testing-skills-with-subagents

The RED-GREEN-REFACTOR workflow defines a Red baseline without the skill, Green skill development, and Refactor iterations to systematically validate process documentation and ensure bulletproof agent performance.

You can capture agent rationalizations by recording their responses word-for-word during simulated pressure scenarios, using these verbatim failures to guide targeted refactoring and close documentation loopholes.

Meta-testing validates the skill's resilience by applying multiple pressures and confirming improvements, ensuring the documentation skill remains bulletproof across various simulated scenarios before deployment.

The best way to close loopholes is to iterate through Refactor phases using captured verbatim rationalizations, applying meta-testing prompts to validate resilience and ensure agents cannot bypass the documentation rules.