What is an Agent Skill?
A skill is packaged know-how for Claude: a SKILL.md of instructions plus optional scripts and files.
An Agent Skill is a directory centered on a SKILL.md file. That file has YAML frontmatter (name, description, allowed tools) and a body of plain-language instructions. Alongside it can sit bundled files: Python scripts, templates, reference docs.
Claude loads a skill on demand: when a task matches the skill’s description, its instructions enter the context and its scripts become runnable. This “progressive disclosure” keeps context clean until the skill is actually needed.
A skill is not an MCP
They are different and complementary. An MCP server gives Claude access to an external system (Gmail, Postgres, GitHub) via a running process. A skill gives Claude the know-how to perform a task and is just files in ~/.claude/skills. They are often used together: the MCP wires up the data, the skill encodes how to work with it.
How to install
Every skill page in the catalog has install tabs: an npx degit command to copy the folder into Claude Code, a manual option, and a plugin install via /plugin. Personal skills live in ~/.claude/skills; project skills in a repo’s .claude/skills.
Security
A skill runs instructions and may execute bundled scripts — treat it like code. Install from sources you trust. Unyly flags on every page whether a skill bundles scripts and which tools it is allowed to call.