Evidence-led guide
What Are Agent Skills?
A practical guide to source files, reusable workflows, compatibility, setup, permissions, and safe evaluation.
Selection criteria
- Clear source and maintainer
- Compatible agent environment
- Understandable setup and dependencies
- Proportionate permissions
- Safe, repeatable example workflow
The short answer
An Agent Skill is a reusable package of instructions and supporting resources that helps an AI agent perform a particular kind of work. A SKILL.md file commonly describes when the Skill applies, how the workflow should run, and which references or scripts may be involved.
A Skill is more than a one-off prompt
A prompt usually asks for something in one conversation. A Skill can encode a repeatable method, route the agent to supporting material, define checks, and bundle scripts or templates. Exact behavior depends on the platform that loads it.
Inspect the source before using it
AI Vitamin indexes original source files and repository facts; a listing is not a security certification. Open the SKILL.md file, read linked files, identify scripts and dependencies, check repository activity, and confirm that your platform supports the instructions.
- Verify the original repository and source path.
- Read setup, dependencies, and compatibility notes.
- Review file, network, account, and execution permissions.
- Test unfamiliar Skills on non-sensitive work first.
When a Skill is useful
Skills are most useful for recurring work that benefits from a stable sequence, specialized references, or consistent quality checks. They are less useful when a request is simple, highly novel, or requires authority the agent should not receive.
Questions people ask
No. A Skill packages guidance and resources for a workflow. An MCP server exposes tools or data through the Model Context Protocol; a Skill may mention one, but the entities are different.
Continue exploring
Move from the decision framework to the relevant directory and adjacent guides.