writing-skills

To prevent AI agent noncompliance, apply test-driven development to skill documentation by creating fail-first pressure scenarios that capture rationalizations. You then write minimal targeted rules to close loopholes and re-verify until the agent complies under time, authority, or sunk-cost pressure.

Test-driven documentation applies RED-GREEN-REFACTOR principles to skill authoring. It defines fail-first baseline tests combining time, sunk cost, and authority pressures to observe agent rationalizations, then translates those observations into minimal intervention instructions.

Pressure test agent instructions by defining fail-first baseline scenarios that combine time, sunk cost, authority, and exhaustion pressures. This observes how agents attempt to rationalize rule violations, allowing you to iterate and refactor explicit loopholes before deployment.

Write reliable skill instructions by translating observed rationalizations into minimal, targeted counters within the SKILL.md file. Iterate using loophole refactoring and re-verification until the skill remains compliant under new rationalizations and discovery-optimized standards.

Test-driven writing optimizes Claude skill discovery by enforcing concise SKILL.md metadata structure and format standards. This ensures reliable selection by search mechanisms when authoring, editing, or deploying discipline-enforcing skills.

Use anti-rationalization techniques when authoring, editing, or deploying discipline-enforcing skills where agents may rationalize violations under time, authority, or sunk-cost pressure. It ensures fail-first baseline testing and explicit loophole refactoring for sustained compliance.