I analyzed 40 OpenClaw use cases — 92.5% don't need OpenClaw

Why I Built This I was looking at the awesome-openclaw-usecases repo — 40 real-world use cases for OpenClaw. It looked like a goldmine of automation ideas. But then I noticed something: almost none of them actually need OpenClaw. The "skills" are just API wrappers. The "memory system" is just file storage. The "sub-agents" are just parallel tasks. Every capability they depend on exists in any capable AI agent. What I Found Classification Count % Agent-agnostic (works anywhere) 20 50% Partial (swap one tool) 17 42.5% OpenClaw-dependent 3 7.5% Only 3 out of 40 use cases genuinely require OpenClaw. The rest? They work with Claude Code, Codex, Cursor, or any agent with shell access + LLM reasoning. The Context Efficiency Problem So I extracted the 37 agent-agnostic use cases. But dumping 37 guides into one skill file would burn ~30,000 tokens every invocation. That's wasteful. Instead, I used an Index Skill Pattern : SKILL.md is just a table of contents (~1,200 tokens) Each use case lives