The AI agent skill ecosystem has a discovery problem. OpenClaw's official registry, ClawHub, now lists over 13,700 skills — modular extensions that let your AI assistant interact with APIs, automate workflows, and perform specialized tasks. But roughly half of those are spam, duplicates, crypto/finance noise, or outright malicious. Finding the good ones means scrolling through pages of junk.
VoltAgent/awesome-openclaw-skills solves this with brute-force curation. The repo catalogs 5,200+ skills that survived filtering for quality, safety, and usefulness — organized into 30 categories with direct install links. At 44,600 stars, it's become the de facto starting point for anyone building with OpenClaw.
But this isn't just about one repo. The broader trend — curated awesome lists for AI agent tools — is reshaping how developers discover and adopt agent extensions. Let's look at what's inside, and how the ecosystem compares.
What OpenClaw skills actually are
If you haven't used OpenClaw yet: it's an open-source, locally-running AI assistant (350k+ stars) that operates directly on your machine. Think Claude Code or Cursor, but fully open-source with support for 25+ LLM providers out of the box.
Skills are OpenClaw's extension mechanism. They're modular packages that give the assistant new capabilities — connecting to external services, running specialized workflows, or adding domain-specific knowledge. Install them via the ClawHub CLI:
clawhub install <skill-slug>
Or drop them manually into ~/.openclaw/skills/ (global) or <project>/skills/ (workspace-scoped). You can also paste a skill's GitHub URL directly into the assistant chat and it handles setup automatically.
The official registry at clawhub.ai hosts everything. The problem is everything includes a lot of noise.
What VoltAgent filtered out — and why it matters
The awesome-openclaw-skills repo didn't just list skills. It excluded 7,200+ entries from the official registry based on:
- Spam and duplicates — identical skills uploaded under different names
- Low quality — skills that don't work, lack documentation, or serve no real purpose
- Crypto and finance — an entire category excluded due to high scam density
- Security flags — skills flagged by automated security audits as potentially malicious
That last point is worth emphasizing. The repo includes a clear warning: skills are "curated, not audited." They recommend running Snyk and VirusTotal scans before installing anything into your local environment. Good advice for any plugin ecosystem, but especially one where extensions run with your local permissions.
What survived the filter: 5,198 skills across 30 categories.
The biggest categories
The distribution tells you where the community's energy is focused:
| Category | Skills |
|---|---|
| Coding Agents & IDEs | 1,184 |
| Web & Frontend Development | 919 |
| DevOps & Cloud | 393 |
| Search & Research | 345 |
| Browser & Automation | 322 |
| Productivity & Tasks | 205 |
| CLI Utilities | 180 |
| AI & LLMs | 176 |
| Image & Video Generation | 170 |
| Git & GitHub | 167 |
| Communication | 146 |
| Transportation | 110 |
| PDF & Documents | 105 |
| Marketing & Sales | 102 |
No surprise that coding-related skills dominate — OpenClaw's primary audience is developers. But the long tail is interesting. Transportation (110 skills), Health & Fitness (87), Shopping & E-commerce (51), and Smart Home & IoT (41) show the assistant expanding well beyond the IDE.
The repo's README shows a preview of roughly 20 skills per category inline, with links to full markdown files in the categories/ directory for the complete listings.
Similar repos worth knowing about
VoltAgent's OpenClaw list is one node in a much larger network of curated AI agent resources. Here are the ones worth bookmarking.
For Claude Code
hesreallyhim/awesome-claude-code (36.9k stars) is the most comprehensive Claude Code resource — skills, hooks, slash commands, agent orchestrators, apps, and plugins. If you're building with Claude Code and only bookmark one repo, make it this one.
VoltAgent/awesome-claude-code-subagents (16.5k stars) focuses specifically on subagents — 100+ specialized agents covering security auditing, code review, documentation generation, and more.
VoltAgent/awesome-agent-skills (14.4k stars) takes a cross-platform approach, listing 1,000+ skills that work across Claude Code, Codex, Gemini CLI, and Cursor. This is increasingly how skills are distributed — targeting the skill specification rather than a single host.
alirezarezvani/claude-skills (9.7k stars) provides 220+ skills and agent plugins with support for 10+ different AI coding assistants.
For MCP servers
punkpeye/awesome-mcp-servers (84.3k stars) is the dominant list for Model Context Protocol servers — the connectors that let AI assistants access databases, APIs, file systems, and cloud services. If you're building integrations rather than skills, start here.
For Codex and other agents
VoltAgent/awesome-codex-subagents (3.6k stars) mirrors the Claude Code subagents list but for OpenAI's Codex, with 130+ specialized agents.
Individual skills that broke out
Some individual skills have become projects in their own right:
- planning-with-files (18.1k stars) — Manus-style persistent markdown planning for AI agents
- humanizer (12.6k stars) — removes signs of AI-generated writing
- Understand-Anything (7.9k stars) — turns codebases into interactive knowledge graphs
- caveman (3.8k stars) — cuts 65% of tokens by having the agent talk like a caveman
- trailofbits/skills (4.3k stars) — security research, vulnerability detection, and audit workflows from Trail of Bits
The convergence pattern
What's happening across all of these lists is convergence. Skills that were built for one agent are increasingly cross-compatible. The awesome-agent-skills repo from VoltAgent explicitly targets Claude Code, Codex, Gemini CLI, and Cursor with the same skill definitions. Tools like cc-switch (39.7k stars) let you run Claude Code, Codex, OpenClaw, and Gemini CLI from a single desktop interface.
This means the 5,200 OpenClaw skills aren't locked into OpenClaw. Many will work — or can be adapted to work — with whatever AI coding assistant you're using. The skill specification is becoming a shared standard, even if it's not formally standardized yet.
What to do with this
If you're using OpenClaw, the awesome-openclaw-skills repo is your starting point. Browse by category, install what's relevant, and scan before you trust.
If you're using Claude Code or another agent, the parallel lists (awesome-claude-code, awesome-agent-skills, awesome-mcp-servers) serve the same purpose. The ecosystem is large enough now that unfiltered registries are unusable — curated lists are infrastructure.
And if you're building skills: publish to the official registries, but also submit to the relevant awesome lists. That's where developers actually discover new tools. The registry is the package manager. The awesome list is the recommendation engine.
The companion site for browsing and searching the full collection is at clawskills.sh.