writing-skills

To prevent agent rationalization, write process documentation using test-driven development by defining pressure scenarios, capturing verbatim failures, and adding explicit negation counters to close loopholes. This structured Skill authoring approach enforces strict compliance by addressing exact observed failures iteratively.

Applying test-driven development to writing agent instructions involves a RED-GREEN-REFACTOR cycle: define pressure scenarios for failures, write minimal documentation to fix them, then close loopholes. This validates instruction reliability against rationalizations before deployment.

Optimizing YAML frontmatter for discoverability requires writing accurate name and description triggers that match user search intent. This ensures reliable Skill triggering and correct behavior activation when users query related tasks without manual invocation.

Test process documentation failures by running an identical baseline scenario without the Skill, then updating the documentation to address the exact verbatim failures observed. This RED phase pressure testing exposes rationalization vulnerabilities before agent deployment.

Use structured Skill authoring guidance when creating or editing process documentation that must resist rationalizations and remain compliant under pressure. This is essential when instructions require reliable behavior enforcement and accurate YAML discoverability triggers.

AI agents bypass process instructions due to "spirit vs letter" workarounds. Counter this by adding explicit negations, foundational principles, and red flags to your documentation, closing loopholes identified during REFACTOR pressure testing.