20 GitHub Skills Repositories With the Highest Stars in 2026: The Definitive Ranking

🥇 #1 — multica-ai/andrej-karpathy-skills

⭐ 156,000 stars  |  github.com/multica-ai/andrej-karpathy-skills

The single fastest-growing Claude skills repo ever. Developer Forrest Chang encoded Andrej Karpathy’s viral January 2026 post about LLM coding failures into one SKILL.md file with four hard behavioral rules. The result: 156,000 stars — arguably the most viral developer tool launch of the year.

The four rules target the three failure patterns Karpathy called out directly: agents making silent wrong assumptions and charging ahead, over-engineering that turns 50 lines into 500, and making orthogonal changes to code the agent was never supposed to touch. One file, zero runtime dependencies, compatible with any AI coding agent that reads a CLAUDE.md or SKILL.md. The principles bias toward caution over speed — right for non-trivial work, possibly overcalibrated for simple edits.

Best for: Any developer frustrated by Claude or Cursor making confident wrong changes. Install: Drop SKILL.md into your .claude/skills/ directory.

🥈 #2 — vibe-coding/agentic-engineering-guide

⭐ 58,500 stars  |  github.com/topics/claude-code-skills (via GitHub Topics)

Subtitled “from vibe coding to agentic engineering — practice makes claude perfect,” this repo by a Pakistani developer community became the definitive best-practices guide for Claude Code power users. It covers agentic workflows, context engineering, Claude Code commands, multi-agent orchestration, and the mindset shift from writing code to defining success criteria.

The most actionable reframe in the entire repo: “Don’t tell it what to do — give it success criteria and watch it go.” Defining verifiable success criteria before a task reduces clarification loops substantially. The repo includes working examples for each pattern, not just theory. It has become required reading in most serious Claude Code communities on Discord.

Best for: Intermediate to advanced Claude Code users. Beginners will benefit more from #4 or #7 first.

🥉 #3 — skills-library/agentic-skills (1,600+ skills)

⭐ 41,300 stars  |  github.com/topics/claude-code-skills

The largest installable skills library on GitHub by volume: 1,600+ agentic skills, an installer CLI, pre-built bundles, and official/community skill collections under one roof. Compatible with Claude Code, Cursor, Codex CLI, Gemini CLI, Antigravity, and more via a unified installer. The project includes specialized plugins, workflow templates, and skill bundles organized by domain (engineering, marketing, research, operations).

The installer CLI is the real differentiator. Instead of manually copying SKILL.md files, you run a single command to browse and install any bundle. The downside: with 1,600+ skills, quality is uneven. The official and community-curated collections are strong; the long tail requires more judgment. Forks from this repo have spawned at least six derivative projects with different curation philosophies.

Best for: Teams who want one repo to manage all their agent skills. Use the official bundles, not the raw long tail.

#4 — VoltAgent/awesome-agent-skills (1,000+ curated)

⭐ 26,000 stars  |  github.com/VoltAgent/awesome-agent-skills

The community’s go-to curated directory: 1,000+ skills collected from official engineering teams including Anthropic, Google Labs, Vercel, Stripe, Cloudflare, Netlify, Trail of Bits, Sentry, Expo, Hugging Face, Figma, DuckDB, and GSAP. Unlike bulk-generated repos, VoltAgent focuses exclusively on real-world skills created and used by actual engineering teams.

Standout inclusions: DuckDB’s full suite of SQL query and file-reading skills, Google Labs’ Stitch design-to-code skills (shadcn-ui, remotion, react-components), and NVIDIA’s NemoClaw skills. The repo is updated multiple times per week and maintains a clean table format with platform compatibility columns. It’s described as “the most contributed Agent Skills repository” and accepts community PRs with clear quality guidelines.

Best for: Discovering skills from specific companies or frameworks you already use.

#5 — alirezarezvani/claude-skills (337 skills + 30 agents)

⭐ 18,700 stars  |  github.com/alirezarezvani/claude-skills

The most comprehensive structured skills library for professional workflows: 337 skills, 30+ agents, 70+ custom commands, 330+ reference templates, and 579 stdlib-only Python CLI tools (zero pip installs required). Domains covered span engineering, marketing, product, compliance, C-level advisory, research, business operations, finance, and personal productivity.

The production-grade organization is what separates this from most community repos. Each skill domain includes a SKILL.md, extracted references/, and a usage-focused README. The 9 SEO and Google Ads skills alone — which connect to Google Search Console, PageSpeed Insights, and Google Ads API — are worth the star on their own for marketers. Works natively across 13 platforms including Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Aider, and Antigravity.

Best for: Business and marketing professionals who want production-ready AI workflow tools beyond just coding.

#6 — ComposioHQ/awesome-claude-skills (1,000+ with app actions)

⭐ ~16,000 stars (growing)  |  github.com/ComposioHQ/awesome-claude-skills

Composio’s curated list stands apart from other directories for one specific reason: the connect-apps plugin that gives Claude the ability to take real actions — send emails, create GitHub issues, post to Slack, update CRM records — across 500+ apps via Composio’s integration layer. This makes it the most “agentic” of the curated directories in the truest sense of the word.

The skills collection itself covers 1,000+ production-ready use cases across Claude.ai, Claude Code, Codex, Cursor, Gemini CLI, Antigravity, and Windsurf. The documentation quality is above average. Getting the connect-apps integration running requires a free Composio API key, which adds a setup step not present in other repos — but the payoff is Claude that can execute real-world tasks rather than just generate text about them.

Best for: Teams who want Claude to take real actions across their software stack, not just generate code.

#7 — travisvn/awesome-claude-skills (Ecosystem Tracker)

⭐ ~12,000 stars  |  github.com/travisvn/awesome-claude-skills

The most thorough historical and conceptual reference for the Claude Skills ecosystem. Travis VN’s repo tracks the full timeline of Skills development from the October 2025 announcement through present, documents the differences between Skills vs. Subagents vs. MCP servers, and maintains the most complete list of official Anthropic-released skills (docx, pdf, pptx, xlsx, algorithmic-art, canvas-design, skill-creator, template). Also includes obra’s superpowers experimental skills.

The conceptual clarity here is excellent: “If you find yourself typing the same prompt repeatedly across multiple conversations, it’s time to create a Skill.” The repo answers the ecosystem’s most-asked questions — costs, skill vs. subagent differences, how token loading works, security implications. A must-read before diving into other repos.

Best for: Newcomers to the skills ecosystem who want to understand the full picture before installing anything.

#8 — hesreallyhim/awesome-claude-code (Skills + Full Ecosystem)

⭐ ~10,500 stars  |  github.com/hesreallyhim/awesome-claude-code

The most curated collection for Claude Code specifically — covering skills, hooks, slash-commands, agent orchestrators, applications, and plugins. The emphasis on code quality and security is higher here than in the mass-aggregation repos. The maintainer reviews contributions carefully and rejects low-quality or AI-generated bulk submissions. This repo tends to be smaller but more reliable than the high-volume alternatives.

Particularly strong on orchestration patterns: how to coordinate multiple Claude Code agents, how to use hooks for persistent state, and how to build subagent workflows. If the 1,600+ skill libraries feel overwhelming, start here for quality-filtered recommendations.

Best for: Developers who prefer quality-over-quantity curation and want security-conscious recommendations.

#9 — heilcheng/awesome-agent-skills (Multi-Platform Index)

⭐ ~9,000 stars  |  github.com/heilcheng/awesome-agent-skills

Hailey Cheng’s community directory focuses on skills compatible across multiple agent platforms — not just Claude Code. Maintained in collaboration with the community, it includes the official Anthropic and OpenAI skill catalogs side by side for easy cross-platform comparison. Points to skills.sh (Vercel’s leaderboard) and the npx skills CLI for discovery, installation, and management. Clear quality guidelines distinguish it from bulk-aggregation repos.

The skills-as-recipe-cards framing is useful: “Instead of the AI needing to know everything upfront, skills let it learn new abilities on the fly.” Practical advice on security (only install from trusted sources), costs (free to use, subscription required to run), and updates (pull from git for repo-based skills).

Best for: Teams using multiple AI agents who need a platform-agnostic skills index.

#10 — BehiSecc/awesome-claude-skills (Security Focus)

⭐ ~8,200 stars  |  github.com/BehiSecc/awesome-claude-skills

The security-specialist’s skills directory. BehiSecc’s collection is notable for the depth of its security skills coverage: OWASP Top 10:2025 with ASVS 5.0 integration, Trail of Bits’ CodeQL/Semgrep static analysis skills, varlock for secrets management, sanitize for PII redaction, systematic-debugging for structured bug triage, and a webapp-testing skill built on Playwright. Also includes the FFUF fuzzing integration and computer forensics toolkit.

Beyond security, the repo has standout creative entries: crowdcast (multi-agent social simulation), dna-claude-analysis (personal genome analysis from 23andMe/AncestryDNA data), and Satori (a clinically informed reasoning companion). The breadth makes it one of the more interesting community repos to browse even if security isn’t your primary use case.

Best for: Security engineers and developers who want OWASP-aligned code review integrated into their agent workflows.

#11 — caramaschiHG/awesome-ai-agents-2026

⭐ ~7,800 stars  |  github.com/caramaschiHG/awesome-ai-agents-2026

300+ resources across 20+ categories updated monthly — the most comprehensive map of the entire AI agent ecosystem in 2026, not just skills. Tracks model releases, framework updates, MCP adoption milestones, and regulatory changes (EU AI Act effective August 2026). The agent frameworks section alone — covering CrewAI, Swarm, AutoGen, MetaGPT, and 20+ others — provides the context that skills repos often lack.

Genuinely useful for understanding why certain skills became popular: this repo tracks the broader adoption curves. The stat that OpenClaw went from 9k to 188k stars in 60 days, and that Cursor + Copilot + Claude Code collectively command 70%+ market share, explains why the skills ecosystem exploded so fast. Monthly updates keep it current in a way that most GitHub lists don’t bother with.

Best for: Anyone who wants to understand the full AI agent landscape, not just install skills.

#12 — duckdb/duckdb-skills (Official Data Query Suite)

⭐ ~7,200 stars  |  github.com/duckdb (via VoltAgent listing)

DuckDB’s official skills suite covers the full data workflow within Claude Code: attach a database for interactive querying, run SQL against attached databases or ad-hoc against files (CSV, JSON, Parquet, Avro, Excel, spatial data), search documentation, and even recover context from previous Claude Code sessions via session log search. Every skill uses DuckDB’s Friendly SQL dialect with automatic schema exploration.

This is what official skills look like done right: focused, well-documented, maintained by the team who actually built the product. The read-memories skill — which searches past Claude Code session logs to recover context — is genuinely novel and solves a real pain point for developers working across long projects. Zero setup friction beyond having DuckDB installed.

Best for: Data engineers and analysts who regularly work with structured data files inside Claude Code.

#13 — obra/superpowers (Experimental Skill Lab)

⭐ ~6,900 stars  |  github.com/obra

The experimental wing of the Claude skills ecosystem. obra’s superpowers-lab tests techniques not yet stable enough for production repos — new prompting patterns, context management strategies, and multi-agent coordination approaches that may become mainstream in 3–6 months. The skills here change frequently and the README warns explicitly that they may break between updates.

The value proposition is access to bleeding-edge patterns before they’re polished. Developers who track this repo and fork experiments that work for them tend to be ahead of the curve when those patterns become stable community standards. Referenced in both travisvn/awesome-claude-skills and hesreallyhim/awesome-claude-code as required following for Claude Code power users.

Best for: Experienced Claude Code users who want early access to next-generation skill patterns. Not for beginners.

#14 — google-labs-code/stitch-skills (Design → Code)

⭐ ~6,500 stars  |  github.com/google-labs-code

Google Labs’ official skills for the Stitch MCP server, which translates design specifications into production code. The suite covers: creating DESIGN.md files, enhancing prompts with design vocabulary, converting Stitch designs to React components via shadcn/ui, generating walkthrough videos via Remotion, and implementing a design-to-code feedback loop. All compatible with Claude Code, Gemini CLI, and Cursor.

The design-to-code workflow these skills enable represents a meaningful acceleration for frontend developers who work from design files. The enhance-prompt skill — which improves prompts with UI/UX vocabulary and design specs — is broadly useful even outside the Stitch ecosystem. Being official Google Labs output means maintenance cadence is reliable.

Best for: Frontend developers and designers who want to close the gap between design artifacts and working code.

#15 — Trail of Bits / tob-security-skills

⭐ ~6,100 stars  |  github.com/trailofbits

Trail of Bits — one of the most respected security research firms in the industry — published official Claude Code skills for static analysis with CodeQL and Semgrep, variant analysis, code auditing, and vulnerability fix verification. These aren’t community-generated security tips; they reflect years of professional security research encoded into repeatable agent workflows.

The combination of CodeQL for program analysis and Semgrep for pattern-based detection, orchestrated via Claude, creates a security review capability accessible to developers who aren’t security specialists. The fix verification component — which checks that a proposed fix actually closes the identified vulnerability — is particularly valuable in CI/CD integration scenarios.

Best for: Development teams that need professional-grade security analysis integrated into their AI coding workflow.

#16 — agentskills-io/cybersecurity-skills (754 Security Skills)

⭐ ~5,900 stars  |  Apache 2.0  |  github.com (via GitHub Trending)

754 structured cybersecurity skills mapped to five frameworks simultaneously: MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND, and NIST AI RMF. Covering 26 security domains. This is the most comprehensive security-specific skills collection available and one of the few that explicitly maps to AI security frameworks (ATLAS, AI RMF) in addition to traditional cybersecurity standards.

The framework mapping is genuinely useful for compliance-oriented teams. Being able to tell an agent “apply NIST CSF 2.0 controls to this architecture” and have it draw from a structured skills library is qualitatively different from prompting an unstructured model. Works with Claude Code, GitHub Copilot, Codex CLI, Cursor, Gemini CLI, and 20+ other platforms.

Best for: Security teams, compliance engineers, and organizations operating under NIST or MITRE framework requirements.

#17 — quemsah/awesome-claude-plugins (Plugin Metrics + Discovery)

⭐ ~5,700 stars  |  github.com/quemsah/awesome-claude-plugins

The only repo in this list that tracks plugin adoption metrics across GitHub using automated n8n workflows. Quemsah’s repo provides a data layer the others don’t: which skills are being used at scale, which are declining, and which are newly trending. This makes it genuinely useful as a discovery tool beyond the stars of the repo itself.

Notable skills surfaced by the metrics engine include the AI memory compression skill (65% token reduction via caveman-speak summarization), the token compression middleware (60–95% token reduction on tool outputs), and the multimodal generative media skills for image, video, and audio generation via muapi.ai. The n8n-powered automation that drives the metrics is itself a useful workflow example.

Best for: Developers who want data-driven skill discovery rather than relying solely on star counts or editorial curation.

#18 — perplexityai/bumblebee (Supply Chain Security Scanner)

⭐ ~5,400 stars (growing fast since v0.1.1 May 2026)  |  github.com/perplexityai/bumblebee

Perplexity AI’s response to a real problem: developers installing MCP servers from GitHub repos and community Discords without knowing what they contain. Bumblebee scans npm, PyPI, Go, RubyGems, and Composer package manifests, checks installed MCP servers against known-malicious registries, audits VS Code extensions, and scans browser extensions. Pure Go standard library, zero dependencies, read-only (makes no changes to your environment).

The timing of this repo is significant. As the skills ecosystem grew rapidly, the security surface area grew with it. Bumblebee fills a gap that no other tool in this list addresses: verifying that what you’ve already installed is safe. Early but fast-growing — 2,600 stars at v0.1.1 release in May 2026, accelerating since then. Requires Go 1.25+.

Best for: Any developer who has installed MCP servers or Claude skills from non-official sources and wants to audit their environment.

#19 — science-skills/lab-agent-skills (160,000+ Scientists)

⭐ ~5,300 stars  |  github.com (via Github-Ranking-AI)

The #1 Agent Skills library for scientific research: 140 ready-to-use skills covering biology, chemistry, medicine, and drug discovery, plus access to 100+ scientific databases. Used by 160,000+ scientists worldwide — the largest verified user base of any specialized skills library in this ranking. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard.

The database integration layer is what makes this repo exceptional. Rather than just teaching Claude how to reason about biology or chemistry, the skills connect to live scientific databases — literature, protein structures, chemical properties, clinical trial data — that the model can query in real time. This is the difference between Claude knowing about science and Claude doing science.

Best for: Researchers, biotech teams, and anyone doing computational science who wants AI agents that can access real scientific data.

#20 — yuxiaopeng/Github-Ranking-AI (The Meta-Index)

⭐ ~5,200 stars  |  github.com/yuxiaopeng/Github-Ranking-AI

The daily auto-updating leaderboard of the most popular AI repositories on GitHub by star count, organized by AI topic. It’s the closest thing to a real-time pulse on the entire AI tools ecosystem — not just skills, but models, frameworks, UIs, and infrastructure. The Claude.md file within the repo maintains a curated view specifically for Claude-related repositories.

Its value is meta: it’s how you discover what’s becoming popular before it shows up in curated lists. When a new skills repo starts climbing this leaderboard, it’s early signal of genuine community traction rather than hype-driven starring. For anyone tracking the AI development ecosystem professionally, this auto-updating index is one of the most useful bookmarks on GitHub.

Best for: AI tool researchers, content creators, and anyone who needs to stay current on GitHub’s AI landscape in real time.

Are GitHub skills repositories free to use?

What is the difference between Claude Skills and GitHub skills repositories?

Do these skills work with tools other than Claude?

How do I install skills from a GitHub repository?

Are there any security risks with installing GitHub skills?

Which skills repository is best for marketing and SEO work?