Skip to content
/

Anthropic Skills (Official)

anthropics-skills · anthropics/skills · ★ 141k · last commit 2026-05-19

Primitive shape 17 total
Skills 17
00

Summary

Anthropic Skills (Official) — Summary

Anthropic's official skills repository is the canonical reference implementation and showcase for the Agent Skills standard, shipping 17 skills organized as two installable Claude Code plugin bundles (document-skills and example-skills), plus document-processing skills that power Claude.ai's native document capabilities (Excel, Word, PowerPoint, PDF).

Problem it solves: Demonstrates what high-quality, production-grade skills look like — spanning creative applications (algorithmic art, canvas design), technical tasks (MCP server building, web app testing, skill creation/evals), and enterprise workflows (branding, internal communications) — and provides the spec template for building compatible skills.

Distinctive trait: The only skill repository to ship the skill-creator meta-skill that teaches agents how to build and eval skills using a full iteration loop (draft→test→quantitative-eval→rewrite), plus live source-available production skills that power claude.ai document features.

Target audience: Developers building skills for Claude.ai, Claude Code, or the Skills API who want a reference for quality patterns; Claude.ai paid users who get these skills bundled.

Scope: 17 skills split across document processing (4 production/source-available) and example skills (13 open-source), available via /plugin install in Claude Code, directly via Claude.ai UI, or via the Claude API Skills endpoint.

Differs from seeds: Unlike superpowers or addyosmani/agent-skills (workflow enforcement frameworks), anthropics/skills is a reference gallery and API demo with no SessionStart hook, no commands, and no workflow sequencing — each skill is a standalone capability demonstrator rather than an integrated development methodology.

01

Overview

Anthropic Skills (Official) — Overview

Origin

Official Anthropic repository, authored by Keith Lazuka (Anthropic) and 13 contributors. Stars: 140,976. Pushed most recently 2026-05-19. No explicit license on the repo root — individual skills carry their own licenses; document skills (docx, xlsx, pptx, pdf) are source-available (not open source), while example skills are Apache 2.0.

Purpose

From the README:

"This repository contains skills that demonstrate what's possible with Claude's skills system. These skills range from creative applications (art, music, design) to technical tasks (testing web apps, MCP server generation) to enterprise workflows (communications, branding, etc.)."

"Many skills in this repo are open source (Apache 2.0). We've also included the document creation & editing skills that power Claude's document capabilities... These are source-available, not open source, but we wanted to share these with developers as a reference for more complex skills that are actively used in a production AI application."

The Agent Skills Standard

The spec lives at agentskills.io/specification (linked from spec/agent-skills-spec.md). The standard defines:

  • Skill format: folder + SKILL.md with YAML frontmatter (name, description) + markdown instructions
  • Required frontmatter: name (unique ID, lowercase hyphen) + description (triggers)
  • Optional: metadata, argument-hint, license

Disclaimer Quoted

"These skills are provided for demonstration and educational purposes only. While some of these capabilities may be available in Claude, the implementations and behaviors you receive from Claude may differ from what is shown in these skills. These skills are meant to illustrate patterns and possibilities. Always test skills thoroughly in your own environment before relying on them for critical tasks."

Partner Skills

Notion is highlighted as a partner skill example ("Notion Skills for Claude").

Documentation Structure

The repository also hosts the spec/ folder containing the canonical Agent Skills specification document (which redirects to agentskills.io), and a template/ folder with a single SKILL.md template for creating new skills.

02

Architecture

Anthropic Skills (Official) — Architecture

Distribution

  • Type: Claude Code plugin marketplace (two plugin bundles) + Claude.ai built-in + Skills API
  • Claude Code install:
    /plugin marketplace add anthropics/skills
    /plugin install document-skills@anthropic-agent-skills
    /plugin install example-skills@anthropic-agent-skills
    
  • Claude.ai: All example skills available to paid plans directly in the UI
  • Claude API: Via the Skills API (see docs.claude.com/en/api/skills-guide)

Required Runtime

  • Python (for webapp-testing scripts via Playwright)
  • None for most skills

Repository Structure

anthropics/skills/
├── .claude-plugin/
│   └── marketplace.json         # Defines 3 plugin bundles
├── skills/                      # 17 skill directories
│   ├── algorithmic-art/
│   ├── brand-guidelines/
│   ├── canvas-design/
│   ├── claude-api/
│   ├── doc-coauthoring/
│   ├── docx/                    # Source-available (production)
│   ├── frontend-design/
│   ├── internal-comms/
│   ├── mcp-builder/
│   ├── pdf/                     # Source-available (production)
│   ├── pptx/                    # Source-available (production)
│   ├── skill-creator/
│   ├── slack-gif-creator/
│   ├── theme-factory/
│   ├── web-artifacts-builder/
│   ├── webapp-testing/
│   └── xlsx/                    # Source-available (production)
├── spec/
│   └── agent-skills-spec.md     # Points to agentskills.io
└── template/
    └── SKILL.md                 # Canonical skill template

Plugin Bundles (marketplace.json)

Three bundles:

  1. document-skills: xlsx, docx, pptx, pdf (source-available, production)
  2. example-skills: algorithmic-art, brand-guidelines, canvas-design, doc-coauthoring, frontend-design, internal-comms, mcp-builder, skill-creator, slack-gif-creator, theme-factory, web-artifacts-builder, webapp-testing (12 skills)
  3. claude-api: claude-api (1 skill)

Skill File Format

YAML frontmatter + markdown:

---
name: <skill-name>           # unique ID, lowercase-hyphen
description: <trigger desc>  # Claude uses this to decide when to activate
license: <optional>
---

Config Files

  • .claude-plugin/marketplace.json — plugin bundles
  • No hooks, no commands, no settings.json hooks
03

Components

Anthropic Skills (Official) — Components

Commands

None.

Skills (17)

Example Skills (Open Source, Apache 2.0)

Name Purpose
algorithmic-art Generate algorithmic/generative artwork using code
brand-guidelines Apply brand voice and design guidelines consistently
canvas-design Create and manipulate canvas-based graphics
claude-api Guidance for building with Claude API/Anthropic SDK
doc-coauthoring Collaborative document writing and editing
frontend-design Build distinctive, polished frontend UI
internal-comms Draft internal communications (announcements, memos)
mcp-builder Build production-quality MCP servers (Python or TypeScript)
skill-creator Create, iterate, and evaluate skills with quantitative benchmarking
slack-gif-creator Generate and send GIFs for Slack
theme-factory Create design themes and color systems
web-artifacts-builder Build interactive web artifacts (HTML/CSS/JS)
webapp-testing Test web apps via Playwright; includes helper scripts

Document Skills (Source-Available, Production)

Name Purpose
docx Create and edit Word documents (.docx)
pdf Read, extract, and process PDF files
pptx Create and edit PowerPoint presentations (.pptx)
xlsx Create, read, and edit Excel spreadsheets (.xlsx)

Subagents

None.

Hooks

None — no hooks.json, no settings.json hook declarations.

MCP Servers

None bundled in the repo. The mcp-builder skill teaches agents how to build MCP servers, but does not ship one.

Templates

1 canonical template: template/SKILL.md

---
name: my-skill-name
description: A clear description of what this skill does and when to use it
---

# My Skill Name

[Add your instructions here that Claude will follow when this skill is active]

## Examples
- Example usage 1

## Guidelines
- Guideline 1

Scripts (within skills)

  • webapp-testing/scripts/with_server.py — Manages server lifecycle for Playwright testing
  • skill-creator/ — Contains eval scripts and description-improver script (referenced in skill but not visible in top-level listing)
05

Prompts

Anthropic Skills (Official) — Prompts

Prompt 1: webapp-testing — Reconnaissance-Then-Action Pattern

File: skills/webapp-testing/SKILL.md Technique: Decision-tree prompt with explicit anti-patterns ("Common Pitfall")

## Decision Tree: Choosing Your Approach

User task → Is it static HTML? ├─ Yes → Read HTML file directly to identify selectors │ ├─ Success → Write Playwright script using selectors │ └─ Fails/Incomplete → Treat as dynamic (below) │ └─ No (dynamic webapp) → Is the server already running? ├─ No → Run: python scripts/with_server.py --help │ Then use the helper + write simplified Playwright script │ └─ Yes → Reconnaissance-then-action: 1. Navigate and wait for networkidle 2. Take screenshot or inspect DOM 3. Identify selectors from rendered state 4. Execute actions with discovered selectors


## Common Pitfall

❌ **Don't** inspect the DOM before waiting for `networkidle` on dynamic apps

Technique analysis: Flowchart-as-decision-tree prompt with an explicit anti-pattern callout using ❌ emoji. The with_server.py --help instruction ("Always run scripts with --help first") trains the agent to use scripts as black boxes rather than reading their source.


Prompt 2: mcp-builder — Phase-Based MCP Development Guide

File: skills/mcp-builder/SKILL.md Technique: Phased implementation guide with URL-loading instructions for live documentation

## 🚀 High-Level Workflow

Creating a high-quality MCP server involves four main phases:

### Phase 1: Deep Research and Planning

#### 1.1 Understand Modern MCP Design

**API Coverage vs. Workflow Tools:**
Balance comprehensive API endpoint coverage with specialized workflow tools. Workflow tools can be more convenient for specific tasks, while comprehensive coverage gives agents flexibility to compose operations.

**Load framework documentation:**

- **TypeScript SDK**: Use WebFetch to load `https://raw.githubusercontent.com/modelcontextprotocol/typescript-sdk/main/README.md`

Technique analysis: Instructs the agent to fetch live documentation URLs during the skill execution — a "dynamic context loading" pattern that keeps the skill fresh without requiring skill updates when SDK docs change.


Prompt 3: skill-creator — Meta-Skill Iteration Loop

File: skills/skill-creator/SKILL.md Technique: Recursive self-improvement prompt with user communication calibration

## Communicating with the user

The skill creator is liable to be used by people across a wide range of familiarity with coding jargon. If you haven't heard (and how could you, it's only very recently that it started), there's a trend now where the power of Claude is inspiring plumbers to open up their terminals, parents and grandparents to google "how to install npm". On the other hand, the bulk of users are probably fairly computer-literate.

So please pay attention to context cues to understand how to phrase your communication! In the default case, just to give you some idea:

- "evaluation" and "benchmark" are borderline, but OK
- for "JSON" and "assertion" you want to see serious cues from the user that they know what those things are before using them without explaining them

Technique analysis: Explicit audience-calibration instruction within a skill — unusually sophisticated social awareness encoding. The "plumbers and grandparents" framing is memorable and concrete.


Prompt 4: webapp-testing — Black Box Scripts Instruction

**Always run scripts with `--help` first** to see usage. DO NOT read the source until you try running the script first and find that a customized solution is absolutely necessary. These scripts can be very large and thus pollute your context window. They exist to be called directly as black-box scripts rather than ingested into your context window.

Technique analysis: Context-window preservation instruction. By banning source-reading of helper scripts, the skill prevents agent context pollution — an unusual meta-instruction about token efficiency built into the skill itself.

09

Uniqueness

Anthropic Skills (Official) — Uniqueness

Differs from Seeds

Fundamentally different from all seeds in intent: anthropics/skills is a reference gallery + API demo, not a workflow enforcement framework. No seed does this.

The closest seed is superpowers (same skill-md format, same Claude Code plugin distribution), but anthropics/skills has no SessionStart hook, no mandatory phases, no TDD enforcement, and no workflow orchestration — it is a collection of standalone capability demonstrators.

The skill-creator skill is closest in spirit to BMAD-METHOD's meta-skills (skills that teach agents how to build more skills), but BMAD's system is much more prescriptive with full persona stacks.

The document skills (docx, xlsx, pptx, pdf) represent something new in the corpus: source-available production code that actively runs in a commercial product (Claude.ai). No other seed ships this hybrid of "educational reference + deployed production code."

Positioning

The authoritative reference for what a well-structured skill looks like per the Agent Skills standard. Its value is:

  1. Being the canonical example that the standard references
  2. Shipping the skill-creator meta-skill for building new skills with evals
  3. Exposing production document skills as reference for complex skill patterns
  4. Bridging Claude Code and Claude.ai as install surfaces

Most Interesting Component

skill-creator — the only skill in any repository in this batch that teaches agents how to create and evaluate other skills, including quantitative benchmarking and description optimization. This is a skills-as-products mindset.

Observable Failure Modes

  1. No cross-tool portability — Claude-only. No Cursor/Gemini CLI/Windsurf support.
  2. No enforcement — skills are suggestions. Without a SessionStart hook or hooks.json, there is no automatic activation guarantee.
  3. Disclaimer gap — the repo explicitly states skills are "for demonstration purposes only" and actual Claude behavior may differ. This creates a trust gap for production use.
  4. Document skills are source-available, not open-source — cannot freely fork or redistribute the most complex production skills.

Explicit Antipatterns

None stated explicitly in the repository itself.

04

Workflow

Anthropic Skills (Official) — Workflow

General Pattern

There is no structured workflow enforced across skills. Each skill is a standalone capability invoked when Claude decides it is relevant based on the skill's description field. There is no session-start hook, no mandatory phase sequencing, and no approval gate system.

skill-creator Workflow (Notable Exception)

The skill-creator skill encodes the only multi-phase iterative workflow in the bundle:

Phase Artifact
1. Capture Intent Requirements notes
2. Interview & Research Clarified requirements + MCP research
3. Draft Skill SKILL.md draft
4. Write Test Prompts Test case set
5. Run Evals (background) Eval results + quantitative metrics
6. Quantitative Eval Draft Benchmark assertions
7. User Review via eval-viewer eval-viewer/generate_review.py output
8. Rewrite Skill Revised SKILL.md
9. Run Description Improver Optimized skill description for better triggering
10. Expand & Re-eval Final validation

This is a full product-quality skill development loop that treats skills as products requiring iteration.

webapp-testing Workflow

Decision tree approach:

User task → Is it static HTML?
    ├─ Yes → Read HTML → identify selectors → Playwright script
    └─ No  → Is server running?
        ├─ No  → with_server.py --help → use helper
        └─ Yes → Reconnaissance (screenshot/DOM) → selectors → actions

Mandates networkidle wait before DOM inspection.

Approval Gates

None at the framework level. The skill-creator workflow includes soft gates (ask user to evaluate results, confirm before rewriting), but these are workflow instructions, not enforced gates.

Phase-to-Artifact Map

Phase Artifact
skill-creator draft SKILL.md
skill-creator eval quantitative benchmark results
webapp-testing Playwright scripts + screenshots
mcp-builder TypeScript or Python MCP server

Spec Format

Not applicable at framework level — individual skills may produce structured output (e.g., MCP server code), but there is no spec artifact format defined.

06

Memory Context

Anthropic Skills (Official) — Memory & Context

State Storage

  • Type: None at the framework level
  • Persistence: Session only (no state files written to disk by the framework itself)
  • Notable exception: webapp-testing produces Playwright screenshots at /tmp/inspect.png; skill-creator may write eval results to disk

Session Context

No SessionStart hook. Skills are activated by Claude's description-matching when a task is submitted — no pre-loaded context injection. Each skill invocation is stateless from the framework's perspective.

Live Documentation Loading

The mcp-builder skill instructs the agent to fetch live URLs during execution:

  • https://raw.githubusercontent.com/modelcontextprotocol/typescript-sdk/main/README.md
  • https://modelcontextprotocol.io/sitemap.xml

This is a dynamic context-loading pattern that substitutes for embedded documentation updates.

Cross-Session Handoff

None — no dedicated mechanism. Any produced artifacts (skill drafts, test results) are left in the working directory.

Context Compaction

No explicit mechanism. The webapp-testing skill includes a specific token-efficiency instruction ("DO NOT read the source... These scripts can be very large and thus pollute your context window") as a workflow-level mitigation.

None.

07

Orchestration

Anthropic Skills (Official) — Orchestration

Multi-Agent Support

No — no subagents, no named personas, no spawn mechanism.

Orchestration Pattern

None — skills are standalone capability demonstrators. There is no pipeline, sequencing, or multi-agent coordination. The skill-creator skill uses background "subagents" in its description (parallel eval runs) but does not define formal agent spawning.

Subagent Definition Format

None.

Isolation Mechanism

None.

Multi-Model

No — single model.

Execution Mode

Event-driven (description-triggered) — Claude activates a skill when the user's task matches the skill's description. No session-level orchestration.

Consensus Mechanism

None.

Prompt Chaining

Weak — the skill-creator cycle has a soft chain (draft → eval → rewrite), but each step requires explicit user interaction; there is no automatic chaining.

Auto-Validators

No — no hooks for automatic test-running, linting, or type-checking.

Git Automation

None.

Execution Mode

Event-driven. Each skill fires when relevant and completes independently.

08

Ui Cli Surface

Anthropic Skills (Official) — UI & CLI Surface

Dedicated CLI Binary

None.

Claude Code Plugin Install

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills

Claude.ai UI

All example skills are available to paid plans on Claude.ai directly. No plugin install needed — they are bundled into the Claude.ai product.

Claude API

Skills can be used and uploaded via the Skills API:

https://docs.claude.com/en/api/skills-guide#creating-a-skill

Local Web Dashboard

None.

IDE Integration

None (no VS Code extension, no Cursor config, no Gemini CLI config). Anthropic's official skills are Claude-only.

Observability

No audit log, no replay. The skill-creator workflow produces eval output files, but this is a workflow artifact, not a framework-level observability feature.

Skills.sh Badge

The repository carries a skills.sh/b/anthropics/skills badge linking to skills.sh/anthropics/skills, suggesting the ecosystem has a skills discovery/registry surface at skills.sh.

Related frameworks

same archetype · same primary tool · same memory type

Superpowers ★ 207k

Enforces spec-first, TDD, and subagent-reviewed development as mandatory automatic workflows rather than optional practices.

OpenAI Skills (Official) ★ 20k

Tiered library of integration and workflow skills for Codex, with a self-bootstrapping skill-installer that lets agents discover…

daymade/claude-code-skills ★ 1.1k

52 production-hardened skills with emphasis on skill craftsmanship — a fork of Anthropic's official skill-creator with security…

Superpowers Lab ★ 344

Experimental skill incubator extending Superpowers with interactive CLI automation, on-demand MCP discovery, semantic…

openspec-skills (chyiiiiiiiiiiii) ★ 6

Ports the OpenSpec proposal/apply/archive SDD lifecycle into three plain SKILL.md files requiring zero runtime dependencies.

Anthropic Skills — IP Guard Delta Report ★ 141k

No ip-guard skill found in the anthropics/skills repo; this entry is a delta report tracking its absence.