testing-skills-with-subagents

Testing skills with subagents applies TDD principles to process documentation by running baseline failure scenarios, implementing targeted fixes, then pressure-testing with combined stressors like time constraints and sunk-cost pressures to verify the skill closes loopholes and enforces compliance.

RED-GREEN-REFACTOR is a test-driven approach where you first run baseline tests without the skill to establish failure modes, write the skill to address those failures, then iteratively refine it by capturing agent rationalizations verbatim and implementing countermeasures.

Yes. This skill enforces verification during skill creation or editing by running baseline and pressure scenario testing with subagents, capturing how agents rationalize non-compliance, and iterating fixes until the skill withstands realistic combined pressures.

Pressure scenario testing creates realistic conditions combining time, sunk-cost, and authority pressures to expose how agents rationalize bypassing your skill; you capture these rationalizations verbatim and implement explicit countermeasures into the skill documentation.

Baseline testing runs your skill against standard failure modes to verify basic functionality. Pressure scenario testing applies combined stressors to expose loopholes agents exploit under real-world conditions, ensuring the skill remains bulletproof even when agents face competing priorities.