VoltAgent’s awesome-agent-skills is an index of Agent Skills, described in the repository’s About text as “A curated collection of 1000+ agent skills from official dev teams and the community, compatible with Claude Code, Codex, Gemini CLI, Cursor, and more.” The README badge puts the figure higher, at “1497+”.
It is MIT licensed, and structurally it is one very large README: a table of contents with roughly fifty sections, most named for the organisation that published the skills underneath it.
How it is organised, and why that matters
The index is grouped by publisher rather than by task. Sections carry names like Anthropic, VoltAgent, Microsoft, OpenAI, Vercel, Cloudflare, Hugging Face, Trail of Bits, SerpApi and Sentry, alongside thematic sections for community and marketing, product management and advertising skills.
Publisher-first grouping is the right call for a list this size. With nearly 1,500 entries, the question you can actually answer by browsing is “what has the vendor I already trust published?” — not “which of these 1,500 is best for my task.” Some sections are substantial in their own right: Microsoft’s covers “133 skills across 6 languages,” the official Claude section lists 17 spanning documents, art, design, web artifacts, MCP building and testing, and TestMu AI contributes dozens of test-framework entries.
Each entry is a one-line description plus a link, either to officialskills.sh or directly to a GitHub folder.
The curation claim, and what it is worth
The README stakes out a position: the collection is “Hand-picked, not AI-slop generated,” and focuses on skills “created and used by actual engineering teams.” There is a Quality Standards section backing that up.
Take it as a statement of intent rather than a guarantee, for two reasons visible in the list itself. First, some entries are not skills at all — the README notes that certain SerpApi and Crawlbase items are “not skills, but they pair with them.” Second, the list carries its own legacy debris: Lettuce is flagged “legacy; prefer Behave” and Protractor “deprecated; prefer Playwright/Cypress.”
Those flags are to the maintainers’ credit — they are doing the work of marking rot rather than hiding it. But they also tell you what a 1,500-entry index inevitably is: a snapshot with a maintenance burden, where some fraction is always behind. Read the flags; they are the most valuable annotations in the file.
What it does not tell you
Two gaps are worth knowing before you rely on it.
Installation is not covered here. The README points to “the table below for paths and documentation” and lists compatible hosts — Claude Code, Codex, Antigravity, Gemini CLI, Cursor, GitHub Copilot, OpenCode, Windsurf “and more” — but the actual setup happens in each linked repository. Expect to follow a link and read someone else’s instructions for every skill you adopt.
Per-skill licences are not stated. The MIT licence covers the index — the list itself — not the skills it points at. Each linked repository carries its own terms, and if you plan to redistribute or build commercially on a skill, that is the licence that governs you.
The README also includes sponsor sections and promotional banners, which is normal for a project this visible but worth knowing when you read a recommendation.
How to use a list this large
Not by reading it. Three approaches that work:
- Jump to the publisher you already have reason to trust and take what they ship.
- Search within the page for your framework or tool name rather than browsing categories.
- Treat the deprecation flags as the signal they are, and check the linked repository’s last commit before adopting anything.
As a map of who is publishing skills and in what volume, this is the most complete single view available, and the maintainers are visibly doing the unglamorous work of annotating decay. Just don’t mistake the index’s licence for the skills’ licence, and read the SKILL.md of anything you install — a skill is instructions your agent will act on, not an inert dependency.
