How do rationalization tables help?

Rationalization tables document the exact excuses an AI gives for ignoring a rule. By adding these excuses to the skill with explicit counters, you prevent the AI from using them in the future.

What is a pressure scenario in AI testing?

A pressure scenario is a test case that includes constraints like tight deadlines, exhaustion, or manager overrides to see if the AI will prioritize speed over following the skill's rules.

What does it mean to apply TDD to skills?

It means writing a scenario where the AI is likely to fail (RED), writing a skill to fix that failure (GREEN), and then refining the skill to close any new loopholes or excuses the AI finds (REFACTOR).

Why is the RED phase mandatory?

If you don't watch the agent fail without the skill first, you don't know if your skill is actually preventing the right failures or if the AI was already behaving correctly by default.

Skill Testing with Subagents

Name: Skill Testing with Subagents
Author: apexbusiness-systems

byapexbusiness-systems

0•

Security & Testing

Applies the RED-GREEN-REFACTOR cycle to process documentation to ensure AI agents follow instructions under high-pressure scenarios.

Testing Skills with Subagents brings the rigor of Test-Driven Development (TDD) to the creation of custom instructions and skills. This framework ensures that your documentation isn't just theoretical by forcing the AI to confront realistic, high-pressure scenarios—such as tight deadlines or sunk costs—where it might be tempted to cut corners. By documenting baseline failures (RED), writing minimal skill updates (GREEN), and systematically plugging loopholes (REFACTOR), users can create bulletproof skills that prevent AI rationalization and maintain discipline in complex workflows.

Key Features

010 GitHub stars

02High-pressure scenario generation (Time, Sunk Cost, Authority)

03Rationalization tracking to document and neutralize AI excuses

04RED-GREEN-REFACTOR framework for process documentation

05Meta-testing methodology for refining instruction clarity

06Template-based pressure testing for compliance verification

Use Cases

01Testing complex workflow instructions for loopholes before team deployment

02Verifying that AI agents adhere to strict TDD protocols under deadline pressure

03Hardening security and compliance skills against 'pragmatic' rationalizations

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add apexbusiness-systems/apex-omnihub testing-skills-with-subagents

For use in Claude.ai and ChatGPT

Download Skill