Agent Skills stable

What it is

A skill is a directory. At its root is SKILL.md with a short YAML header (name, description) and a Markdown body of instructions. Next to it can sit scripts, templates and reference files. Anthropic introduced skills for Claude in October 2025 and published the format as an open standard in December 2025. Within weeks OpenAI, Microsoft, Cursor, JetBrains and Google’s Gemini CLI supported it. Around 40 products do now.

How it works

Skills rely on progressive disclosure:

1. At startup the agent loads only each skill’s name and description.
2. When a task matches, it reads the full SKILL.md.
3. Only if the instructions call for it does it open supporting files or run scripts.

That keeps the context small until a skill is needed, which is the main reason skills beat stuffing everything into the system prompt.

When to use it

• A workflow you repeat: release checklists, a house style, a deploy procedure.
• Something you want to share across tools and teammates without each tool having its own copy.

If the instruction applies always in a repo, it belongs in AGENTS.md or CLAUDE.md instead.

Watch out for

• The description is what triggers the skill. A vague one means it never fires or fires wrongly.
• Skills can run scripts. Treat a third-party skill like a dependency: read it, pin it.