Is skills.sh Worth It in 2026? Security, Licensing and Maintenance Compared
Skills.sh is a free, open directory from Vercel that makes installing agent skills across 20+ coding agents effortless with one npx command, but it is a discovery surface with no private team distribution, no version pinning, and no review gate — inclusion is not a security guarantee. SkillzBench (currently in pre-launch) is being built as a free alternative that adds version pinning to commit SHAs, license checks, security scanning and sandboxed benchmarking on top of discovery. If you just want to browse popular public skills, skills.sh works; if you need to know a skill is safe, licensed and actually improves your agent before it touches your business, a verification layer like SkillzBench is the piece skills.sh deliberately leaves to you.
TL;DR
- Skills.sh is Vercel's free, open directory for agent skills, launched January 2026, with an MIT-licensed CLI that installs skills across 20+ agents including Claude Code, Cursor, Codex, GitHub Copilot, Windsurf and Zed.
- Installation is one command — npx skills add owner/repo — with no sign-in required, and the CLI detects your agent and writes to the correct directory.
- Security is partial: every indexed skill is scanned by Gen Agent Trust Hub, Socket and Snyk, and skills failing all partner audits are excluded — but skills.sh itself states inclusion is not a security guarantee, and there is no human review gate.
- Maintenance is the weak spot: installs point at whatever the author's repo currently contains, with no publish step freezing a version, so there is no meaningful rollback and no diff between what you reviewed and what you installed.
- Everything on skills.sh is public — there are no private skills, organizations, member roles or audit trails, which limits its usefulness for standardizing a business team.
- SkillzBench (pre-launch) is being built to fill exactly those gaps: pinning every version to a commit SHA, license-checking, security-scanning, quality-judging, and sandbox-benchmarking every skill with and without it to publish a Lift score.
- SkillzBench is free in full — directory, MCP connector, benchmarks and alerts — funded by advertising that never touches scores.
Head to head
| Dimension | skills.sh | SkillzBench |
|---|---|---|
| Pricing | Free directory and MIT-licensed CLI; no account needed to install | Free everything — directory, MCP connector, benchmarks, alerts |
| Ease of use | One command: npx skills add owner/repo; CLI auto-detects the agent 🏆 | Free MCP server connects any agent to chosen skills; per-agent loadouts |
| Security | Automated partner scans (Gen Agent Trust Hub, Socket, Snyk); inclusion not a guarantee; no review gate | Security scan of every version pinned to a commit SHA; addresses ~26% vulnerability rate in published research 🏆 |
| Licensing | License info surfaced for inspection; no license-check workflow | License check built into verification of every version 🏆 |
| Maintenance & versioning | Installs point at live repos; npx skills update pulls latest; no frozen versions or rollback | Pin to commit SHA; track upstream changes, re-scan, re-benchmark, notify 🏆 |
| Team features | None — everything public; no orgs, roles or audit trail | Per-agent loadouts; agent only sees chosen skills and tools 🏆 |
| Agent compatibility | 20+ agents: Claude Code, Cursor, Codex, Copilot, Windsurf, Zed, Gemini, Cline and more 🏆 | Designed to connect any agent via one free MCP server; supports Claude Code, Codex, Cursor, Gemini CLI and chat-app agents |
| Maturity | Live since January 2026; 83,627 skills, 8M+ installs; Official section with 6,588 skills from 102 organisations 🏆 | Pre-launch; waitlist live, product not yet public |
The contenders
Skills.sh is Vercel's open directory and package manager for AI agent skills, launched in January 2026. Skills are SKILL.md-based packages of procedural knowledge — instructions, scripts and templates — that teach coding agents how to handle specific tasks, and skills.sh lets developers discover them via a leaderboard ranked by install count and trending momentum, then install them with a single command: npx skills add owner/repo. The CLI is MIT-licensed open source, requires no account, auto-detects which of 20+ supported agents you are using (Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Zed, Gemini, Cline and more), and writes the skill to the correct directory for that agent. Every indexed skill is scanned by Gen Agent Trust Hub, Socket and Snyk, and a skill that fails every partner audit is excluded. The directory has grown sections for Packs, Topics, Official and Audits, with the Official section listing 6,588 skills from 102 organisations. It is infrastructure, not a hosted service: free, GitHub-based, and built for discovery rather than governed team distribution.
pricing 0
- + Completely free with no sign-in required to install, and the CLI is MIT-licensed open source
- + Genuinely cross-agent: the same skill installs into 20+ agents, so you can switch tools and keep your skills
- + Frictionless one-command installation that auto-detects the agent and writes to the right directory
- + Automated partner security scanning (Gen Agent Trust Hub, Socket, Snyk) with exclusion of skills that fail every audit
- + Strong discovery signal: install-count leaderboard with hourly deduplication, plus an Official section with 6,588 skills from 102 organisations
- + Simple, inspectable Markdown-based format — no proprietary builder or hosted runtime
- − No review gate: a skill appears automatically once the CLI reports installs, and skills.sh states it cannot guarantee the quality or security of every listed skill
- − No version pinning or rollback — installs point at whatever the author's repo currently contains, so what you reviewed may differ from what you installed
- − Everything is public: no private skills, organizations, member roles, access control or audit trail for teams
- − Popularity is anonymous install telemetry and does not prove correctness, maintenance quality or fit for your project
- − Skills can contain executable scripts and tool permissions, making installation a supply-chain decision rather than harmless prompt copying
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 — currently in pre-launch and actively being built. It exists because agent skills are everywhere but nobody can tell which ones actually help and which are risky: published research found roughly 26% of skills carry at least one vulnerability. SkillzBench's pipeline is designed around five pillars: Discover (continuously crawl and dedupe skills from GitHub), Verify (pin every version to a commit SHA, validate, license-check, security-scan and quality-judge), Benchmark (test every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts, where users toggle skills and tools per agent and the agent only sees what was chosen), and Stay current (track upstream changes, re-scan, re-benchmark and notify). A 0–100 trust score is computed from security, quality, license, maintenance and community signals; Lift is always shown separately from ratings, and money never touches scores. Everything on the platform — the directory, the MCP connector, benchmarks and alerts — is free, with revenue coming from advertising around blog articles that never influences scores or listings.
pricing 0
- + Purpose-built for the security problem: addresses research finding ~26% of skills carry at least one vulnerability, with security scanning on every pinned version
- + License checking built into verification, so licensing risk is surfaced before a skill reaches your agent
- + Version pinning to commit SHAs plus upstream change tracking, re-scanning and re-benchmarking directly solves the maintenance drift problem
- + Sandboxed benchmarking with published Lift separates 'does this skill actually help' from popularity
- + Per-agent loadouts mean an agent only sees the skills and tools explicitly chosen for it — a meaningful permission boundary
- + Completely free, with advertising revenue explicitly firewalled from scores and listings
- − Pre-launch: the product is not yet public, so none of its scanning, benchmarking or MCP capabilities can be used today — only the waitlist/newsletter is live
- − As a newcomer it lacks the install base, ecosystem momentum and third-party validation that skills.sh has accumulated since January 2026
- − Independent verification claims depend on the platform's own scanning and judging pipeline being live and rigorous at launch
Verdict
Winner: none
Skills.sh wins on breadth, maturity and one-command convenience today; SkillzBench is designed to win on security, licensing and maintenance verification, but is pre-launch. They solve different halves of the same problem.
| Option | Overall | Breakdown |
|---|---|---|
| skills.sh | 7 | : 10: 9: 5: 4: 4: 3 |
| SkillzBench | 7 | : 10: 7: 8: 8: 8: 4 |
Security compared: automated scans vs. per-version verification
Licensing and maintenance: the drift problem
Does the skill actually help? Benchmarking vs. install counts
Which should you pick?
- :
- :
- :
- :
- :
FAQ
Is skills.sh free to use?
Yes. Browsing the skills.sh directory and using the MIT-licensed skills CLI have no listed subscription fee, and installing skills requires no account — only the public API is rate-limited. Note that costs can still come from the AI agent, model usage, or external services an installed skill invokes, so 'free directory' does not mean 'free everything downstream.'
Is skills.sh safe? Are its skills security-checked?
Partially. Every indexed skill is scanned by Gen Agent Trust Hub, Socket and Snyk, and a skill that fails every partner audit is excluded from the directory. However, skills.sh itself states it cannot guarantee the quality or security of every listed skill, there is no human review gate, and skills can contain executable scripts and tool permissions — so you should read SKILL.md and inspect every script before installing.
Which AI agents does skills.sh support?
The site lists 20+ agents, including Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Zed, Gemini, Cline, AMP, Goose, OpenCode, Roo, Trae, VS Code and others. The same skill installs into each of them because the CLI writes each agent's native format. Aider is notably not on the supported list, and a skill's instructions or scripts may not behave identically across every client.
What is the difference between skills.sh and SkillzBench?
Skills.sh is a live, free discovery directory: it helps you find and install community skills fast, with automated partner security scans but no version pinning, license checks, review gate or team controls. SkillzBench is a free, pre-launch platform being built to add the verification layer on top: every version pinned to a commit SHA, license-checked, security-scanned, quality-judged and sandbox-benchmarked, with a 0–100 trust score and published Lift, plus a free MCP connector and per-agent loadouts.
Can I use skills.sh for private, internal company skills?
No. Skills.sh is a public directory built around a public leaderboard — everything is public, and there are no organizations, member roles, access controls or audit trails. The CLI itself can install from private repositories, URLs or local paths, but the directory offers no governed team distribution. Teams needing private skills with versioning and review gates need a different kind of registry.
How do I know if an agent skill actually improves my AI agent?
On skills.sh, the honest answer is you test it yourself on low-risk tasks — install counts prove popularity, not effectiveness, and one analysis of 47,150 public skills found an average quality score of just 6.2 out of 12. SkillzBench is being built to answer this systematically: sandboxed benchmarking of every skill with the same model, with and without the skill, publishing the Lift in percentage points so you can see measured impact rather than guessing.