Why apply TDD to documentation?

TDD ensures that documentation is not just 'good advice' but a functional solution to a documented failure mode, preventing agents from ignoring rules under pressure.

How does the Refactor phase work for skills?

In the Refactor phase, you attempt to find new excuses or loopholes the agent might use to circumvent the new skill, then update the skill to close those gaps.

Can I use this for project-specific conventions?

This skill is best for broad, reusable patterns. For project-specific conventions, it is recommended to use a CLAUDE.md file instead.

What is the 'writing-skills' capability?

It is a specialized framework for Claude Code that treats the creation of agent instructions like software development, using testing to verify effectiveness.

What is a pressure scenario in this context?

A pressure scenario is a prompt given to a subagent that includes constraints like tight deadlines or high sunk costs to see if the agent will attempt to bypass standard procedures.

Agentic Skill Writing TDD

Name: Agentic Skill Writing TDD
Author: withzombies

bywithzombies

•

学習とドキュメンテーション

Applies Test-Driven Development principles to the creation of Claude Code skills to ensure instructions are robust and loophole-free.

The writing-skills capability transforms process documentation into a rigorous engineering task by applying the Red-Green-Refactor cycle to Claude Code skills. Instead of guessing what instructions an agent needs, this skill mandates creating 'pressure scenarios' with subagents to document baseline failures before writing a single line of guidance. This ensures that every rule added to a skill addresses a verified failure mode, resulting in bulletproof instructions that resist AI rationalizations, time-pressure excuses, and edge-case shortcuts.

主な機能

01Red-Green-Refactor cycle for process documentation

02Standardized XML schema for skill definitions

0358 GitHub stars

04Pressure testing via automated subagent scenarios

05Requirement for baseline failure documentation before skill creation

06Identification and mitigation of AI rationalization patterns

ユースケース

01Developing new specialized Claude Code skills that require high reliability

02Hardening existing documentation against agent shortcuts and 'laziness'

03Standardizing team-wide coding conventions with verifiable compliance

主な機能

01Red-Green-Refactor cycle for process documentation

02Standardized XML schema for skill definitions

0358 GitHub stars

04Pressure testing via automated subagent scenarios

05Requirement for baseline failure documentation before skill creation

06Identification and mitigation of AI rationalization patterns

ユースケース

01Developing new specialized Claude Code skills that require high reliability

02Hardening existing documentation against agent shortcuts and 'laziness'

03Standardizing team-wide coding conventions with verifiable compliance