testing-skills-with-subagents

Test skills with subagents using RED-GREEN-REFACTOR cycles to verify resilience under combined pressure scenarios (time, sunk cost, authority). This approach identifies rationalization loopholes and ensures skills reliably guide AI behavior even when compliance is tempted.

TDD for documentation applies Test-Driven Development to process documentation by creating pressure-scenario baselines, implementing skills, and refactoring based on failure patterns. This ensures skill clarity, effectiveness, and resistance to AI shortcuts in high-stakes scenarios.

Test new skills immediately after creation to catch flaws, misinterpretations, and compliance gaps before deployment. Subagent testing under realistic pressure scenarios reveals how skills perform when agents face time constraints, sunk-cost reasoning, or authority challenges.

Yes. Automated testing with subagents generates multiple combined pressure scenarios, captures rationalization patterns, codifies countermeasures, and produces failure and compliance reports—enabling systematic verification across RED-GREEN-REFACTOR cycles.

Skills become robust through explicit countermeasures identified during pressure testing. Baseline failure observation reveals common rationalizations; systematic refinement ensures the final skill documentation blocks those loopholes and holds under combined pressure scenarios.

Use pressure-scenario testing to surface rationalization patterns agents use to bypass skill intent. Document these patterns explicitly, then refactor skill wording to block them. Automated compliance reporting validates that countermeasures hold across repeated pressure tests.