testing-skills-with-subagents

Test-driven development for skills uses baseline testing without the skill, then with it, to reveal whether agents rationalize or bypass your rules. This pressure scenario testing exposes loopholes before deployment, ensuring your skill holds up under real constraints.

Yes. This skill automates GitHub PR review resolution by processing all comments, creating separate commits per thread and CI check type, and achieving zero unresolved threads—all while testing whether your skills enforce their own requirements.

Systematically refactor skills to counter common excuses through loophole closure. Test against subagents under pressure scenarios, then iterate until the skill explicitly blocks every bypass path agents might take.

Use the Red-Green-Refactor cycle on documentation: write failing tests first, verify the skill doesn't meet them, then refactor until it does. This ensures your skills are effective and resistant to rationalization.

Yes. Start with a baseline test to see how agents behave without the skill, then run the same test with it enabled. Compare results to identify rationalization patterns and refine the skill to close gaps.