Jonathon's AI Wiki

anthropics/skills — Canonical Anthropic Skills Repository

8 min read

Source: Anthropics Skills Repo 2026 04 26

Repo: https://github.com/anthropics/skills Stars: 124,272 (at time of ingest 2026-04-26) Forks: 14,541 Open standard pointer: agentskills.io License: Mixed — most skills Apache 2.0; the four document skills (docx, pdf, pptx, xlsx) are source-available (not OSS)

The canonical Anthropic-published skills repository — Anthropic’s reference implementation of Agent Skills alongside the official spec (spec/agent-skills-spec.md) and a starter template (template/). Bundles 17 example skills covering creative, technical, and document workflows. Distributed as a Claude Code plugin marketplace (/plugin marketplace add anthropics/skills) with two installable plugins: document-skills (the four production document tools) and example-skills (the rest). The README explicitly redirects readers to agentskills.io as the home of the open standard, framing this repo as one implementation rather than the standard itself.

Key Takeaways

  • It’s both a marketplace and a reference. Run /plugin marketplace add anthropics/skills to register the marketplace, then install document-skills or example-skills directly. Same source tree doubles as a study reference for skill authors.
  • Two officially shipped Claude Code plugins live here. document-skills@anthropic-agent-skills and example-skills@anthropic-agent-skills. These are the Anthropic-vetted “starter pack” for developers who want to read real skills before authoring their own.
  • The four production document tools are source-available. docx, pdf, pptx, xlsx are the actual skills powering Claude’s document capabilities in the consumer product. Source-available (readable for reference) but not OSS — different license tier from the other 13 skills, which are Apache 2.0.
  • The Agent Skills spec lives in spec/agent-skills-spec.md. This is the formal contract every Anthropic Skill conforms to: YAML frontmatter requirements (name, description), folder layout, scripts/resources conventions. Pointing at spec/ is more reliable than referencing the docs page when authoring a skill that needs to be portable.
  • agentskills.io is the open-standard home. The README leads with “for information about the standard, see agentskills.io” — reinforcing that Skills is a cross-vendor open standard that other implementers (OpenAI, Google, Microsoft per claudeskills.info’s collection counts) target. This repo is Anthropic’s implementation; the spec is the open layer.
  • Same skills are pre-installed on Claude.ai paid plans. Pro/Max/Team/Enterprise users can invoke these without any plugin install — the repo is the “show your work” version.
  • Apache 2.0 (most skills) means commercial reuse is allowed. Fork, embed, customize. Document skills sit under a separate source-available license — read-only reference material, not redistributable.
  • Disclaimer is load-bearing. “Skills are provided for demonstration and educational purposes only … the implementations and behaviors you receive from Claude may differ from what is shown.” Production reliance requires testing in your own environment — even when the skill is officially Anthropic-published.
  • 124k stars, 14.5k forks. One of the largest Anthropic-published repos on GitHub. Active (last push the day of this ingest).

The 17 example skills

SkillCategoryNotes
algorithmic-artCreative & DesignGenerative art with Claude
brand-guidelinesEnterpriseApply brand systems consistently
canvas-designCreative & DesignVisual canvas workflows
claude-apiDevelopment & TechnicalClaude API patterns
doc-coauthoringEnterpriseMulti-author document flow
docxDocument (source-available)Production Word doc tool
frontend-designDevelopment & TechnicalFront-end engineering patterns
internal-commsEnterprise & CommunicationInternal communication templates
mcp-builderDevelopment & TechnicalGenerate MCP servers (essential-mcp-servers)
pdfDocument (source-available)Production PDF tool
pptxDocument (source-available)Production PowerPoint tool
skill-creatorMetaGuide to authoring effective skills
slack-gif-creatorCreative & DesignSlack-friendly animated assets
theme-factoryCreative & DesignGenerate visual themes
web-artifacts-builderDevelopment & TechnicalBuild interactive web artifacts
webapp-testingDevelopment & TechnicalTest web applications
xlsxDocument (source-available)Production Excel tool

How it fits the Claude Code stack

  • vs Agent Skills Overview: The overview article documents Anthropic’s docs page. This article documents the GitHub repo — the actual code, the spec file, the plugin marketplace registration. Different artifacts, both canonical.
  • vs Claude Code Skills Ecosystem: Ecosystem is the map. anthropics/skills is the reference territory — when you want a real SKILL.md to crib from, you read this repo, not the docs page.
  • vs Building Skills Guide: The guide explains how to author a skill. anthropics/skills shows what authored skills look like in production. Pair them.
  • vs Skill Design Patterns: The patterns article extracts five abstract patterns from skill examples. This repo is one of the example sources those patterns are extracted from.
  • vs Plugins and Marketplaces: This repo is a plugin marketplace (the .claude-plugin/ folder registers anthropic-agent-skills with two plugins). When the plugins-and-marketplaces article references “the official registry,” claude-plugins-official and anthropic-agent-skills are both Anthropic-run, but distinct marketplaces.
  • vs Superpowers: Both ship via Claude Code plugins, but Superpowers is a methodology bundle (closed workflow) while anthropics/skills is à-la-carte examples (pick the doc skill you need). Different shapes.
  • vs Claude Skills Hub (claudeskills.info): The Hub aggregates from many publishers including Anthropic. anthropics/skills is one of the collections the Hub features (16 official Anthropic skills card on the Hub vs 17 directories in the actual repo — the Hub may not surface every skill, or may have a slight count drift).
  • vs Corey Haines Marketing Skills / Five Claude Skills Recipe: Those are community / curator skill bundles. anthropics/skills is the publisher-of-Claude’s own bundle. The trust gradient runs Anthropic > publisher > community.

Installation paths

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills    # docx + pdf + pptx + xlsx
/plugin install example-skills@anthropic-agent-skills     # the other 13

After install, mention the skill by name in a prompt: “Use the PDF skill to extract the form fields from path/to/some-file.pdf”. Claude Code routes to the skill automatically.

Claude.ai

These skills are already pre-installed for paid plans. Invoke by name. To upload custom skills, follow Using skills in Claude.

Claude API

Anthropic exposes its pre-built skills, plus custom-skill upload, via the Skills API. Reference: Skills API Quickstart.

Authoring a basic skill (template)

The template ships at template/ in the repo. Minimum required structure:

---
name: my-skill-name
description: A clear description of what this skill does and when to use it
---
 
# My Skill Name
 
[Instructions Claude follows when this skill is active]
 
## Examples
- Example usage 1
- Example usage 2
 
## Guidelines
- Guideline 1
- Guideline 2

Frontmatter requires only two fields:

  • name — lowercase, hyphens for spaces, unique within scope
  • description — complete; this is what Claude sees when deciding whether to invoke the skill

Body is regular Markdown — instructions, examples, guidelines. See Agent Skills Overview for the formal spec field requirements (64-char name limit, 1024-char description limit, etc.).

Partner skills

The README highlights one officially-blessed partner skill set:

The “we may highlight some of them here” framing implies this section will grow. Watch the README for new partner additions.

When to read this repo (vs the docs)

  • You’re authoring your first skill and want a battle-tested example to mimic — read template/ plus 2-3 skills from skills/ that share your domain.
  • You need to see how Claude’s production document tools are structured — read skills/pdf/SKILL.md (source-available reference; treat as reading-only).
  • You’re verifying the formal spec a third-party tool needs to conform to — read spec/agent-skills-spec.md directly, not paraphrased docs.
  • You want to install Anthropic’s complete starter pack into Claude Code in one motion — /plugin install example-skills@anthropic-agent-skills.

Tradeoffs

  • The disclaimer cuts both ways. “Demonstration and educational purposes” means you can’t blindly trust that Anthropic’s deployed Claude.ai behavior matches what you read here — production routing may diverge.
  • Mixed licensing requires care. If you’re forking a document skill into a commercial product, you’re under the source-available license, not Apache 2.0. Confirm per-folder.
  • Two plugins, one repo. The split between document-skills and example-skills means installing “everything” is two /plugin install commands — a small but real friction.
  • Spec lives here, but standard lives elsewhere. If agentskills.io and the in-repo spec ever drift, the standard wins per the README’s framing. Watch for divergence.

Try It

  1. Register the marketplace: /plugin marketplace add anthropics/skills.
  2. Install the example pack: /plugin install example-skills@anthropic-agent-skills. Restart Claude Code.
  3. Trigger one skill by mention: ask Claude to “use the mcp-builder skill to scaffold an MCP server for our internal X API.” Watch how the skill auto-loads and which sub-files it reads.
  4. Open skills/skill-creator/SKILL.md in your editor as a meta-reference. Use it to author your first project-local skill.
  5. Read spec/agent-skills-spec.md once before authoring anything you intend to publish — the spec catches edge cases the docs page glosses over.
  6. Browse skills/pdf/ (source-available) to see what a real-world production skill looks like — the structure, the helper scripts, the failure-mode handling. Don’t redistribute the code; do absorb the patterns.

Open Questions

  • Will the partner-skills section grow into a curated marketplace tier (Anthropic + partner-vetted), or stay as occasional callouts? The “we may highlight” wording is intentionally non-committal.
  • How does the repo’s spec/agent-skills-spec.md synchronize with agentskills.io’s standard? If a vendor implements against agentskills.io and Anthropic’s spec diverges, what’s the precedence?
  • Why is the count between this repo (17 skills) and claudeskills.info’s “Anthropic Official Skills” collection (16 skills) off by one? Probably a counting quirk (template not counted, or one skill excluded), but worth confirming if the gap matters.
  • Is there a release/tag cadence for the production document skills, or are they updated in-place as Claude.ai changes? Source-available means we can read; we can’t necessarily diff for behavior changes.

Graph View

Backlinks