What it is
anthropics/skills is Anthropic's public repository of Agent Skills: folders with a SKILL.md file plus scripts that Claude loads only when a task needs them. It holds 19 skills, among them docx, pdf, pptx, xlsx, skill-creator, frontend-design, mcp-builder, webapp-testing and claude-api, along with the Agent Skills specification and a template.
In Claude Code you add it as a plugin marketplace (`/plugin marketplace add anthropics/skills`) and install document-skills or example-skills. That matters because Anthropic's pre-built document skills run natively on claude.ai, the Claude API, Claude Platform on AWS and Microsoft Foundry, but not in Claude Code.
Our review
This is less a product than a reference: Anthropic's own working skills, readable in full, which is the most useful thing a skill author can study. The repo has about 183,200 GitHub stars and 21,700 forks. It has no tagged releases; commits go straight to main, most recently on 9 October 2026, and lately they are mostly updates to the claude-api skill.
The licensing is split and easy to miss. The docx, pdf, pptx and xlsx skills, the ones behind Claude's file features, are source-available: their licence ties use to your agreement with Anthropic and bars copying them outside Anthropic's services or making derivative works. The other examples are Apache 2.0. A company that wants its own branded report generator should start from template-skill or brand-guidelines, not from docx.
The design argument is progressive disclosure. According to Anthropic's docs, each installed skill costs about 100 tokens of metadata at startup, the SKILL.md body (under 5,000 tokens) loads only when triggered, and bundled scripts run without their code entering context. Installing the whole example set is therefore cheap until something fires.
Two caveats. The README says the skills are for "demonstration and educational purposes" and that Claude's real behaviour may differ. And the runtime changes by surface: in Claude Code a skill has the same network access as any program on your machine, while on the API it runs in a sandbox with no network. Anthropic's own advice is to install skills only from sources you trust, which this one is; the third-party packs that copy its layout deserve the audit.
What it leaves out is workflow. There is no planning, review or test discipline here, which is why Superpowers and similar packs tend to be installed alongside it rather than instead of it.
Where it earns its place
A developer working in Claude Code
Fill in PDF forms and edit Word contracts from the terminal
Run `/plugin install document-skills@anthropic-agent-skills` and ask for the PDF or Word skill by name. No fee beyond normal Claude usage; the skills' Python scripts run on your machine with your file and network access.
A platform engineer adding spreadsheet export to a product built on the Claude API
Generate formatted xlsx reports from model output
Reference the pre-built `xlsx` skill_id with the code execution tool, and read the repo's xlsx folder to see what the skill does. Note that Agent Skills are not covered by zero data retention arrangements, which rules them out for some regulated data.
An operations lead at a consulting firm
Make every consultant's Claude output follow the house report template and brand rules
Build a custom skill from template-skill and brand-guidelines. On the API it is shared workspace-wide; on claude.ai each user must upload it separately and admins cannot push it centrally, so the API or a Claude Code plugin is easier to roll out.
Before you commit
- Setup
- In Claude Code: `/plugin marketplace add anthropics/skills`, then install document-skills or example-skills; a skill is usable as soon as it is installed. For your own skill, copy the template folder and fill in name and description.
- Integrations
- Claude Code plugin marketplace; the same SKILL.md format works on claude.ai (zip upload under Settings > Features) and the Claude API (/v1/skills), but custom skills do not sync between those surfaces.
- Data and privacy
- Skills run wherever Claude runs them: locally with full network access in Claude Code, in a no-network sandbox on the API. On the API, Agent Skills are excluded from zero data retention. Example skills are Apache 2.0; the four document skills are source-available under Anthropic's terms.
- Cost at scale
- Free to install. The cost is context and compute: about 100 tokens per installed skill at startup and up to 5,000 more when one triggers; on the API, skills also need the code execution tool and are billed under its pricing.
- Limits
- About 1,430 open issues and pull requests, no tagged releases, and a README that disclaims production reliability. claude.ai custom skills are per user only, with no central admin management.
What's good
- Production document skills from Anthropic, readable in full
- Agent Skills spec, template and skill-creator in one place
- About 100 tokens per skill until triggered
- Only route to the pptx, xlsx, docx and pdf skills inside Claude Code
What's not
- Document skills forbid derivative works
- README calls the skills demonstration quality
- No releases or versioning; main changes under you
- No coding workflow
Best for
Anyone writing skills, and Claude Code users who need Office and PDF handling.
Alternatives
Anthropic Skills: common questions
- Is Anthropic Skills worth it?
- Read it before writing a skill of your own, and install it if Claude Code has to handle Office files and PDFs; just don't mistake the document skills for something you can fork. We rate it 4.4 out of 5.
- How much does Anthropic Skills cost?
- Free. Example skills are Apache-2.0; the document skills are source-available, not open source.
- What are the downsides of Anthropic Skills?
- Document skills forbid derivative works; rEADME calls the skills demonstration quality; no releases or versioning; main changes under you; no coding workflow.
- Who is Anthropic Skills best for?
- Anyone writing skills, and Claude Code users who need Office and PDF handling.
- How much does Anthropic Skills cost for a team or in production?
- Free to install. The cost is context and compute: about 100 tokens per installed skill at startup and up to 5,000 more when one triggers; on the API, skills also need the code execution tool and are billed under its pricing.
- Is Anthropic Skills safe for company data?
- Skills run wherever Claude runs them: locally with full network access in Claude Code, in a no-network sandbox on the API. On the API, Agent Skills are excluded from zero data retention. Example skills are Apache 2.0; the four document skills are source-available under Anthropic's terms.
- What are the best Anthropic Skills alternatives?
- The closest alternatives we have reviewed are Claude Code official plugin marketplace, Impeccable and Superpowers.
Answers checked Oct 23, 2026.
Make Anthropic Skills? Show this rating on your site
