Are agent skills the same across models?

Largely yes — SKILL.md is an open standard shared by Claude, OpenAI Codex, and Gemini CLI / Antigravity. The main difference is where each tool looks for installed skills.

What's the difference between a skill and an MCP server?

A skill adds instructions (know-how); an MCP server adds tools (actions and data connections). They're complementary — you often use both.

Do I need to code to make an agent skill?

No. A basic skill is just markdown with a short frontmatter. Scripts are optional. A generator like Skill Builder can produce one for you.

Are 'agent skills' the same as an AI skills course?

No. Agent skills are files that extend an AI agent. 'AI skills' in a careers/training context refers to human upskilling — an unrelated meaning.

What Are Agent Skills? A Plain-English Definition of SKILL.md

Q: What is a SKILL.md file?

It's the single required file of an agent skill: a YAML frontmatter block (name + description) followed by markdown instructions. Optional scripts and reference files can sit alongside it in the skill folder.

Agent skills are a simple, open way to give an AI agent reusable know-how: a folder with a SKILL.md file. This page defines the concept, shows how skills differ from prompts, tools, and MCP, and clears up the common mix-ups.

What are agent skills? (the one-sentence definition)

An agent skill is a reusable, self-contained instruction package — a folder with a SKILL.md file — that teaches an AI agent how to perform a specific task. The agent loads the skill only when it's relevant, so you can give an agent dozens of skills without overloading its context.

What is a SKILL.md file? Anatomy of a skill

SKILL.md is the one required file. It has:

YAML frontmatter with a name (lowercase, kebab-case) and a description (what the skill does + when to use it).
A markdown body with the actual instructions.
Optional bundled resources — scripts, reference docs, or templates the skill can point to.

How agent skills work: progressive disclosure

Skills stay cheap because of progressive disclosure. The agent always sees the lightweight name + description of every installed skill. It only pulls the full body into context when a task matches — and only reads bundled resources if the instructions ask for them. This is what lets skills scale where a giant system prompt wouldn't.

Agent skills vs prompts vs tools vs MCP vs subagents

Prompt — one-off text you send now. A skill is reusable and loads itself when relevant.
Tool — an action the agent can call (e.g. run code). A skill is *instructions*, not an action.
MCP server — connects the agent to external tools/data. MCP adds capabilities; a skill adds know-how. They work together.
Subagent — a separate agent with its own context. Heavier; a skill is just a file.
Custom instructions — global preferences. A skill is task-specific and loads on demand.

Is it an open standard? Portability across Claude, Codex, and Gemini

Yes. The SKILL.md format is an open standard, so the same skill can run on Claude (Claude Code), OpenAI Codex, and Gemini CLI / Antigravity, with small differences in where each tool looks for skills. That portability is the whole point — write once, run on any compatible agent. See the tool-specific guides: Claude · Codex.

Which models support agent skills? (and why Gemini "Gems" are not)

Claude, OpenAI Codex, and Gemini CLI / Antigravity all support SKILL.md agent skills. Gemini "Gems" are not agent skills — a Gem is a no-code custom chatbot you build at gemini.google.com, not an installable SKILL.md. Don't confuse the two.

Agent skills vs human "AI skills"

Searching "AI skills" often returns training courses for *people*. That's a different meaning. Agent skills are files that extend an *agent*, not a curriculum for humans. This page is about the former.

Create your first agent skill

Ready to build one? Use the free Skill Builder, grab a copy-paste template, or browse real skills.