The Weekly Index #3: The Rise of Claude Skills: What Our Index of 2,471 OpenClaw Tools Shows

Issue #3 of The Weekly Index. This week: a deep look at the Claude Skills and OpenClaw ecosystem, the fastest-growing category in our index that still does not get enough attention.

📊 This Week's Numbers

• 2,471 OpenClaw Skills indexed (clawsecure source)
• 1,341 VS Code AI extensions indexed (see this week's ranking →)
• 62 is the highest score any VS Code extension earns in our model, the field is competitive but level at the top
• 52.8% of scored tools still have security score ≤ 2, our ongoing concern, addressed in issue #2

🔦 Spotlight: The OpenClaw Skills Ecosystem

If you have been watching the MCP server ecosystem grow, you may have missed a parallel movement: the OpenClaw Skills format, distributed through ClawHub and the clawsecure registry. We have 2,471 OpenClaw skills in our index, more than the number of MCP servers in the original awesome-mcp-servers list.

The format is different from MCP. OpenClaw skills are skill files that define agent behaviors, personas, and capabilities in a structured YAML/markdown format. They do not require a running server process, you add them to your agent's context and they shape how it responds.

The highest-scored OpenClaw skill in our index is Ecommerce (score: 35/100), which provides an agent with e-commerce operations capabilities including payment security, inventory management, and checkout flows. That score might look low compared to MCP servers (which often score 80-100), but the scoring model penalizes lower GitHub star counts, OpenClaw skills tend to live on ClawHub rather than GitHub.

📈 Trend: Claude Skills vs MCP Servers, When to Use Which

The question we get most often from developers new to the ecosystem:

Use an MCP server when: you need to connect your agent to an external system (database, API, file system, browser). MCP is the right tool for any integration that requires real-time data or state changes.

Use a Claude Skill when: you want to shape how your agent thinks or responds without an external integration. Skills are good for: adding domain expertise, defining personas, enforcing output formats, or building reusable reasoning patterns.

They compose well. A research agent might use an MCP server to browse the web (Firecrawl) and a Claude Skill to define how it synthesizes and presents findings.

🆕 Top 5 OpenClaw Skills Worth Adding to Your Agent

1. security-sentinelUnified security scanner for OpenClaw workspaces. Audits dependencies, secrets, and code patterns.
2. docx-skillGenerate Word documents from text. Simple but surprisingly useful for report-generating agents.
3. feishu-cardSend rich interactive cards to Feishu. For teams using Feishu/Lark as their communication layer.
4. claw-controlKanban for AI agents. Coordinate multiple agent tasks with a structured task board.
5. clawdzapDecentralized P2P messaging between agents. For multi-agent architectures that need inter-agent communication.

⚙️ Use Case Corner: How to Add a Skill to Your Claude Project in 5 Minutes

1. Pick a skill from our index or from ClawHub
2. Download the skill file (usually a .md or .yaml file)
3. In your Claude Project, go to Project Instructions and paste the skill content
4. The skill is now active for all conversations in that project

Skills do not require API keys, running servers, or any infrastructure. The tradeoff: they only work in Claude Projects, not in arbitrary MCP-compatible clients.

👥 Community Picks

• obsidian-skills10k GitHub stars, Claude skill for Obsidian integration. If you use Obsidian as a knowledge base, this turns Claude into a note-aware assistant.
• GPT Researcher25k stars, autonomous research agent for browser-based deep dives. The most-starred non-MCP tool in our browser category.

---

Next week: we ran security analysis across 4,133 MCP servers. The results are surprising. See you Thursday. Past issues