Plain-English definitions of the key terms in the AI agent skills ecosystem — from SKILL.md files to MCP servers to skill stacks.
An Agent Skill (or just "skill") is a structured instruction file that gives an AI agent a defined workflow for a specific task. Instead of prompting the agent from scratch each time, a skill tells it exactly how to approach a problem — what steps to follow, what output to produce, and what quality standards to apply. Skills are repeatable, shareable, and version-controlled.
SKILL.md is the standard filename for an Agent Skill instruction file. It's a markdown file stored in a GitHub repository that contains the skill's instructions, context, and workflow steps. When you install a skill into Claude Code or another agent, the agent reads this file to understand how to perform the task. The .md extension means it's plain text — you can read exactly what the skill does before installing it.
Claude Code is Anthropic's official CLI (command-line interface) for Claude. It lets you interact with Claude directly from your terminal, run it on your codebase, and extend its behavior with Agent Skills. Claude Code is one of the primary platforms that Agent Skills are designed for — skills can be installed into a project's .claude directory and Claude Code picks them up automatically.
Claude.ai is Anthropic's web interface for Claude. Some Agent Skills are designed for Claude.ai's Projects feature, where you can upload a SKILL.md file as a project document and Claude will follow its instructions throughout your conversation. Skills for Claude.ai typically work without any installation — you paste or upload the file directly.
MCP stands for Model Context Protocol — an open standard that lets AI agents connect to external tools and data sources. An MCP server is a separate process that provides Claude (or other agents) with new capabilities: running code, querying databases, fetching live data, or controlling applications. MCPs extend what an agent can do; skills extend how an agent thinks and works. They complement each other. See the guide: Agent Skills vs MCP Servers.
A Skill Stack is a curated collection of individual Agent Skills that work together for a specific professional workflow. For example, the Sales Stack combines account research, call prep, outreach drafting, and pipeline review skills into a coherent set. Stacks are opinionated recommendations — they tell you which skills to install together to cover an entire workflow rather than picking skills one by one.
Source-verified means a skill listing links directly to the real GitHub repository and raw SKILL.md file it was drawn from. Every skill on findskills.co is source-verified — you can inspect the exact instruction file, check the commit history, and verify the license before installing anything. This is the opposite of "prompt marketplaces" that sell opaque system prompts with no way to see what you're actually getting.
The install command is the terminal command you run to add a skill to your agent setup. For skills from repos that use the npx skills package, the command typically looks like: npx skills add owner/repo --skill path/to/skill. For skills that don't use that tooling, you install them by copying or downloading the SKILL.md file directly into your .claude/skills/ directory (or equivalent for your agent).
Some skills include reference files — additional documents that the skill uses as context during execution. These might be style guides, compliance checklists, example outputs, or lookup tables. Reference files are stored alongside the SKILL.md in the source repository and are downloaded with the skill when you install it. On findskills.co, skills with reference files show a file explorer tab where you can browse them before installing.
A system prompt is text given to an AI model before the conversation starts that shapes how it responds. Agent Skills work differently — they're typically loaded as context or instructions that the agent reads and follows, rather than as a raw system prompt injection. The distinction matters: system prompts are global and invisible to the user; skills are explicit, readable, and scoped to specific tasks.
Gemini CLI is Google's command-line interface for the Gemini model family. Some Agent Skills listed on findskills.co are compatible with Gemini CLI in addition to Claude Code — these skills are written in a model-agnostic format and work wherever you can provide a system instruction or context file.
OpenAI Codex CLI is OpenAI's terminal-based coding agent. Skills compatible with Codex CLI follow a similar pattern to Claude Code skills — they're instruction files that guide the agent's approach to specific coding or workflow tasks. On findskills.co, skill pages display which agent platforms each skill supports.
🔍
Browse 350+ source-verified Agent Skills by category, audience, or agent platform.
Browse skills →