Is selling Claude Code skills worth it in 2026?

What just changed with Anthropic's own plugin directory?

maybe worth building's radar caught four separate skill and plugin repositories climbing GitHub's trending page in the same stretch that Anthropic's official directory, anthropics/claude-plugins-official, sat at 34,321 stars and shipped commits daily. The directory launched publicly on May 22, 2026, and it isn't a passive list. As of this week it holds 39 plugins built and maintained by Anthropic itself (a security scanner, a code-review tool, a commit-message generator, a frontend-design skill, and more) plus 15 external partner integrations that had to clear Anthropic's own quality bar, including NetSuite, Carta, PayPal, and Unity. Every one of them installs with a single command, and every one is free.

That matters because it answers the question a would-be skill seller has to ask first: who else is already giving away the thing I want to charge for? For the categories Anthropic decided to own itself, the answer is now Anthropic. A generic code-review skill, a git-workflow skill, a commit-message skill: all shipped, all free, all one command away, no marketplace required.

Why are the most popular skills repos all free?

Because a skill is a text file, and the entire buying audience for one already lives on GitHub, where the going rate for a good idea is a star, not a dollar. mattpocock/skills sits at 237,847 stars. ponytail, a skill that just tells the agent to stop over-engineering things, is at 112,381. ai-job-search is at 36,341, book-to-skill at 25,856, archify at 17,610, and claude-obsidian at 13,328. Four of those, archify, ai-job-search, claude-obsidian, and ponytail, add up to 179,660 stars between them, and every one is open source.

None of that is a business, but it is the competition a paid skill has to beat. When the free version has six-figure social proof and the paid version has none, most builders take the free one and paste in their own tweaks. A $9 skill sitting next to a 200,000-star free repo doing something similar isn't an underdog story. It's a pricing mistake.

Where is the funded money in agent skills actually going?

Not into a storefront for individual skills. Arcade.dev raised a $60M Series A in June 2026, led by SYN Ventures with Morgan Stanley and Wipro joining, on top of a $12M seed the year before: $72M total. On August 5, 2026, Arcade acquired Smithery, the registry that already lists 19,626 skills for free discovery. Arcade's own benchmark of the ecosystem it just bought into, run across more than 43,400 MCP servers and 219,069 tools, found only 0.5% earned an "A" grade for reliability. That's the problem Arcade is actually selling a fix for: production governance and trust for enterprises deploying agent tools, not a checkout page for a $12 skill.

So the receipts point in one direction. The money that showed up with a real check attached went to the layer that sits above individual skills (discovery, trust, governance sold to companies), which we covered in full in the agent skills registry verdict. Nobody has raised a real round to run a Gumroad for SKILL.md files.

Can an individual builder actually make a living selling skills?

A few can. Most can't, and the numbers explain why. Agensi, a small Amsterdam-based marketplace where creators keep 70% of each sale, publishes its own distribution: the median listed skill earns under $50 a month, while the top individual skills clear $500 to $3,000 a month, and that top slice takes most of the platform's revenue. Agent37, a competing platform that pays creators 80%, launched on Hacker News in December 2025 on exactly this observation from its founder: "I noticed there are 1,000+ MCP skills on GitHub with zero way for creators to charge for them." Both platforms solved the payment problem. Neither solved the harder one underneath it.

That harder problem showed up in a Hacker News thread from July 30, 2026, titled "Why you should sell your AI Skills?", linking to a third marketplace, Capabase. A commenter, mathgeek, wrote the pain plainly: "Love the idea, but in practicality I'd want to see a reproducible example that end users can't just take a description of the skill and throw it at an LLM to get most of the way there for free. You'd also need to prove to a buyer that an LLM isn't already better at producing these skills from scratch... Personally I suspect the value of any such skills rapidly decreases over the course of days/weeks once any marketplace signals the ones that are popular." The marketplace's own founder replied and didn't push back. That's the whole problem in one thread: a listing that proves a skill is popular is also the listing that hands a competitor the description needed to copy it for free.

When is selling Claude Code skills worth it?

• Sold as a service, not a file. Keep the hard, undocumented parts off GitHub and off any marketplace listing. The buyer pays for the outcome, not for a document they could paste into an LLM themselves.
• Sold to one business, not to strangers. A company paying for a proprietary internal skill built around its own systems isn't shopping a marketplace of comparable options. It's buying custom work, priced like custom work.
• Bundled with implementation. The skill is the delivery mechanism for consulting or an integration, not the standalone product. This is the same shape the AI implementation business verdict lands on: productize the repeatable parts, but sell the outcome.

When it isn't

• A general-purpose skill sold to the open market. If it's genuinely useful and broadly applicable, a free version with more stars than yours already exists or will exist within weeks.
• Anything Anthropic could plausibly bundle next. Code review, commit messages, git workflows, frontend polish: Anthropic's own 39-plugin directory already claims that ground, and it adds new categories on a weekly commit cadence.
• A marketplace listing you expect to stay hidden. The moment a paid skill's description is public, it's a spec sheet a buyer can hand to their own agent instead of paying you.

The test to run before you build

Our standard bar is a two-part receipt: a real company or raise proving the space, and a real person naming the pain. This idea clears both, and both point the same direction. The space receipt is Arcade.dev's $72M and its acquisition of a 19,626-skill directory, spent on trust infrastructure for enterprises, not a consumer storefront. The pain receipt is mathgeek's July 30, 2026 comment naming the exact failure mode: a popular listing becomes the blueprint a free LLM call replicates. Claude Code itself went through the same layer-ownership fight a year earlier over who controls the coding tool, covered in the Cursor vs Claude Code vs Windsurf verdict, and the pattern repeats here: the platform keeps the commodity layer for itself.

Here's the one way this is wrong. If Anthropic ships an actual in-product payment rail, letting a creator charge inside the plugin directory itself, the trust and distribution problem that kills every current marketplace disappears in a single release. That hasn't happened. Every paid option today still routes through a third party the buyer has to trust separately from Claude Code itself, which is exactly the friction a platform-native option would remove.

Frequently asked questions

Is there a funded company selling AI agent skills to individuals?

Not one that raised money to do exactly that. Arcade.dev has raised $72M total and bought Smithery (August 5, 2026), a directory of 19,626 free skills, but Arcade sells governance and production infrastructure to enterprises. The consumer marketplaces that do sell individual skills, Agensi, Agent37, and Capabase, are small and unfunded as of this writing.

What is a Claude Code skill?

A markdown file, usually SKILL.md, that teaches Claude a specific capability or workflow. It can ship alongside commands, sub-agents, and MCP configs inside a plugin. Anthropic's own directory lists 39 first-party plugins and 15 vetted external partners, all installable free.

How much can you actually earn selling a Claude Code skill?

Per Agensi's own published numbers, the median listed skill earns under $50 a month. The top individual skills clear $500 to $3,000 a month, and that top slice takes most of the platform's revenue. It's a hits business with a long, thin tail.

Why did Anthropic's own plugin directory change the math?

It closed the exact gap a paid skill seller was filling: a trustworthy, discoverable place to get a good skill. The directory now ships 39 Anthropic-built plugins plus 15 partner integrations, including NetSuite, Carta, PayPal, and Unity, preinstalled and free.

Why are the biggest skills repositories all free?

A skill is a text file distributed on GitHub, where the going currency is a star, not a dollar. mattpocock/skills sits at 237,847 stars and ponytail at 112,381, neither charging a cent. A free skill with real reach beats a paid one nobody has heard of.

What would actually make selling Claude Code skills worth it?

Selling a narrow, proprietary skill as a service to one specific business, not a general file to the open market. Keep the hard parts off GitHub, bundle it with implementation, and charge a company for an outcome instead of charging a stranger for a document.

What's the one way this verdict is wrong?

If Anthropic ships a real in-product payment rail for skill creators, the trust and distribution problem killing every current marketplace disappears overnight. No official marketplace supports a price tag today, so every paid option still runs through a third party the buyer has to trust separately.