The OpenClaw Skill Ecosystem: How 13,000+ AI Skills Got Built in 6 Months

Six months ago, the idea of an npm-style package registry for AI agent behaviors seemed like a research project. Today, ClawHub lists 13,729 skills with over 1.5 million combined downloads — and the numbers keep climbing.

This is how it happened, and why it matters for every developer building with AI.

What OpenClaw Actually Is

OpenClaw is an open-source personal AI assistant that runs locally on your machine. Mac, Windows, Linux. It connects to Anthropic Claude, OpenAI models, or local models via Ollama. It integrates with over 50 services out of the box: WhatsApp, Telegram, Discord, Slack, iMessage, Gmail, GitHub, Spotify, Obsidian, and more.

The core philosophy is captured in three words: your assistant, your machine, your rules.

What makes OpenClaw different from every other AI tool isn't the model integration or the local-first architecture. It's skills.

The SKILL.md Format: Dead Simple by Design

A skill is not a plugin. It's not a compiled binary. It's a directory with a markdown file.

That's the entire format. Every skill in the ClawHub registry follows the same pattern: a folder containing SKILL.md with YAML frontmatter at the top, followed by plain markdown instructions that teach the AI when and how to act.

The frontmatter looks like this:

---
name: skill-identifier
description: What the skill does
homepage: https://example.com
user-invocable: true
metadata:
  openclaw:
    requires:
      bins: ["node"]
      env: ["API_KEY"]
    os: ["darwin", "linux"]
---

Below the frontmatter: natural language instructions. The AI reads them and learns new behaviors.

This simplicity is not an accident. Peter Steinberger, OpenClaw's creator, built the format to be authored by anyone — including the AI itself. OpenClaw can generate new skills through conversation. You describe what you want; it writes the SKILL.md; you install it. The barrier between "I wish this AI could do X" and "this AI now does X" collapsed to a single conversation.

How ClawHub Works

ClawHub (clawhub.ai) is the official skill registry. Think npm, but for AI behaviors. The CLI interface should feel immediately familiar to any developer:

npm i -g clawhub
clawhub search "productivity"
clawhub install obsidian
clawhub list
clawhub publish ./my-skill/

The backend runs on Convex for database and file storage. Search uses OpenAI's text-embedding-3-small embeddings with vector search — which means you can search semantically, not just by exact skill name. The frontend is built with TanStack Start.

Publishing requires GitHub OAuth. That's the only gatekeeping. Anyone with a GitHub account can publish a skill.

The Numbers Behind 13,729 Skills

The raw registry count is 13,729. But that number is misleading in isolation.

The VoltAgent community curated a filtered list of 5,366 skills — removing spam, duplicates, low-quality entries, and outright crypto scams that made it into the registry. The LeoYeAI team maintains a more selective list of 339+ best-in-class skills, updated weekly.

This is the pattern of every open package registry. npm has 2.5 million packages, maybe 50,000 that developers actually use. The signal-to-noise ratio in early ecosystems is always poor. The community builds curation layers on top.

The category breakdown tells the real story of where developer energy went:

Coding tools dominate. Developers built skills for developers first — the standard pattern for any new developer platform.

What the Download Numbers Reveal

The top downloaded skill, Capability Evolver, has 35,581 downloads. That's a skill that makes OpenClaw better at improving itself — a meta-skill. Second is Wacli at 16,415, a utility wrapper. Third is Self-Improving Agent at 15,962, which teaches the AI to generate and refine its own instructions.

The pattern across the top 20 is striking: the most downloaded skills are not productivity tools or integrations. They're AI enhancement skills — capabilities that make the agent itself more capable.

• Capability Evolver: 35,581 downloads
• Self-Improving Agent: 15,962 downloads
• Proactive Agent: 7,010 downloads
• Auto-Updater Skill: 6,601 downloads

Developers aren't just extending OpenClaw. They're building self-improving loops into it.

The second cluster is workflow connectors: GitHub (10,611), Google Workspace via Gog (14,313), Obsidian (5,791). Real work lives in these tools. Skills bridge the AI to where the work already happens.

The GitHub Archive

The openclaw/skills GitHub repository archives every skill ever published to ClawHub. At last count: 91,882 commits. This is the most complete data source for anyone building analytics, search tools, or quality metrics on top of the ecosystem.

The VoltAgent awesome-openclaw-skills repository at GitHub curates 5,366 skills across 30+ categories and remains the most useful discovery layer for developers looking to find high-quality skills without wading through the full registry.

Why It Grew This Fast

Three reasons.

First, the format is trivially easy to understand. Any developer who can write markdown can write a skill. The learning curve is measured in minutes, not days.

Second, OpenClaw can generate skills from conversation. The ecosystem bootstrapped itself — users asked for capabilities, the AI generated skills, those skills got published. The creation loop is closed inside the tool being extended.

Third, 220,000+ GitHub stars brought an enormous developer community to the ecosystem fast. Peter Steinberger's personal following, his technical blog posts on agentic engineering, and significant press coverage created a flywheel that most open-source projects spend years building.

What Comes Next

The ecosystem is now at the stage where curation matters more than creation. The raw skill count will continue growing. The real value creation will happen in quality layers: curated lists, safety audits, quality scoring, and discovery tools that surface the right skill for the right job.

Tencent's SkillHub launch in March 2026 — a curated, CDN-accelerated mirror for Chinese developers — signals that the ecosystem has grown large enough to attract major platform investment. When large technology companies start building mirrors and distribution infrastructure around an open-source ecosystem, the bet on that ecosystem's longevity becomes significantly safer.

The SKILL.md format is not locked to OpenClaw. The same pattern works for any AI agent that can read and follow markdown instructions. The ecosystem built around one platform has already begun exporting its ideas to others.

That's what healthy ecosystems do.