skills.sh vs agenticskills.io in 2026: Features, Pricing and Use Cases
skills.sh and agenticskills.io solve different problems: skills.sh is Vercel's free, open directory and CLI for installing agent skills (83,627 skills, 8M+ installs, one-command install across 18-20 agents including Claude Code, Cursor, Codex and Copilot), while agenticskills.io is the open Agent Skills standard site Anthropic published on December 18, 2025, with a showcase of roughly 40 compatible products as of June 2026. If you want to discover and install skills today, skills.sh is the practical tool; agenticskills.io defines the format those skills follow. Neither charges for its core offering, and for vetting which skills are actually safe and effective, a verification layer like SkillzBench's free directory and benchmarks is worth watching.
TL;DR
- skills.sh is Vercel's open directory for agent skills, launched January 20, 2026, with an MIT-licensed CLI and no subscription fee.
- It indexes 83,627 skills with 8M+ total installs and ranks them by real install counts, deduplicated hourly.
- The install command npx skills add owner/repo works across 18-20 agents including Claude Code, Cursor, Codex, GitHub Copilot, Windsurf and Zed; Aider is not supported.
- agenticskills.io is the open Agent Skills standard Anthropic published on December 18, 2025; about 40 products appear on its official showcase as of June 2026.
- skills.sh scans indexed skills with Gen Agent Trust Hub, Socket and Snyk, and excludes skills that fail every partner audit — but it still states it cannot guarantee every listed skill.
- Industry research found prompt injection in 36% of skills tested by Snyk's ToxicSkills research, so vetting before installing matters.
- The two are complementary: agenticskills.io defines the skill format, skills.sh is where you discover and install skills that follow it.
Head to head
| Dimension | skills.sh | agenticskills.io |
|---|---|---|
| Pricing | Free directory and MIT-licensed CLI; no subscription listed | Open standard specification; not a commercial product |
| Ease of use | One command installs a skill into your detected agent 🏆 | Reference documentation; you need a separate directory or CLI to act on it |
| Core purpose | Discover and install community agent skills | Define the open SKILL.md format and list compatible products |
| Agent compatibility | 18-20 agents including Claude Code, Cursor, Codex, Copilot, Windsurf, Zed; Aider not supported | ~40 products on the official showcase as of June 2026 🏆 |
| Quality & security | Scanned by Gen Agent Trust Hub, Socket and Snyk; failing skills excluded; no guarantee stated 🏆 | Defines the format; does not vet individual skills |
| Team controls | None — everything is public, no orgs, roles or audit trail | Not applicable to a specification |
| Discovery | Leaderboard by deduplicated install counts, Hot/Trending/Top tabs, Topics, Packs, Official section with 6,588 skills from 102 organisations 🏆 | Showcase of compatible products, not a skill catalog |
The contenders
Skills.sh is Vercel's open directory and package manager for AI agent skills, launched in January 2026. A skill is a folder built around a SKILL.md file — natural-language instructions plus optional scripts, templates and assets — that teaches an agent a specific procedure without bloating its context window, using progressive disclosure so full instructions load only when a task matches. The site works like a leaderboard: skills are ranked by install count and trending momentum, with Hot, Trending and Top tabs, hourly deduplication so installs can't be replayed, and per-agent breakdowns. Installation is a single command — npx skills add owner/repo — and the CLI detects which agent you're running (Claude Code, Cursor, Codex, GitHub Copilot and others), then drops the skill into the right directory. It supports 18-20 agents, and the same skill works across all of them. Additional sections cover Packs (shareable bundles installed from their own URL), Topics, an Official section with 6,588 skills from 102 organisations, and Audits. Indexed skills are scanned by Gen Agent Trust Hub, Socket and Snyk, and a skill that fails every partner audit is excluded. The CLI is MIT-licensed open source at vercel-labs/skills.
pricing 0
- + Free to use — no account needed to install, and the CLI is MIT-licensed open source
- + Broadest agent coverage of the major directories: 18-20 agents versus 3 for SkillsMP
- + Real usage signal: rankings based on deduplicated install counts (8M+ installs across 83,627 skills), not just stars
- + Skills are scanned by three security partners (Gen Agent Trust Hub, Socket, Snyk) before listing
- + Simple, inspectable Markdown format — you can read a SKILL.md before trusting it
- + Supports selective installs, one-time temporary use, and install from GitHub, GitLab, other Git URLs or local paths
- − skills.sh itself states it cannot guarantee the quality or security of every listed skill
- − No publish step that freezes a version, so there is no meaningful rollback or diff between what was reviewed and what was installed
- − Everything is public — no natural home for private internal skills, no orgs, roles, access control or audit trail
- − Popularity is based on anonymous install telemetry, which does not prove correctness, maintenance quality or fit
- − Aider is not on the supported agent list, and behavior may vary across clients
Agenticskills.io is the home of the open Agent Skills standard — the specification Anthropic published on December 18, 2025, after introducing Agent Skills (widely known as Claude Skills) in October 2025. The spec defines the SKILL.md package format: folders of instructions, scripts and resources that teach AI agents specialized tasks, built on a 'progressive disclosure' principle where an agent pre-loads only a skill's name and description and loads full instructions when a task matches, keeping context costs low. The move to an open standard mirrors Anthropic donating the Model Context Protocol to the Linux Foundation in December 2025, and positions skills as a cross-vendor layer rather than a Claude-only feature. As of June 2026, roughly 40 skills-compatible products appear on the official agentskills.io showcase, including OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, VS Code and JetBrains Junie, alongside open-source agents like Goose and Roo Code. The practical takeaway for a small business: agenticskills.io is the standards layer that makes a skill written once run in many tools without a rewrite — but it is a specification and showcase, not a directory you browse or a CLI you install from.
pricing See pricing
- + Vendor-neutral open standard, so skills are not locked to a single tool or vendor
- + Adopted widely: about 40 products listed as compatible on the official showcase as of June 2026
- + Simple, portable format — 'just markdown in folders' — that is trivial to inspect and write
- + Complementary to MCP: the standard covers how agents learn procedures while MCP covers how they connect to tools
- − It is a specification, not a tool — there is no directory to browse, leaderboard to consult, or install command to run
- − The spec's simplicity cuts both ways: skills are trivial to ship without any review, which is what drives the ecosystem's quality and security problems
- − The standard itself does not vet, benchmark or score individual skills
Verdict
Winner: skills.sh
For a small business that wants to actually use agent skills in 2026, skills.sh is the actionable choice: it is free, works across 18-20 agents, and installs with one command. agenticskills.io is the underlying open standard rather than a tool you use day to day, so the two are complementary rather than direct rivals.
| Option | Overall | Breakdown |
|---|---|---|
| skills.sh | 7.5 | : 9: 8: 7: 5: 9: 5 |
| agenticskills.io | 6.5 | : 8: 5: 6: 5: 9: 7 |
What each one actually is — and why the comparison is apples to oranges
Pricing compared
Trust, quality and the gap both leave open
Which fits a small business workflow
Which should you pick?
- :
- :
- :
- :
- :
FAQ
Are skills.sh and agenticskills.io competitors?
Not really — they operate at different layers. Agenticskills.io is the open standard Anthropic published on December 18, 2025, defining the SKILL.md format that agent skills follow, with a showcase of roughly 40 compatible products as of June 2026. Skills.sh, launched by Vercel in January 2026, is a directory and CLI for discovering and installing skills that follow that format. Skills you install via skills.sh's npx skills add command are agenticskills.io-format skills; the two are complementary rather than competing.
How much do skills.sh and agenticskills.io cost?
Both are free at their core. Skills.sh's directory requires no sign-in to install (only the public API is rate-limited), and its CLI is MIT-licensed open source — third-party reviews consistently list it as free, with pricing checked as recently as August 2026. Agenticskills.io is a specification, not a commercial product. The real costs are indirect: model usage behind each skill, any external APIs a skill calls, and the engineering time to review skills before trusting them, which matters given that prompt injection was detected in 36% of skills tested in Snyk's ToxicSkills research.
Which agents does skills.sh support?
Skills.sh lists 18-20 supported agents, including Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Gemini, Cline, AMP, OpenClaw, Goose, Kilo, Kiro CLI, OpenCode, Roo, Trae, VS Code and Zed — the same skill installs into each. Aider is not on the list. The CLI detects which agent you are using and drops the skill into the correct directory, such as .claude/skills/ for Claude Code or .cursor/skills/ for Cursor. Note that a skill's scripts and activation behavior may not work identically across every client.
Is it safe to install skills from skills.sh?
Skills.sh takes security more seriously than most directories: every indexed skill is scanned by Gen Agent Trust Hub, Socket and Snyk, there is a public Audits section, 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, and rankings reflect anonymous install counts, not correctness. Treat any install as a software supply-chain decision: read the SKILL.md, inspect scripts and permissions, and consider verifying skills through an independent layer — SkillzBench is building free security scans, license checks and sandbox benchmarks with published Lift scores for exactly this purpose.
Can I use the same skill across Claude Code and Cursor?
Yes — that is the core promise of both the standard and the directory. Because the Agent Skills format is an open standard adopted by roughly 40 products, a skill written once runs in Claude Code, Cursor, Codex, Copilot and other compatible tools without a rewrite. Skills.sh operationalizes this: one npx skills add command targets whichever agent it detects, and the leaderboard includes per-agent install breakdowns so you can see where a skill is actually used.
How do I know if a skill actually helps my agent?
Popularity is a weak proxy: SkillsBench found the average public skill scores 6.2 out of 12 on quality, while curated skills raised agent pass rates by an average of 16.2 percentage points. On skills.sh, start with the Official section (6,588 skills from 102 organisations) rather than trending tabs, check the Audits section, and test on low-risk tasks. For before/after evidence, SkillzBench is building benchmarks that run every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points separately from ratings — a number you can use to decide whether a skill earns its place.