Anthropic Skills: The Complete Guide to the Official Agent Skills Standard
What if every AI coding assistant — Claude Code, Cursor, GitHub Copilot, VS Code, Gemini CLI, and OpenAI Codex — could share the same skills? Anthropic Skills is the official public repository from the creators of Claude, defining the Agent Skills open standard that has been adopted by 16+ major AI tools. With 85,000+ GitHub stars, 9,000+ forks, and backed by Anthropic itself, this isn't just another config pack — it's the industry-defining specification for how AI agents learn specialized tasks.
Skills are folders of instructions, scripts, and resources that agents load dynamically to improve performance on specialized tasks. From creating branded documents to testing web applications to generating MCP servers, skills teach AI agents how to complete specific tasks in a repeatable, portable way — across any compatible tool.
What Are Agent Skills?
Agent Skills solve a fundamental problem: AI agents are powerful generalists but struggle with specialized, repeatable tasks. You might need Claude to follow your company's brand guidelines when writing documents, analyze data using your organization's specific workflows, or automate personal tasks with precise instructions.
A skill is simply a folder with a SKILL.md file containing YAML frontmatter and markdown instructions:
---
name: my-skill-name
description: A clear description of what this skill does and when to use it
---
# My Skill Name
[Instructions that the agent will follow when this skill is active]
## Examples
- Example usage 1
- Example usage 2
## Guidelines
- Guideline 1
- Guideline 2
The frontmatter requires only two fields:
name— A unique identifier (lowercase, hyphens for spaces)description— A complete description of what the skill does and when to use it
Beyond the basic SKILL.md, skills can optionally include:
scripts/— Executable scripts the agent can runreferences/— Reference documentationassets/— Static assets (images, templates, etc.)
The Open Standard: agentskills.io
Anthropic didn't just create skills for Claude — they published an open standard at agentskills.io. Launched in December 2025, the specification defines a portable, interoperable format that any AI tool can adopt.
Progressive Disclosure
One of the most important design decisions in the specification is progressive disclosure. Instead of loading all skill content upfront (which would waste context window tokens), the system works in three stages:
- Discovery — Only skill names and descriptions are loaded initially
- Activation — Full
SKILL.mdcontent is loaded when a task matches a skill - On-Demand — Scripts, references, and assets are accessed only as needed
This prevents "context rot" — the gradual degradation of agent performance when too much irrelevant context is loaded. It's the same principle behind ECC's strategic compaction, but built into the skill format itself.
Skills vs Tools
The specification draws a clear distinction:
- Tools are external systems (APIs, databases, file systems) an agent can invoke to execute actions
- Skills are procedural knowledge that dictates how an agent should approach problems and use those tools
Think of it this way: an MCP tool gives Claude access to your database, but a skill teaches Claude your organization's data analysis methodology.
Industry Adoption
As of early 2026, the Agent Skills standard has been adopted by 16+ major AI tools:
| Platform | Status |
|---|---|
| Claude Code | ✅ Native (plugin marketplace) |
| Claude.ai | ✅ Native (paid plans) |
| Claude API | ✅ Native (Skills API) |
| Cursor | ✅ Adopted |
| GitHub Copilot | ✅ Adopted |
| VS Code | ✅ Adopted |
| OpenAI Codex | ✅ Adopted (structurally identical) |
| Gemini CLI | ✅ Adopted |
| Amp | ✅ Adopted |
| Letta | ✅ Adopted |
| OpenCode | ✅ Adopted |
| RooCode | ✅ Adopted |
OpenAI's adoption is particularly significant — they implemented a "structurally identical architecture," validating Anthropic's design decisions. This means a skill you create for Claude Code will work in GitHub Copilot, Cursor, Gemini CLI, and beyond.
What's Inside the Repository
The repository contains 17 official skills organized into four categories:
Creative & Design Skills
| Skill | Description |
|---|---|
| algorithmic-art | Generate algorithmic art and creative visuals |
| canvas-design | Design layouts and visual compositions |
| frontend-design | Frontend UI/UX design patterns and implementation |
| slack-gif-creator | Create animated GIFs for Slack communications |
| theme-factory | Generate consistent design themes and color systems |
| web-artifacts-builder | Build interactive web artifacts and components |
Development & Technical Skills
| Skill | Description |
|---|---|
| claude-api | Best practices for using the Claude API |
| mcp-builder | Generate MCP (Model Context Protocol) servers |
| webapp-testing | Test web applications with structured methodologies |
| skill-creator | Meta-skill: create new skills from scratch |
Enterprise & Communication Skills
| Skill | Description |
|---|---|
| brand-guidelines | Apply brand guidelines to communications |
| doc-coauthoring | Collaborative document creation workflows |
| internal-comms | Internal communications and messaging |
Document Skills (Production)
These are the actual skills that power Claude's document capabilities in production:
| Skill | Description |
|---|---|
| docx | Create and edit Word documents |
| Create and process PDF files | |
| pptx | Create PowerPoint presentations |
| xlsx | Create and manipulate Excel spreadsheets |
Note: The document skills are source-available (not open source), shared as reference implementations for complex production skills.
How to Use Skills
In Claude Code
Register the repository as a plugin marketplace:
/plugin marketplace add anthropics/skills
Then install skills:
# Install document skills
/plugin install document-skills@anthropic-agent-skills
# Install example skills
/plugin install example-skills@anthropic-agent-skills
After installation, just mention the skill naturally: "Use the PDF skill to extract the form fields from path/to/some-file.pdf"
In Claude.ai
All example skills are already available to paid plans. To upload custom skills, follow Using Skills in Claude.
Via the Claude API
Use pre-built skills or upload custom ones via the Skills API Quickstart.
Creating Custom Skills
Creating a skill is straightforward:
Step 1: Create the Folder Structure
my-custom-skill/
├── SKILL.md # Required: instructions and metadata
├── scripts/ # Optional: executable scripts
├── references/ # Optional: reference documentation
└── assets/ # Optional: static assets
Step 2: Write SKILL.md
---
name: code-review-standards
description: Enforces your team's code review standards, including naming conventions, error handling patterns, and documentation requirements.
---
# Code Review Standards
## When to Apply
Apply these standards when reviewing any Pull Request or code change.
## Naming Conventions
- Use camelCase for variables and functions
- Use PascalCase for classes and interfaces
- Prefix private methods with underscore
## Error Handling
- Always use custom error classes
- Log errors with structured metadata
- Never swallow exceptions silently
## Documentation
- All public functions need JSDoc comments
- Include @param and @returns annotations
- Add usage examples for complex functions
Step 3: Install and Use
Add the skill to your Claude Code configuration, upload to Claude.ai, or include via the API. The agent will automatically activate the skill when it detects a relevant task.
Partner Skills
As the ecosystem grows, partners are creating official skills for their platforms:
- Notion — Notion Skills for Claude: teaches Claude how to work with Notion's API and data structures effectively
More partner skills are expected as the standard gains adoption.
Anthropic Skills vs Alternatives
Each approach to agent configuration has genuine strengths:
| Feature | Anthropic Skills | Everything Claude Code | Antigravity Awesome Skills | Custom CLAUDE.md | ClawHub Marketplace |
|---|---|---|---|---|---|
| Official Standard | ✅ agentskills.io | Community standard | Aggregation | Per-project | Community |
| Platform Adoption | 16+ tools | 4 tools | 6+ tools | Claude Code only | OpenClaw |
| Skill Count | 17 official | 60+ integrated | 1,200+ aggregated | Custom | Growing marketplace |
| Progressive Disclosure | ✅ Built-in | ❌ | ❌ | ❌ | ❌ |
| Production Document Skills | ✅ docx/pdf/pptx/xlsx | ❌ | ❌ | ❌ | ❌ |
| Hook Automations | ❌ | ✅ Full system | ❌ | ❌ | ❌ |
| Security Scanning | ❌ | ✅ AgentShield | ❌ | ❌ | ❌ |
| Continuous Learning | ❌ | ✅ Instincts | ❌ | ❌ | ❌ |
| Backed By | Anthropic | Community (62K ⭐) | Community | Individual | Community |
| Specification | ✅ Full spec | README | Index | ❌ | ❌ |
| API Support | ✅ Skills API | CLI only | ❌ | ❌ | ❌ |
| Ease of Creation | ✅ Very simple | Complex | Varies | ✅ Simple | Varies |
When to choose Anthropic Skills: You want the official, industry-standard format for portable skills that work across 16+ AI tools — especially if you build skills for teams, publish for the community, or need production document capabilities.
When to choose Everything Claude Code: You need a complete performance optimization system with hooks, security scanning, continuous learning, and 60+ integrated skills — particularly if you want an opinionated, all-in-one setup for Claude Code.
When to choose Antigravity Awesome Skills: You want the largest aggregated catalog (1,200+ skills) from multiple providers across 6+ tools, with freedom to mix and match.
When to choose Custom CLAUDE.md: You want simple, per-project instructions without the overhead of a skill format — great for quick, one-off customizations.
When to choose ClawHub Marketplace: You use OpenClaw and want to discover and install community-created skills through a marketplace interface.
The Agent Skills Specification
The full specification at agentskills.io defines:
Required Fields
name— Unique skill identifierdescription— What the skill does and when to use it
Optional Directories
scripts/— Executable code the agent can runreferences/— Documentation and contextassets/— Static files (images, templates, data)
Key Design Principles
- Simplicity — A skill can be as simple as a single
SKILL.mdfile - Progressive disclosure — Context is loaded on-demand, not all at once
- Portability — Write once, use across any compatible AI tool
- Self-contained — Each skill is a complete, independent unit
- Composability — Multiple skills can be active simultaneously
FAQ
What are Agent Skills?
Skills are folders of instructions, scripts, and resources that AI agents load dynamically to improve performance on specialized tasks. They teach agents how to complete specific tasks in a repeatable, portable way.
Who created this?
Anthropic — the company behind Claude. This is the official reference implementation of the Agent Skills standard.
Which AI tools support skills?
16+ tools including Claude Code, Claude.ai, Claude API, Cursor, GitHub Copilot, VS Code, OpenAI Codex, Gemini CLI, Amp, Letta, OpenCode, and RooCode.
How many skills are included?
17 official skills across four categories: Creative & Design (6), Development & Technical (4), Enterprise & Communication (3), and Document Skills (4).
Are the document skills open source?
The document skills (docx, pdf, pptx, xlsx) are source-available but not open source. They're shared as reference implementations. All other skills are Apache 2.0 licensed.
Can I create my own skills?
Yes. A skill requires only a folder with a SKILL.md file. The specification is deliberately simple to encourage community contributions.
How do skills differ from tools (MCP)?
Tools execute actions (call APIs, query databases). Skills provide the expertise and knowledge about how to approach problems and use those tools effectively. They're complementary.
Is the standard truly open?
Yes. The Agent Skills specification is published at agentskills.io and has been adopted by competitors including OpenAI, Google, and Microsoft — making it a genuine industry standard.
Conclusion
Anthropic Skills represents something rare in AI: a standard created by one company that the entire industry has adopted. With 85,000+ stars and 16+ compatible tools, the Agent Skills specification has become the lingua franca for teaching AI agents specialized tasks. The progressive disclosure system, simple folder-based format, and "write once, use everywhere" portability make it practical for everyone from individual developers to enterprise teams.
Whether you're building custom skills for your team's workflows, publishing skills for the community, or simply browsing the 17 official skills for inspiration, Anthropic Skills is the canonical starting point for agent specialization in 2026.
Explore Anthropic Skills on GitHub