writing-skills

You test AI agent skills by applying Test-Driven Development principles to create pressure scenarios that expose rationalization loopholes. This methodology helps you develop realistic tests combining time, sunk cost, and authority pressures to ensure agents remain compliant under stress.

The TDD approach for AI skill development adapts the RED-GREEN-REFACTOR cycle to documentation. You write failing pressure test scenarios, create minimal skill documentation to pass those tests, and iteratively refactor to close any compliance loopholes the agent finds.

You document AI agent failures by recording exactly how the agent bypasses rules during pressure testing scenarios. Capturing these specific rationalization attempts allows you to iteratively refactor the skill documentation to explicitly address and close those loopholes.

Yes, you can use TDD to enforce code review standards by writing failing test scenarios first, creating the skill to enforce the standard, and then testing it under combined pressures. This ensures agents do not find loopholes to bypass the review rules.

Rationalization countermeasures involve identifying common excuses agents use to bypass rules and explicitly addressing them in the skill documentation. By testing skills under authority, time, and sunk cost pressures, you can refactor the documentation to close those specific loopholes.