Anthropic Skills Repo Review 2026: How It Handles Skill Security and Licensing
The Anthropic Skills Repo (github.com/anthropics/skills) is Anthropic's official, openly published collection of 17 Agent Skills — including the document skills (docx, pdf, pptx, xlsx) that power Claude's file handling — plus example skills like skill-creator and mcp-builder. The repo itself costs nothing to browse, fork, or install, though using the skills inside Claude products requires a Claude plan (Free, Pro at $17/month billed annually, or Max from $100/month). For small businesses, its main strengths are official maintenance and production-grade quality; its main gaps are that licensing varies by skill (the document skills are source-available, not open source) and there is no built-in security scanning or benchmarking of what you install.
TL;DR
- The Anthropic Skills Repo at github.com/anthropics/skills is Anthropic's official, openly published set of 17 Agent Skills, including the production docx, pdf, pptx, and xlsx skills that power document handling in the Claude apps.
- The repo itself is free to browse, fork, and install; running the skills requires a Claude plan — Free ($0), Pro ($17/month billed annually, $20 monthly), or Max (from $100/month).
- Skills use progressive disclosure: Claude sees only a name and description at startup, then loads the full SKILL.md and bundled scripts only when a task matches.
- Licensing is mixed: the four document skills are source-available (not open source) for developer reference, which limits how much you can reuse them in your own products.
- The repo is a reference implementation, not a vetted marketplace — there is no per-skill security scanning, sandbox benchmarking, or trust scoring for the skills you install.
- SkillzBench is building exactly that layer: version-pinning to commit SHAs, security scans, license checks, sandbox Lift benchmarks, and a free MCP connector to load only chosen skills per agent.
- Community skills built on the format (like the code-review skill with ~170 cloners in its first weeks) show the ecosystem is growing fast — and why vetting matters.
Head to head
| Dimension | SkillzBench | Anthropic Skills Repo |
|---|---|---|
| Pricing | Free — directory, MCP connector, benchmarks, and alerts all free, ad-funded 🏆 | Free repo access; Claude Pro $17/mo (annual) or Max from $100/mo to run skills at scale |
| Security & trust | Security scan, commit-SHA pinning, and 0-100 trust score per skill 🏆 | Official Anthropic provenance; no per-skill scanning or trust scoring |
| Licensing clarity | Automated license-check on every listed skill 🏆 | Mixed: document skills are source-available, not open source |
| Skill breadth | Whole-ecosystem crawl of GitHub skills, official and community 🏆 | 17 official skills across five categories |
| Benchmarking | Sandbox tests each skill with the same model, publishing Lift in percentage points 🏆 | None — no with/without skill performance data |
| Agent compatibility | Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents via free MCP 🏆 | Claude Code, Claude apps, Claude API; SKILL.md format read by other agents |
| Setup effort | Toggle skills per agent in loadouts; MCP connector handles delivery | One-line plugin install or copy folders into .claude/skills/ |
| Maturity | Pre-launch; waitlist live, product being built | Live and maintained by Anthropic since late 2025 🏆 |
The contenders
The anthropics/skills GitHub repository is where Anthropic publishes its official Agent Skills in source form. A skill is a folder with a SKILL.md file at its root — YAML frontmatter (name, description) plus instructions, optional scripts, reference docs, and assets — that Claude loads on demand when a task matches the skill's description. The repo splits into two groups: document-skills (docx, pdf, pptx, xlsx) which are the same production skills that power document creation in the Claude apps, and a set of example skills including skill-creator, mcp-builder, webapp-testing, artifacts-builder, brand-guidelines, canvas-design, frontend-design, and internal-comms. Anthropic shipped Agent Skills in late 2025, and the SKILL.md convention has since been adopted by other agents and registries. The repo doubles as a Claude Code plugin marketplace, so skills can be installed with a single /plugin command or copied into a project's .claude/skills/ directory. Its design teaches two patterns worth copying: bundled deterministic scripts for mechanical file transformations, and reference files that keep SKILL.md bodies short.
pricing 0/month
- + Official provenance: maintained by Anthropic and the single best reference for how well-written skills look
- + Free to browse, fork, and install; document skills are the same ones powering Claude's production file handling
- + Well-engineered patterns — scripts over raw generation, reference files keeping SKILL.md short — that transfer to your own skills
- + One-line installation in Claude Code via the plugin marketplace, or copy folders into .claude/skills/
- + Format is spreading beyond Claude, so skills you learn here work with other SKILL.md-reading agents
- − Mixed licensing: the four document skills are source-available, not open source, limiting commercial reuse
- − No per-skill security scanning, sandbox testing, or trust scoring — you install at your own judgment
- − Only 17 official skills; anything beyond that means venturing into unvetted community repos
- − No benchmarking of whether a skill actually improves task outcomes
- − Running skills in Claude products requires a paid plan for meaningful usage (Pro $17/month annual and up)
SkillzBench is AIQ Labs' free, independent directory of AI agent skills plus a free MCP tool that connects any agent to the skills you choose. It exists because agent skills (SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents) have proliferated faster than anyone can vet them — published research found roughly 26% of skills carry at least one vulnerability. SkillzBench is organized around five pillars: Discover (continuously crawling and deduplicating skills from GitHub), Verify (pinning every version to a commit SHA, validating, license-checking, security-scanning, and quality-judging), Benchmark (testing every skill in a sandbox with the same model, with and without the skill, and publishing the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts, so an agent only sees the skills and tools you toggled on), and staying current (tracking upstream changes, re-scanning, re-benchmarking, and notifying users). A 0-100 trust score combines security, quality, license, maintenance, and community signals, with Lift always shown separately and money never touching scores. Everything is free, funded by blog advertising.
pricing 0/month
- + Free across the board — directory, MCP connector, benchmarks, and alerts — with ad-funded, never user-charged revenue
- + Addresses the documented risk that ~26% of skills carry at least one vulnerability
- + Version pinning to commit SHAs means what you tested is what you run
- + Lift benchmarks show whether a skill actually helps, measured with and without it
- + Agent-agnostic: works with Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents
- − Pre-launch: the directory, scanning, benchmarking, and MCP connector are still being built, with only the waitlist live
- − No published benchmark results or live skill counts yet
- − Does not itself publish skills — it vets and delivers skills from across the ecosystem, including Anthropic's
Verdict
Winner: none
The Anthropic Skills Repo is the highest-quality starting point for agent skills because it is official, maintained, and free to access — but it is a raw repository, not a vetted marketplace. Small businesses that want security scans, license checks, and before/after benchmarks on any skill (official or community) will want a verification layer on top.
| Option | Overall | Breakdown |
|---|---|---|
| Anthropic Skills Repo | 8.2 | : 9: 9: 6: 6: 7: 7 |
| SkillzBench | 8 | : 10: 9: 9: 8: 7: 5 |
Skill security: what the repo does and doesn't give you
Licensing explained: open source vs source-available
What the repo costs in practice
How the repo's skills actually work
Which should you pick?
- :
- :
- :
- :
- :
- :
FAQ
Is the Anthropic Skills Repo free to use?
Yes — the repository at github.com/anthropics/skills is free to browse, fork, and install, and the skills themselves carry no charge. The costs come from where you run them: skills are available on every Claude plan tier including Free ($0), but meaningful production usage requires Pro at $17/month billed annually ($20 monthly), Max from $100/month, or Team seats at $20/seat/month billed annually. If you run the skills in other SKILL.md-compatible agents, your cost is that agent's usage instead.
Are Anthropic's skills open source?
Partly. The example skills — skill-creator, mcp-builder, webapp-testing, artifacts-builder, brand-guidelines, and the creative and design skills — are open for reuse and forking. The four document skills (docx, pdf, pptx, xlsx) are source-available rather than open source: you can read them for developer reference, but their license does not grant the same reuse rights. Check each skill's license before building a product on its code.
How secure are skills from the official repo compared to community skills?
Official provenance is a strong trust signal — these skills are maintained by Anthropic and the document skills run in Claude's production apps, giving them real-world scrutiny. But the repo ships no security tooling: no per-skill vulnerability scanning, sandbox testing, or trust scores. That matters because published research found roughly 26% of agent skills carry at least one vulnerability, and skills can bundle executable scripts. Any skill from outside the official repo should be scanned, version-pinned, and tested before an agent runs it.
Can I use Anthropic's skills with agents other than Claude?
Increasingly, yes. Anthropic shipped Agent Skills in late 2025, and the SKILL.md folder convention has since been adopted by other agents and registries. The repo's skills install natively in Claude Code (via the plugin marketplace or .claude/skills/ folders) and run in the Claude apps and API, but the format itself — a SKILL.md file with frontmatter plus optional scripts and references — is what other tools read. SkillzBench's free MCP connector is being built to deliver vetted skills to Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents from one place.
Do I need a paid Claude plan to use skills?
Skills are listed as available on every Claude plan, including the Free plan at $0, so you can try them without paying. In practice, heavier skill usage — especially the document skills that run bundled scripts and code execution — consumes usage quickly, and the Pro plan at $17/month billed annually or Max from $100/month provides the higher limits needed for regular work. Team plans add organization-wide skills deployment for rolling skills out across a company.
How can I tell whether a skill actually helps before installing it?
The Anthropic repo provides no performance data — no with/without comparisons or quality measurements — so you'd evaluate by reading the SKILL.md and trying it manually. SkillzBench is building exactly this measurement: every skill it lists is tested in a sandbox with the same model, with and without the skill, and the resulting Lift is published in percentage points, always shown separately from ratings. Its 0-100 trust score covers security, quality, license, maintenance, and community signals, and money never touches scores.