bee:testing-skills-with-subagents

Testing AI agent skills for rationalization involves applying Test-Driven Development (TDD) principles using pressure scenarios to verify compliance under realistic, high-pressure conditions. This process systematically documents and addresses agent excuses for bypassing rules.

The TDD cycle for skill development follows a RED-GREEN-REFACTOR sequence. It creates tests using pressure scenarios like time, sunk cost, and authority to ensure AI agent skills remain robust and resistant to rationalization before deployment.

Pressure testing validates agent behavior by developing realistic scenarios that mimic real-world temptations to cut corners. It uses multiple pressures like time and authority to test skill compliance and systematically document rationalizations.

Yes, you can use TDD to verify a coding standards skill by creating pressure scenarios that mimic real-world temptations to cut corners. This ensures the skill effectively prevents violations and remains robust against rationalization.

Rationalization tables should be used during the REFACTOR phase of skill development to systematically document and address agent excuses for bypassing rules. They help bulletproof skills against rationalization after initial pressure testing reveals vulnerabilities.