testing-skills-with-subagents

Test skills using TDD's RED-GREEN-REFACTOR cycle: run baseline tests without the skill to document failures, implement the skill to address specific failures, pressure-test with combined stressors to verify compliance, then refactor to close rationalization loopholes. This prevents agents from bypassing critical guidance before deployment.

TDD for documentation applies RED-GREEN-REFACTOR to skill creation: RED captures baseline agent failures verbatim, GREEN writes targeted guidance addressing those failures, and REFACTOR explicitly counters observed rationalizations. This bulletproofs skills against agent bypass attempts by closing loopholes systematically.

Combine multiple pressures—time constraints, sunk-cost framing, authority claims—in test scenarios to simulate real compliance challenges. Run the scenario with your skill active and verify the agent follows rules despite pressure, then identify new rationalizations and refactor to plug them.

Test skills before deployment during creation and editing phases to verify they work as intended and resist agent rationalization. Pre-deployment testing catches ineffective guidance early, preventing the deployment of skills that agents can bypass under pressure.

Yes, this Skill requires foundational knowledge of TDD principles and the RED-GREEN-REFACTOR cycle. Familiarity with test-driven development ensures you can map the testing phases correctly and structure quality assurance for your custom skills.

Document exact agent rationalizations during baseline RED testing, implement GREEN guidance addressing those specific failures, then pressure-test to uncover new workarounds. Iteratively refactor by adding explicit counters to each new rationalization until the skill remains compliant under sustained pressure.