writing-skills

Apply TDD to AI agent skills by authoring SKILL.md files with YAML frontmatter, running failing pressure scenarios to capture agent failures, then iterating until compliance is achieved. This prevents brittle, untested agent behaviors.

A pressure scenario is a test case designed to make an agent fail without the skill's guidance. You capture agent failures verbatim, author the SKILL.md to address them, then re-run the scenario to confirm compliance and resistance to rationalization.

Make agent skills discoverable by including YAML frontmatter with name and description fields, adding keyword-rich discovery terms, and following Claude Search Optimization guidance to ensure skills are findable and token-efficient.

Prevent AI agents from rationalizing by embedding explicit anti-rationalization counters and discipline-enforcing guardrails within the SKILL.md file. These counters address known agent failure modes and enforce compliance during deployment.

A deployment checklist for agent skills should include verification of YAML frontmatter, passing pressure scenarios, anti-rationalization counters, keyword-rich discovery terms, and confirmation that agents comply with documented guidance under pressure.

Yes, you can use TDD to verify existing agent skills by running failing pressure scenarios against them, capturing non-compliance, editing the SKILL.md to address failures, and re-running until agents comply with the updated guidance.