testing-skills-with-subagents

Testing AI skills with subagents applies test-driven development to expose rationalization loopholes. The RED-GREEN-REFACTOR cycle validates skills against real agent behavior, ensuring rules persist even under pressure like time constraints or authority conflicts, making automation reliable and consistent.

Pressure testing combines realistic scenarios—time limits, sunk costs, authority figures, exhaustion—to reveal skill weaknesses. This method identifies where agents bypass rules, then explicitly plugs those loopholes so skills remain bulletproof under maximum stress.

TDD for documentation uses RED-GREEN-REFACTOR to test skill descriptions against actual agent behavior. Write test scenarios first, verify the skill fails without the documented behavior, then refine the skill until it passes—ensuring documentation matches reality.

Yes. Scenario design and verification workflows test subagent-based skills for compliance, audit trails, and consistent behavior. Testing methodology coverage ensures skills resist agent rationalization and perform reliably across automation workflows.

Use scenario design, pressure testing, and RED-GREEN-REFACTOR verification to test subagent skills. Mandatory usage announcements and audit trails enforce compliance. Integration with writing skills consolidates skill development and ensures agent behavior aligns with skill intent.