What is Red-Green-Refactor for Claude skills?

It is a methodology where you first observe an agent failing a task without the skill (Red), write the minimal skill content to fix the failure (Green), and then refine the skill to close any logic loopholes or new excuses the agent generates (Refactor).

How do I stop an AI from rationalizing away rules?

By documenting the exact excuses (rationalizations) the AI makes during testing and adding explicit negations, such as 'No exceptions for pragmatism' or 'Do not keep code as reference,' directly into the skill documentation.

Why does this skill use pressure scenarios?

AI models often follow rules in simple academic tests but may bypass them in real-world situations. Pressure scenarios involving deadlines or 'sunk costs' reveal if a skill is actually bulletproof.

Is this skill for testing software or testing prompts?

This skill is specifically designed for testing 'process documentation' and custom Claude skills to ensure the AI follows specific behavioral patterns and constraints during development.

Subagent Skill Testing

Name: Subagent Skill Testing
Author: yusoofsh

byyusoofsh

セキュリティとテスト

Validates and bulletproofs Claude skills using a Red-Green-Refactor cycle and high-pressure subagent scenarios.

概要

Subagent Skill Testing brings the rigor of Test-Driven Development (TDD) to AI process documentation and skill creation. By simulating high-pressure scenarios—such as tight deadlines, sunk costs, or conflicting authority—this skill helps developers identify where Claude might rationalize ignoring rules. It guides users through a baseline failure test (RED), drafting targeted skill content (GREEN), and iterating to close logic loopholes (REFACTOR), ensuring that custom skills remain compliant and effective even in complex real-world situations.

主な機能

RED-GREEN-REFACTOR workflow for prompt engineering
Baseline testing to verify natural model behavior before skill application
Rationalization tables to identify and neutralize AI excuses
0 GitHub stars
High-pressure scenario generation (time, cost, and authority constraints)
Meta-testing framework for auditing skill clarity and instruction following

ユースケース

Refining complex multi-step workflow instructions to prevent corner-cutting
Verifying safety and compliance skills before production deployment
Hardening internal coding standards like mandatory TDD or linting rules

概要

主な機能

RED-GREEN-REFACTOR workflow for prompt engineering
Baseline testing to verify natural model behavior before skill application
Rationalization tables to identify and neutralize AI excuses
0 GitHub stars
High-pressure scenario generation (time, cost, and authority constraints)
Meta-testing framework for auditing skill clarity and instruction following

ユースケース

Refining complex multi-step workflow instructions to prevent corner-cutting
Verifying safety and compliance skills before production deployment
Hardening internal coding standards like mandatory TDD or linting rules