Glossary

The vocabulary of the Agent Skills ecosystem, defined once. Each term has a stable anchor you can link to.

Agent Skill

A reusable instruction file that teaches an AI agent to do one task well. It ships as a SKILL.md file, loads automatically when a request matches its description, and works across Claude Code, Cursor, Codex, and 60+ other agents.

SKILL.md

The open file format an Agent Skill ships in: YAML frontmatter (name, description, optional license) followed by Markdown instructions. Agents read the description to decide relevance, then load the body on demand.

Skill Pack

A curated bundle of related Agent Skills installed together, so one command sets up a whole workflow (for example a release pipeline or a writing toolkit). SkillMD lists official and community packs.

MCP server

A program that exposes live tools and data to an AI agent over the Model Context Protocol. Where a skill adds knowledge, an MCP server adds capabilities; SkillMD runs one so agents can search and install skills directly.

Verified skill

A skill published by an official, vetted first-party source such as Anthropic, Google, Microsoft, or Hugging Face. The verified checkmark marks the publisher's identity; it is separate from the per-skill safety review.

Capability flags

Per-skill labels showing what a skill's instructions can do: docs-only, executes scripts, makes network calls, or reads secrets. They let you judge a skill's blast radius before installing it.

Progressive disclosure

The loading model behind skills: the agent keeps only each skill's name and description in context, and pulls in the full instructions only when a request matches. Many skills stay installed at near-zero context cost.

Skills registry

A searchable index of Agent Skills with install tooling, like a package registry for agent knowledge. SkillMD is an open, free registry where every public skill passes a safety review before listing.

New to all of this? Start with What is an Agent Skill? or the SKILL.md format reference.