
skills.sh
skills.sh is Vercel's open directory of agent skills — reusable, installable capabilities that give AI coding agents procedural knowledge. Browse a live install leaderboard, check partner security audits, then add any skill with a single command.
What is skills.sh?
skills.sh is an open directory of agent skills — reusable capabilities that hand an AI coding agent the procedural knowledge it needs for a specific job. A skill lives in a public GitHub repository as a SKILL.md file, and the directory indexes every public skill that ships through the open-source skills CLI. Vercel operates the site, and states that the CLI, the ingestion pipeline and the directory itself are all open source.
The centrepiece is a live install leaderboard. Rankings come from anonymous, deduplicated install counts the CLI reports back when users opt in, with hourly deduplication to stop anyone inflating a number. Three views are offered: All Time, Trending over 24 hours, and Hot, which compares the last hour against the same hour a day earlier. At the time of review the board showed 1,213,824 cumulative installs, led by find-skills from vercel-labs/skills at 3.1M, grill-me from mattpocock/skills at 962.8K and frontend-design from anthropics/skills at 816.4K.
Every skill page carries its source repository, install count, originating organisation, the agents the skill is most used on, and combined security audit results from three named partners — Gen Agent Trust Hub, Socket and Snyk. Skills that fail every partner audit are excluded from the directory entirely.
Browsing is organised three ways: twenty agent pages covering Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Gemini, Cline, AMP, Antigravity, OpenClaw, Droid, Goose, Kilo, Kiro CLI, Nous Research, OpenCode, Roo, Trae, VS Code and Zed; eight curated topics spanning React, Next.js, Design & UI, Mobile, Agent workflows, Databases, Testing and Marketing; and an Official section listing roughly a hundred organisations that publish their own skills, among them Anthropic, Microsoft, GitHub, OpenAI, Sentry, Cloudflare, AWS, Stripe, Supabase and Figma.
Two further mechanisms round it out. Packs bundle public skills, private files or zip archives and connected GitHub repositories into one unlisted install link, created from a Vercel account and shareable with a Vercel team; installing a pack needs no sign-in. A public REST API under /api/v1/ exposes the same catalogue data the site renders, authenticated with a Vercel OIDC token instead of a generated key. The service is free throughout: no pricing page, and no amount anywhere on the site.
What it does
- Browse a live leaderboard of agent skills ranked by anonymous, deduplicated install counts
- Install any listed skill into a compatible AI agent with a single npx command
- Check combined Gen Agent Trust Hub, Socket and Snyk audit verdicts before installing
- Filter the catalogue by AI agent, by topic, or by the organisation publishing the skill
- Bundle public skills, private files and GitHub repositories into one shareable pack
- Query the whole catalogue programmatically through a public REST API
- Publish your own skills and track their adoption with a README install-count badge
When to use skills.sh / When not to
A quick filter to help you decide if skills.sh is the right fit.
When to use skills.sh
- Developers already working with an AI coding agent who want to extend it with ready-made procedural knowledge
- Engineering teams that need to distribute an internal set of skills, mixing public entries, private files and GitHub repositories in one install link
- Skill authors chasing visibility, since listing happens automatically through install telemetry and a README badge is provided
- Product companies publishing official skills for their own technology, alongside the hundred or so organisations already in the Official section
- Developers building on Vercel who want catalogue data programmatically, authenticated with their project's OIDC token rather than a generated key
When not to use skills.sh
- Anyone looking for an AI agent itself — skills.sh catalogues what agents can be taught, it does not provide the agent
- Teams that need a vetted, warranted catalogue, since the terms explicitly disclaim any guarantee of skill quality or safety
- Users of agents outside the twenty the CLI supports, for whom nothing here installs
- Non-technical users, as the entire workflow runs through a command line and a Git repository
- Anyone needing confidential distribution, because packs are unlisted rather than access-controlled and anyone holding the URL can install them
How to use skills.sh
A typical end-to-end flow, from setup to results.
- Open the directory and browse the All Time, Trending or Hot leaderboard — no account is required
- Narrow the list through an agent page, one of the eight topics, or a publishing organisation
- Open a skill page to see its source repository, install count, top agents and partner audit verdicts
- Read the skill's SKILL.md in its source repository before installing it anywhere that matters
- Run npx skills add owner/repo — for example npx skills add vercel-labs/agent-skills — to download the skill and configure it for your agent
- Run npx skills update later to pull the current version of every installed skill, packs included
- Set DISABLE_TELEMETRY=1 if you would rather your installs were not counted in the rankings
- Sign in with Vercel to create a pack, then add public skills, private files and connected GitHub repositories
- Share the pack link and install it anywhere by passing that URL to npx skills add, no sign-in needed
- For programmatic access, enable OIDC Federation on your Vercel project, install @vercel/oidc and call getVercelOidcToken() inside the request handler
Pros & Cons
Pros
- Free throughout, with no account needed to browse or to install
- A single command installs a skill, and npx means the CLI itself never has to be installed
- Broad coverage, with around a hundred official publishers and 1,213,824 cumulative installs shown on the board
- Security audits from three named partners visible per skill, and skills failing every audit are removed
- Vendor-neutral across twenty agents rather than tied to a single one
- CLI, ingestion pipeline and site are open source, and corrections go through a public pull request
- Telemetry documented field by field and switched off with one environment variable
Cons
- The terms disclaim any guarantee of skill quality or safety, so the audit never replaces reading the SKILL.md
- Rankings measure installs, not quality or sustained use
- No email address is published anywhere, leaving forms and GitHub as the only routes
- No GDPR mention, no DPA, no subprocessor list, no hosting location and no minimum age
- Packs are unlisted but not access-controlled, so the URL is effectively the access
- The public API authenticates with a Vercel OIDC token, which in practice targets apps deployed on Vercel
- Interface and content are in English only
Pricing & Plans
skills.sh is free of charge in its entirety. The site publishes no pricing page — the /pricing path returns a genuine 404 — and no amount, currency or billing unit appears anywhere across the pages reviewed. Browsing the directory and installing skills or packs require no account whatsoever. Creating packs requires a Vercel account, and the public API requires a Vercel project with OIDC Federation enabled, but neither is presented as a paid capability. There is consequently no lowest price point to report.
- browse the leaderboard
- search the catalogue
- read audit verdicts and install any skill or pack with the CLI
- without an account
- create
- share and delete packs combining public skills
- private files and connected GitHub repositories
- with Vercel teams as the sharing scope
- call the public API at 600 requests per minute per team and project
- authenticated with the project's own token
Data, GDPR & hosting
A consolidated view of how skills.sh handles your data.
GDPR overview
There is no mention of the GDPR anywhere on skills.sh. The regulation is never named — neither to claim compliance nor to disclaim it — and the same holds for the CCPA. No Data Processing Agreement is published or offered, no subprocessor list exists, no Article 27 EU representative is designated and no Data Protection Officer is named. The site publishes no email address at all, so privacy questions and data deletion requests are routed to a contact page offering only forms and GitHub, with no stated procedure or deadline. No minimum age is given and no hosting country or region is disclosed. What stands in place of formal commitments is a narrow factual statement: four fields per install event, no session content, a deduplication fingerprint discarded hourly, and analytics without cookies or persistent identifiers.
Who owns the data?
The terms are explicit that skills.sh owns nothing it displays. Skills remain the property of their authors and are distributed under whatever licence sits in the source repository; the site states plainly that it does not own, host or relicense skill content, and that linking to a skill there transfers no rights. Packs belong to the Vercel account that created them, with Vercel teams as the sharing scope. Everything else — listings, install counts, per-agent breakdowns and audit verdicts — is public catalogue data, surfaced identically on the site and through the public API. No ownership claim is made over anything a visitor contributes.
Reuse rights
Reuse is broadly permitted and, in places, actively encouraged. The terms invite reasonable use of the public API, caching results on your own infrastructure included, and impose no restriction beyond a per-IP rate limit; only programmatic abuse, scraping that bypasses that limit, or use materially degrading the service for others may draw an IP-level block. Anything visible on a skill page is public and available through the same API, so no permission is needed to read or republish catalogue data. Skill content itself is a separate question: what you may do with it is governed by each author's repository licence, not by skills.sh, which neither owns nor relicenses it.
Data retention & training
Hosting summary
skills.sh discloses no hosting jurisdiction. No page names a country, a region or a data centre for the install telemetry, the aggregated counters or the API request logs, and no data residency commitment of any kind is offered. The privacy page mentions Vercel Analytics and Speed Insights without saying where either processes its data. The only locational signals available are technical rather than contractual: the domain resolves to a single A record geolocated in the United States, on infrastructure attributed to Amazon and flagged as an anycast node — indications about serving infrastructure, not a statement about where data comes to rest. The skills themselves are not hosted by skills.sh at all; the FAQ states they live in GitHub repositories, so their storage follows GitHub's arrangements rather than the directory's. Anyone with a data residency requirement should treat this as undocumented and ask directly.
Things to keep in mind
Risks and trade-offs to weigh before adopting skills.sh.
- Skills are third-party instructions and code that an agent executes on your machine; the terms themselves advise reading the SKILL.md and source repository before installing on a system that matters
- An audit badge is a signal, not a warranty: at the time of review several heavily installed skills carried Medium Risk verdicts and a number of entries were still marked Pending
- Install counts are easily mistaken for quality scores; they measure adoption reported by an opt-out telemetry and nothing more
- Packs are unlisted but not access-controlled, and the documentation explicitly warns against putting secrets or credentials in one
- Handing an agent ever more procedural knowledge can quietly erode the habit of checking its output — the convenience of a one-line install is exactly what makes review easy to skip
- Authenticated API requests are logged with team, project and environment identifiers taken from the verified token
- The absence of any GDPR, DPA, subprocessor or hosting disclosure leaves organisations with compliance duties without documentation to rely on
Setup & Integrations
Technical difficulty
Very low for the main use. Installing a skill takes one command, no account and no prior CLI installation, provided Node.js and a supported agent are already in place. Creating a pack is a handful of steps behind a Vercel sign-in, plus a GitHub connection to import repository skills. Customising a repository page means adding one JSON file at the repo root. The public API is the only genuinely technical part: it requires enabling OIDC Federation on a Vercel project, installing the @vercel/oidc helper and calling it inside the request handler rather than at module scope.
Deployment
Integrations
Supported languages
Behind skills.sh
Fundraising
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What exactly is a skill?
How do I install a skill?
Which AI agents are supported?
How much does skills.sh cost?
How is the leaderboard ranked?
Is any personal data collected?
Can I turn the telemetry off?
Are the listed skills safe to install?
What is a pack?
Who operates skills.sh?
Should you pick skills.sh?
skills.sh has become the default entry point to the open ecosystem of agent skills, and it earns that position mainly through restraint. It does not sell anything, does not lock you to one agent, and does not pretend the catalogue is curated. What it offers instead is a single command that works across twenty agents, a leaderboard that is honest about measuring installs rather than quality, and security audits from three named partners printed next to every entry — with skills that fail all three removed outright. Around a hundred organisations, Anthropic, Microsoft, GitHub, OpenAI, Cloudflare and AWS among them, publish official skills here, which is the clearest signal of how central the directory has become.
The limits are worth naming plainly. The terms disclaim every guarantee, so the audit badges are a signal and not a warranty: reading a skill's SKILL.md before letting an agent run it remains the user's job, and the site says so itself. Install counts invite being read as quality scores, which they are not. Packs are unlisted rather than protected, and the documentation warns against putting secrets in them. The legal surface is unusually thin for a site of this reach: no GDPR mention, no DPA, no subprocessor list, no hosting location, no minimum age and not a single published email address — an organisation with compliance obligations will find nothing here to rely on.
Taken for what it is — a free, open-source index and installer, not a trusted third party — skills.sh is a genuinely useful piece of infrastructure. Treat its verdicts as a starting point for your own review rather than a substitute for it.
- Choosing a selection results in a full page refresh.
- Opens in a new window.