writing-skills

To document agent behaviors as testable SKILL.md entries, write a failing baseline scenario with subagents, draft a minimal SKILL.md implementing the skill, then re-run the baseline to verify agent compliance.

Test-driven development for writing agent skills applies red-green-refactor cycles to skill documentation, mapping baseline failures to skill content and countermeasures across subagents to ensure repeatability and guardrails.

The red-green-refactor cycle applies to skill writing by capturing a baseline failure, drafting minimal skill content to pass it, and refactoring countermeasures to ensure agent compliance under pressure-scenario testing.

Structured SKILL.md templates require Markdown YAML frontmatter and consistent sections for quick discovery, with optional scripts or references placed in scripts/ or references/ directories as needed.

No dependencies are required to start writing test-driven agent skills; the workflow requires only Markdown frontmatter and structured sections, using subagents to run baseline and pressure-scenario compliance tests.

Use combat testing for agent skill documentation when you need to verify compliance under stress, guiding pressure-scenario testing with subagents to ensure the documented behaviors hold up before deployment.