writing-skills

You test agent skills by applying a TDD process using subagents to run baseline pressure scenarios, capturing rationalizations, and verifying compliance. Write a SKILL.md, run scenarios without it, then add minimal instructions until agents comply.

TDD for agent skills involves writing failing baseline pressure scenarios, adding minimal SKILL.md content to address failures, and iteratively closing loopholes. This RED-GREEN-REFACTOR cycle ensures documentation reliably enforces correct agent behavior.

Write a SKILL.md with YAML frontmatter containing name and description fields. Focus on discovery by optimizing description writing, keyword coverage, and token efficiency to ensure skills are found and loaded only when relevant.

Yes, you can use subagents to execute baseline tests for skill authoring. Run pressure scenarios with subagents to capture rationalizations where agents bypass rules, then re-verify compliance after updating your SKILL.md instructions.

Agents ignore instructions when skill documentation lacks tested enforcement. Use pressure scenarios to capture rationalizations, then iteratively add minimal instructions closing loopholes until the subagent tests confirm strict compliance.

Yes, YAML frontmatter containing the name and description is required in a SKILL.md file. These discovery-focused fields ensure the skill is correctly indexed, found by agents, and loaded only when contextually relevant.