonsong-Labs

•

Learning & Documentation

Standardizes the creation of robust AI instructions using a rigorous test-driven development workflow.

This skill transforms how process documentation and AI behaviors are authored by applying core Test-Driven Development (TDD) principles to the writing process. Instead of creating passive guides, it guides users through a Red-Green-Refactor cycle: establishing a failing baseline with subagent pressure scenarios, drafting minimal effective instructions, and systematically closing rationalization loopholes. It ensures every skill is verified against actual agent behavior before deployment, resulting in highly reliable, compliant, and production-ready AI capabilities.

Key Features

013 GitHub stars

02Anti-rationalization tables to close agent loopholes

03Pressure-testing framework using subagent scenarios

04Mandatory verification checklists for skill deployment

05Red-Green-Refactor workflow for process documentation

06Standardized metadata and authoring templates

Use Cases

01Creating new domain-specific capabilities for Claude Code

02Standardizing best practices for prompt engineering and skill development

03Hardening team documentation against AI shortcuts and 'spirit vs letter' errors

Key Features

013 GitHub stars

02Anti-rationalization tables to close agent loopholes

03Pressure-testing framework using subagent scenarios

04Mandatory verification checklists for skill deployment

05Red-Green-Refactor workflow for process documentation

06Standardized metadata and authoring templates

Use Cases

01Creating new domain-specific capabilities for Claude Code

02Standardizing best practices for prompt engineering and skill development

03Hardening team documentation against AI shortcuts and 'spirit vs letter' errors