testing-skills-with-subagents

Apply TDD to skill documentation by testing under pressure, catching failures, and closing loopholes.

• Test skills using RED-GREEN-REFACTOR cycle
• Run scenarios without skill to observe failures (RED phase)
• Write skill to fix observed failures (GREEN phase)
• Re-test under pressure to ensure compliance (VERIFY GREEN)
• Refactor skill to close new loopholes (REFACTOR phase)

• Run baseline tests without skill to see failures (RED)
• Document rationalizations exactly as they occur
• Use multiple pressures (time, sunk cost, authority, etc.) in scenarios
• Force concrete choices (A/B/C) not open-ended questions
• Ensure skill prevents rationalizations that bypass rules
• Iterate by refactoring to close new loopholes (REFACTOR)

1. 1.Create pressure scenarios with 3+ combined pressures
2. 2.Run without skill to observe failures and rationalizations
3. 3.Write minimal skill to address observed failures
4. 4.Re-run with skill to verify compliance under pressure
5. 5.Identify new rationalizations and update skill to close loopholes
6. 6.Repeat until agent consistently complies under all pressures