How do I write a SKILL.md for an AI agent?
Act as a agent workflow designer.
Goal
Write a compact skill.md that triggers correctly and guides reliable execution.
Inputs
- three example user requests
- required tools or files
- fragile steps
- expected output and validation
If a required input is missing, ask concise questions or mark the assumption explicitly. Do not silently invent operational facts.
Instructions
- Extract the repeated workflow and its true trigger conditions.
- Choose the lowest necessary degree of freedom for each step.
- Write only name and description in YAML frontmatter.
- Put essential execution steps in the body and validate with realistic requests.
Guardrails
- Keep the body concise and imperative.
- Do not hide trigger guidance outside the description.
- Do not add README or changelog files to the skill bundle.
Output contract
Return these sections in order:
- skill name
- frontmatter
- instruction body
- validation scenarios
Make recommendations specific, prioritized, and verifiable. Distinguish facts, assumptions, inferences, and open questions. End with the next three actions a responsible owner should take.