1,352 downloads, $0 revenue. 5 platforms already dead. Selling files doesn't pay—selling API services does.
Here's the short version: selling skill files doesn't make money. Selling API services does.
I have dozens of custom skills for Claude Code. I listed a few on marketplaces to test, and tried every monetization platform I could find. Here's what the data says.
This is the fundamental contradiction of the entire space:
Skills on ClawHub are markdown files. Users review the source before installing (for security), and once they've read it, they can just copy the whole thing.
To charge money, you need users to trust the skill is safe → open-source the code for review → users read it and grab the free version on GitHub → nobody pays.
App Store apps ship as compiled binaries. You can't extract the source. Skills are different. They're plain text. Copy-paste and walk away.
I listed three free skills on ClawHub as a test:
| Skill | Downloads | Actual Installs | Revenue |
|---|---|---|---|
| Systematic Debugging | 1,352 | 10 | $0 |
| Humanizer (AI text denoiser) | 118 | 0 | $0 |
| Content Pipeline | 121 | 0 | $0 |
1,352 downloads sounds decent until you see that only 10 people actually installed it. That's a 0.74% conversion rate.
For comparison, a competing skill (debug-pro) launched five weeks earlier: 12,950 downloads, 112 installs, 0.86% conversion. About the same—most downloads are probably bots and indexers. Real human installs sit below 1%.
From what I can see, three groups are making money, but none of them are "skill creators":
Platform operators. Top apps in the ClawHub ecosystem like OpenClaw Pro and Donely pull in tens of thousands per month. They sell full solutions, not individual skills.
Service providers. People who help businesses set up OpenClaw and build custom agents report $5K-12K/month based on public posts. They're selling expertise and time.
Early token participants. SkillMarket's $SKILL token and Claw Mart's ecosystem token may have appreciated for early holders. But that's token speculation, not skill revenue.
Who's not making money? One creator listed 30+ paid skills—total earnings: $208. Another built multiple products: $0. My own test: $0. These aren't outliers.
| Platform | Model | Key Data |
|---|---|---|
| ClawHub | Official store, 50% rev share | $274K/mo ecosystem |
| skills.sh | Vercel official directory, free | 89,000+ skills |
| SkillMarket | Pay-per-call API (x402/Solana), 70% | 847 skills |
| LarryBrain | $29.99/mo subscription, 50% | 134 subscribers |
| Claw Mart | Paid marketplace, $20/mo creator fee | ~$195K in 5 weeks |
| ComposioHQ awesome list | Open-source directory | 12K+ GitHub stars |
| skillsmp.com | Aggregator | Early stage |
Five platforms are already gone: SkillPort, AgentPowers, JustSkillPay, Agent37, SkillDock.
In a gold rush, even the shovel sellers aren't guaranteed to survive.
Skill-as-API. Package your capability as a pay-per-call API. The backend runs models or data processing. Users don't install anything—they make a call, you charge per request. The moat is in your servers and data, not the code file itself. Even if someone copies your skill file, they can't run it—the compute and data live on your side.
Vertical data skills. Generic skills (debugging, formatting, writing assistance) have zero moat—anyone can build a near-identical version. But if your skill depends on domain-specific data (financial signals, industry reports, proprietary datasets), the value lives outside the file.
Enterprise custom work. A $3 off-the-shelf skill won't sustain a business. A $3K custom agent solution can. But you've shifted from a product business to a service business, and that doesn't scale.