On October 22, 2026, Anthropic did something companies almost never do with a hit product: it gave it away. The Agent Skills standard — the thing that lets an AI agent load a packaged capability on demand — now belongs to a neutral foundation, not to Anthropic. And the timing is the tell: more than 70 agent platforms had already adopted it before the ink was dry.
Agent Skills started as an experiment in Claude Code. Many across the industry have come to adopt it, so we at Anthropic found it more fitting to donate it to the @AgenticAIFdn, next to MCP. https://t.co/rn33HmNxav
— Boris Cherny (@bcherny) October 22, 2026
The destination is the Agentic AI Foundation (AAIF), which sits under the Linux Foundation. Anthropic co-founded AAIF with Block and OpenAI last year, and it is the same home it chose for the Model Context Protocol (MCP). So Agent Skills now stands beside MCP under governance no single vendor controls. There is a catch to this generosity worth examining — more on that below.
What a "Skill" actually is
Strip away the branding and a Skill is a portable instruction kit. It bundles whatever an agent needs to do one job well — guidance, scripts, references — so the agent can pull it in only when the task calls for it, rather than carrying every capability all the time. Think of it as a plug-in for behaviour instead of code.
Agent Skills began inside Claude Code as an experiment. In Boris Cherny's words, "Agent Skills started as an experiment in Claude Code. Many across the industry have come to adopt it, so we at Anthropic found it more fitting to donate it to the @AgenticAIFdn, next to MCP." That adoption is the whole justification: once rivals build on your format, keeping it proprietary starts to work against you.
The numbers that make this real
Standards are cheap to announce and expensive to prove. Agent Skills has the proof. Since Anthropic published the standard in December 2025, 1 million skills have been published and nearly 280 million have been installed — and the roster of adopters runs through ChatGPT, Cursor, GitHub Copilot and VS Code.
Here's a back-of-envelope figure that makes the scale tangible: with 1 million skills published and 280 million installs, each published skill has been installed about 280 times on average. That is not a library gathering dust; it is a format people actively reach for.
AAIF's land-grab year
The donation did not land alone. AAIF announced four new projects on the same day: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba. The additions push the foundation to 10 hosted projects in its first year, up from the three it launched with in December 2025.
A foundation a year old is already the parking lot for the industry's agent plumbing. (Chart: BougainWell · Data: Agentic AI Foundation)
Read the contributor list again: Anthropic, OpenAI, Hugging Face, Alibaba. These companies compete hard on models and products. They are cooperating on the connective tissue — the layer that decides whether an agent built on one vendor's stack can use a capability written for another's. That split is deliberate, and it is where the sector is heading.
The announcement was staged at AGNTCon + MCPCon North America, running October 22-23 at the San Jose McEnery Convention Center — the same event built around MCP. Pairing the Skills news with the MCP crowd is not a coincidence; it frames Skills as the second pillar of the same open-agent story.
So what's the catch?
Giving away a standard is not charity, and treating it as such misreads the move. Here is the open loop paid off.
When your format becomes the default that everyone builds on, you win even without owning it. Anthropic's leverage shifts from controlling the spec to being the company that understands it best and ships the best agent on top of it — a direction already visible in Claude Duet's shift from waiting to watching. A neutral standard also removes the single biggest objection a cautious enterprise raises — the fear of betting on one vendor's proprietary format. Hand governance to a foundation and that objection evaporates, which widens the whole market Anthropic sells into.
Why this matters beyond the headline
The pattern now is unmistakable. MCP handles how agents connect to tools and data. Agent Skills handles what capabilities an agent can pick up and use. Both sit in the same neutral foundation. Put together, developers get an interoperable agent layer that no single company can fence off — and that is the precondition for agents to move between platforms instead of being locked to one.
For a developer, building on Agent Skills no longer means betting on Anthropic's roadmap; it means betting on a shared standard with 70-plus platforms behind it. That is a very different risk calculation, and it is exactly the calculation open standards are designed to change. The same logic applies to the AI adoption gap that holds most organisations back: shared, interoperable infrastructure lowers the bar for every team to plug in.
What to Take Away
- Anthropic donated the Agent Skills standard to the Agentic AI Foundation, placing it under neutral Linux Foundation governance next to MCP.
- Adoption was already huge before the hand-off: 70+ platforms, 1 million skills published, and nearly 280 million installs since December 2025 — about 280 installs per published skill.
- The donation came with three rivals' projects (OpenAI, Hugging Face, Alibaba), pushing AAIF from 3 to 10 hosted projects in a year.
- Giving away a winning standard is strategy, not generosity: neutrality removes the vendor-lock objection and expands the market Anthropic competes in.
- Watch the maintenance, not the announcement — true independence shows up in how the standard is actually governed over time.
Chart(s): BougainWell, built from the Agentic AI Foundation. This article is for general information only and is not investment advice.
Cover image: Anthropic, via their announcement at anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation.
Sources
All analysis and opinions in this article are BougainWell's own.
