testing-skills-with-subagents

To stress-test agent documentation, you execute RED-GREEN-REFACTOR cycles that capture rationalizations and iteratively refine skills to close loopholes. This requires baseline scenarios to observe failures without the skill, then re-running against the same scenarios to confirm compliance.

The RED-GREEN-REFACTOR cycle for skill validation is a test-driven approach where you run baseline tests without the skill, implement the skill to achieve compliance, and iteratively refactor the documentation to close loopholes. It ensures skills actually improve agent behavior under pressure.

You capture rationalizations during testing by logging exact failures and agent justifications when the skill is challenged. This identifies specific loopholes in the documentation that need iterative refactoring to ensure the skill resists rationalization.

The best way to validate that skills improve agent behavior under pressure is by executing meta-testing guidance with baseline scenarios. You observe agent failures without the skill, then re-run against the same scenarios to confirm compliance and measure improvements.

Yes, you need baseline scenarios to run test-driven development on process documentation. Baseline testing without the skill is required to observe initial agent failures and establish a control group for measuring iterative improvements during refactoring.