How do I know when a skill is bulletproof?

A skill is bulletproof when the agent chooses the correct option under maximum pressure, cites specific skill sections as justification, and acknowledges its own temptation to fail but follows the rule anyway.

What is an AI rationalization?

A rationalization is an excuse an AI agent makes to bypass a rule, such as claiming 'I am following the spirit of the rule' while violating the letter of it.

What is the Red-Green-Refactor cycle for skills?

It is a process where you first run a test without the skill to watch the agent fail (Red), write the minimal skill needed to pass (Green), and then refine the skill to close loopholes (Refactor).

How do pressure scenarios work?

Pressure scenarios combine multiple stressors like tight deadlines, management pressure, and exhaustion to see if the agent will prioritize speed over your established skill rules.

When should I not use this skill?

Avoid using this for simple reference skills like API documentation or syntax guides that don't contain specific rules for the agent to follow or violate.

Agent Skill Stress Tester

Name: Agent Skill Stress Tester
Author: NeoLabHQ

byNeoLabHQ

•

542

•

セキュリティとテスト

Applies a rigorous Red-Green-Refactor cycle to verify and bulletproof AI agent skills against rationalization and rule-breaking under pressure.

The Agent Skill Stress Tester is a specialized utility designed for context engineers and developers who need to ensure their Claude Code skills are followed reliably. By treating process documentation like software code, it implements a TDD (Test-Driven Development) workflow where you first observe an agent failing a task without the skill, then iteratively refine the skill to close logical loopholes. It is specifically designed to counter common AI 'rationalizations'—such as skipping tests due to deadlines or sunk costs—by using multi-pressure scenarios to ensure the agent remains compliant even in high-stakes environments.

主な機能

01Provides meta-testing prompts to improve documentation clarity based on agent feedback

02Implements a structured Red-Green-Refactor cycle for skill development

03Generates complex pressure scenarios involving time, authority, and sunk costs

04Identifies and documents verbatim AI rationalizations to build counter-rules

05542 GitHub stars

06Includes a comprehensive 'bulletproofing' checklist for final deployment

ユースケース

01Identifying hidden loopholes in complex agent instructions before team deployment

02Hardening compliance-heavy skills like TDD requirements or security protocols

03Debugging why an agent is consistently ignoring specific guardrails or guidelines

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add neolabhq/context-engineering-kit test-skill

For use in Claude.ai and ChatGPT

Download Skill