writing-skills

Test agent compliance by running RED-GREEN-REFACTOR style pressure tests on real scenarios where agents may rationalize shortcuts or skip required steps. Observe baseline failures, apply minimal skill revisions targeting exact rationalizations, and iterate until verification shows consistent compliance.

Pressure testing for skill authoring validates that agents follow documentation rules when incentivized to cut corners under time pressure, sunk cost, authority, or exhaustion. You need it before deploying or updating any agent Skill expected to be followed reliably under real conditions.

Close documentation loopholes by creating explicit negations, counters, and red flags for rationalizations agents use to bypass rules. Re-test with pressure scenarios after each revision until compliance remains stable and agents can no longer rationalize shortcuts.

Improve skill discoverability by requiring a rich, trigger-focused third-party description in YAML frontmatter. This discovery-oriented authoring approach ensures the Skill surfaces when relevant conditions arise, rather than relying on the Skill name alone for search visibility.

Yes, you can apply test-driven development to skill authoring by turning documentation writing into RED-GREEN-REFACTOR iterations. Start by observing baseline agent failures, make minimal revisions addressing exact rationalizations, then refactor until pressure-scenario compliance tests pass consistently.

Agents skip required steps under pressure because the documentation lacks explicit negations and counters for rationalizations like time pressure, sunk cost, or authority. Without iterative pressure-scenario testing and loophole closure, agents rationalize shortcuts even when they understand the intended rule set.