A skill for reviewing a document might include an editing workflow, examples, and a script for reading the file. A host can discover the skill from its short description, then load detailed instructions when relevant. The open Agent Skills format organizes these packages around a SKILL.md file.
A skill supplies procedural context. A tool supplies an interface for an action or lookup, and MCP can expose such interfaces. A skill can explain when and how to use tools without being interchangeable with them.
In episode 72, at 3:03, AMD’s Anush Elangovan describes skills in AMD’s ROCm installation and model-serving workflow. This is Elangovan’s account of the integration. The episode names Svix and Walrus Memory as the season’s presenting sponsors in its opening sponsor read.
Our practical check for any installed skill is to review both the instructions and bundled code, then test whether it improves representative tasks; installing a package is not evidence of quality or safe behavior.
Sources
- Anthropic: Equipping agents with Agent Skills — Instructions, progressive disclosure and bundled-code risks; checked October 7, 2026.
- Agent Skills: Open format overview — Defines SKILL.md metadata, instructions and optional scripts, references and assets; checked October 7, 2026.
