What are the core skill types supported?

The framework categorizes skills into Technique (recognition/application), Pattern (usage constraints), and Reference (information retrieval).

What is the TDD approach for Claude skills?

It follows a RED-GREEN-REFACTOR cycle where you first document an agent's failure (RED), write the minimal skill to fix it (GREEN), and then refine it by finding new loopholes (REFACTOR).

How do I know when to create a new skill versus updating CLAUDE.md?

Create a skill if the technique applies broadly across different projects; use CLAUDE.md for rules that are specific only to the current repository.

Why shouldn't I include workflow steps in a skill description?

Including workflows in the description creates a shortcut that agents may take, causing them to skip the detailed logic contained in the body of the skill.

Claude Skill Development & TDD

Name: Claude Skill Development & TDD
Author: ReinaMacCredy

byReinaMacCredy

•

Learning & Documentation

Implements a test-driven development framework for creating, validating, and refactoring specialized AI skills.

About

The Writing Skills skill provides a rigorous methodology for developing high-quality Claude Code capabilities by treating AI instructions as testable code. It enforces a strict RED-GREEN-REFACTOR cycle, ensuring that every new skill is preceded by a documented failure case to prove its necessity. By focusing on 'When to Use' triggers and avoiding common anti-patterns like narrative descriptions or workflow shortcuts, this skill helps developers build a library of reliable, reusable, and loophole-free AI behaviors that improve agent performance across complex tasks.

Key Features

Loophole-plugging strategies through pressure testing
Anti-pattern detection for narratives and multi-language maintenance
Strict frontmatter validation for 'When to Use' triggers
RED-GREEN-REFACTOR workflow for instruction engineering
Standardized skill categorization (Technique, Pattern, Reference)
20 GitHub stars

Use Cases

Standardizing documentation practices across a multi-agent development team
Validating AI agent performance through documented failure scenarios
Building a new library of domain-specific instructions for Claude Code

About

Key Features

Loophole-plugging strategies through pressure testing
Anti-pattern detection for narratives and multi-language maintenance
Strict frontmatter validation for 'When to Use' triggers
RED-GREEN-REFACTOR workflow for instruction engineering
Standardized skill categorization (Technique, Pattern, Reference)
20 GitHub stars

Use Cases

Standardizing documentation practices across a multi-agent development team
Validating AI agent performance through documented failure scenarios
Building a new library of domain-specific instructions for Claude Code