What are Agent Skills?

Agent Skills are portable folders for agents: a SKILL.md (required name + description) plus optional scripts/references/assets. The spec is at agentskills.io — not the same thing as the Claude Code product.

How is this different from Claude Code or MCP?

Claude Code is a terminal agent product that can load Skills. MCP is how tools connect. Agent Skills are how you write a triggerable playbook. They stack.

What's the minimum SKILL.md?

The spec requires name and description in frontmatter. name is 1–64 lowercase alphanumerics/hyphens and must match the folder. description decides whether it triggers.

Can I treat the Anthropic examples repo as Apache-2.0?

Not as a blanket. The README says many skills are Apache-2.0, but the docx/pdf/pptx/xlsx document skills are source-available, not OSS. Check that folder's terms.

Agent SkillsOpen spec · Fast-evolvingApache-2.0（规范）

Portable SKILL.md playbooks for agents

Agent Skills are an open spec: a folder plus SKILL.md (name / description) that agents load on demand. Spec at agentskills.io. This hub is the format, not how to install Claude Code.

Updated 2026-09-04. Same-day snapshot of the official repo.

173.9k

GitHub stars

20.6k

Forks

Open spec · Fast-evolving

Status

Apache-2.0（规范）

License

Open GitHub Repo Open the spec Start with Quickstart

How to Write a Minimal Skill

3 steps to a SKILL.md folder per the open spec.

Make a folder

Folder name is the skill name: lowercase, digits, hyphens, 1–64 chars, matching frontmatter name.

mkdir pdf-processing

Write SKILL.md

Required name + description. Steps in the body; details in references/. Keep the main file short.

---
name: pdf-processing
description: Extract text and tables from PDFs. Use when handling PDF files.
---

Check the spec

Layout, fields and progressive disclosure: agentskills.io/specification.

https://agentskills.io/specification

Pitfalls

⚠️

Don't write a Claude Code manual

This hub is the SKILL.md convention. Installing Claude Code and triggering skills belongs on /claude-code and Claude's own docs.

⚠️

A weak description never fires

The spec uses description as the discovery hook. Say when, for whom, and what — not “general assistant”.

⚠️

Example-repo licenses are mixed

Don't treat all of anthropics/skills as Apache-2.0 for a commercial product. The document skills are explicitly not OSS.

Beside MCP & Claude Code

Based on the official spec and repo README. Skills are a format, MCP a tool protocol, Claude Code one product that consumes Skills.

Tool	What it is	License (verified)	Solves	Does not solve
Agent Skills	SKILL.md folder convention	Spec Apache-2.0	Triggerable playbooks for specialized tasks	Not an installable agent product
MCP	Tool-connection protocol	See the MCP hub	Any client to any tool	Doesn't prescribe playbooks
Claude Code	Terminal coding-agent product	Proprietary / closed	Edit repos and run commands	Not the Skills spec itself

Official Directory

Official

Specification (agentskills.io)
Open Agent Skills spec and overview.
Spec repository
Spec source, Apache-2.0.

Docs & Examples

Specification text
Layout, SKILL.md frontmatter, progressive disclosure.
Anthropic examples repo
Claude-side reference skills. Licenses are mixed; docx/pdf/etc. are source-available, not OSS.

Claude platform docs (one implementation)
How Skills work in Claude API / Claude Code / Claude.ai — one implementation, not the spec.
Anthropic engineering post
Official write-up of the Agent Skills shape.

Latest Updates

2026-09-04 · agentskills.io · anthropics/skills
Agent Skills: portable SKILL.md folders
The spec lives at agentskills.io. Anthropic's examples repo is 170k+ stars; the spec repo is Apache-2.0. This hub is the format, not how to use Claude Code.

What Agent Skills Are: the SKILL.md Spec, Not Claude Code

LaoJin's Pick

163.0k ★ · 2026-07

Anthropic Agent Skills（官方）

Anthropic's official Agent Skills repo — the de facto reference set for the SKILL.md format: document processing, design and office automation with official implementations. Agent skills are moving from a Claude-only feature to an industry-wide convention.

FAQ

What are Agent Skills?: Agent Skills are portable folders for agents: a SKILL.md (required name + description) plus optional scripts/references/assets. The spec is at agentskills.io — not the same thing as the Claude Code product.
How is this different from Claude Code or MCP?: Claude Code is a terminal agent product that can load Skills. MCP is how tools connect. Agent Skills are how you write a triggerable playbook. They stack.
What's the minimum SKILL.md?: The spec requires name and description in frontmatter. name is 1–64 lowercase alphanumerics/hyphens and must match the folder. description decides whether it triggers.
Can I treat the Anthropic examples repo as Apache-2.0?: Not as a blanket. The README says many skills are Apache-2.0, but the docx/pdf/pptx/xlsx document skills are source-available, not OSS. Check that folder's terms.

For Enterprise & Dev Teams

Need to deploy a team-level Agent system?

Open-source frameworks are free, but turning a runtime into an enterprise asset requires access control, RBAC sandboxes, and private RAG integrations. Laojin Global provides commercial-grade architecture and deployment support.

Schedule a Tech Call View Solutions

Cite This Hub

We encourage you to capture this hub into your knowledge base, research, or feed it straight to an AI agent under CC BY 4.0.

Raw Markdown Endpoint (For AI Agents)

Text Citation (For Articles/Reports)

Laojin Global. "Agent Skills Hub". Last updated 2026-09-04. Available at: https://www.laojinchuhai.com/en/agent-skills

How this hub stays current

Version and links live in src/content/agent-skills.ts; star snapshots sync via GitHub API.