writing-skills

Test-driven development for skills follows the RED-GREEN-REFACTOR cycle: first run a failing test with an agent lacking the skill, observe failures, write the skill to address those failures, then refactor to close loopholes. This ensures skills are robust and resistant to agent rationalization before deployment.

Rationalization bulletproofing explicitly identifies and counters common ways agents justify incorrect behavior or skip steps. By enforcing discipline through pressure scenarios and red-flag checks during skill authoring, you prevent agents from finding workarounds and ensure skills work as intended.

Verify skills by running baseline and passing test artifacts that validate frontmatter (name, description), structured sections, explicit rationalization counters, and red-flag checks. Subagent pressure scenarios serve as test cases to confirm the skill handles edge cases and pressure situations correctly.

Deploy-ready skills require a complete TDD workflow: define pressure scenarios as failing tests, write the skill to pass those tests, refactor to close new loopholes, and validate all test artifacts pass. This pressure-testing approach with subagents ensures the skill is production-ready.

Yes, the TDD workflow applies to both new and edited skills. Run pressure scenarios against the current skill to identify failures, modify the skill to address those failures, then refactor and re-test to ensure no new loopholes emerge from your changes.

CSO guidelines in the skill-writing process ensure skills are token-efficient and easily discoverable through search. Applying CSO during authoring makes skills more findable while reducing unnecessary token usage in documentation and frontmatter.