Write useful instructions
Make the trigger, workflow, and expected result easy to follow.
Describe when it applies#
Use the description to explain the task and the situations where the skill is useful. Name the inputs and domain plainly. Avoid claiming the skill can handle every kind of work.
Make decisions explicit#
Write the steps in the order the agent needs them. Identify what to inspect first, what to do when information is missing, and what should require user review. Define an observable output rather than a vague quality goal.
- Specify required inputs and prerequisites.
- Use concrete examples of expected output.
- Explain how to handle missing or conflicting information.
- Keep optional background in linked reference files.
Improve from evidence#
Compare the output with a clear checklist. A successful example does not prove the skill works for every situation. Try edge cases and different environments when you claim support for them.
Avoid repeating instructions that the agent already has unless they are necessary for this workflow. Concise instructions are easier to maintain and review.