How does the REFACTOR phase work for AI skills?

During the refactor phase, you capture the specific excuses or rationalizations the AI used to bypass a rule and add explicit negations or 'Red Flags' to the skill to prevent those specific loopholes.

What is a 'pressure scenario' in skill testing?

A pressure scenario is a test case that includes constraints like tight deadlines, sunk cost, or conflicting authority to see if the AI will prioritize the skill over convenience.

Which skills should I avoid testing with this method?

You don't need to test pure reference skills, such as API documentation or syntax guides, that don't have rules to violate or incentives for agents to bypass.

What is the core principle of this skill?

The core principle is that if you haven't watched an agent fail without the skill (the RED phase), you cannot know if the skill actually prevents the right failures.

Skill Testing with Subagents

Name: Skill Testing with Subagents
Author: franroa

byfranroa

0•

Seguridad y Pruebas

Applies Test-Driven Development (TDD) principles to AI skills to ensure they remain effective under pressure and resist rationalization.

Testing Skills with Subagents brings the rigor of software testing to prompt engineering and process documentation. By utilizing a Red-Green-Refactor cycle, this skill allows developers to create 'pressure scenarios'—realistic, high-stakes situations where an AI agent might be tempted to cut corners or ignore protocols. By observing baseline failures and iteratively refining instructions to close rationalization loopholes, users can transform fragile guidelines into bulletproof AI skills that maintain high discipline and compliance in production environments.

Características Principales

01Pressure-testing methodology using time, cost, and authority constraints

02Standardized scenario formats with concrete choice-based validation

03Meta-testing workflows to improve instruction clarity

04Rationalization tracking to identify and plug instruction loopholes

050 GitHub stars

06Red-Green-Refactor framework specifically for process documentation

Casos de Uso

01Hardening skills that enforce strict discipline like TDD or security protocols

02Validating complex instructions where agents might seek 'pragmatic' shortcuts

03Refining documentation-heavy skills to ensure they are followed under tight deadlines

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add franroa/chezmoi testing-skills-with-subagents

For use in Claude.ai and ChatGPT

Download Skill