Guide
How to choose an AI agent skill
Choosing a skill is a product decision at the workflow level. The right package should match the user's task, be clear enough to inspect, and have safety boundaries that fit the environment.
Start with workflow fit
A skill should describe the exact type of work it supports. Broad labels are less useful than concrete use cases, expected inputs, and verification steps.
If the summary could apply to anything, keep searching. A good skill should make the agent's behavior more predictable, not more vague.
Check metadata and risk
Review the title, category, tags, quality score, risk level, package size, and hash. These fields help users compare skills before downloading a zip.
Avoid packages that request unknown credentials, destructive commands, or external services that are not required for the task.
Test in a small scope
After reviewing SKILL.md, test the skill on a small task before trusting it with production files or sensitive context.
If a skill works well, keep it in a stable local skills directory and use the ShowUskills detail page to find related packages for adjacent workflows.