testing-skills-with-subagents

To test AI agent compliance and robustness, apply a Test-Driven Development cycle adapted for skills, simulating high-pressure scenarios like time and authority pressures to expose rationalizations and iteratively refactor documentation to plug loopholes.

Skill TDD is an adaptation of the software Test-Driven Development cycle used to rigorously test AI skills. It prevents unintended AI behavior by capturing verbatim agent excuses during simulated pressures and refactoring documentation to enforce adherence to defined principles.

You expose AI agent rationalizations by simulating real-world scenarios that combine time, sunk cost, and authority pressures. This rigorous testing method forces the agent to generate verbatim excuses, revealing loopholes in the skill documentation.

Yes, TDD for AI skills works effectively with coding standards enforcement. You can test an enforcement skill by simulating a developer rushing to meet a tight deadline with a large existing codebase, ensuring it prevents cutting corners.

The best way to close loopholes in AI skill documentation is through an iterative RED-GREEN-REFACTOR cycle. This process captures verbatim agent excuses generated under pressure and refines the documentation until full compliance is achieved.