Anthropic Agent Skills Joins the Agentic AI Foundation: What Changes for Teams Building Agents

Anthropic Agent Skills, the open standard that packages specialized knowledge and workflows for AI agents, now belongs to a neutral foundation. On October 22, 2026, at the AGNTCon + MCPCon North America conference in San Jose, Anthropic announced it is donating the standard to the Agentic AI Foundation (AAIF) under the Linux Foundation, the same body that already hosts the Model Context Protocol. The move puts the most widely adopted agent capability format under vendor-neutral governance, and it matters for any team that builds or runs agents. Source

What happened on October 22

Anthropic is donating the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain to the AAIF. Engineers from Anthropic, Google, OpenAI, and other companies will keep maintaining the standard as individuals, with David Soria Parra and Mahesh Murag named as Lead Maintainers. Anyone can keep using or contributing to the standard without joining the foundation. Source

The same day, the AAIF announced that Agent Skills is one of four newly hosted projects. OpenAI contributed Open Responses, Hugging Face contributed OpenEnv, and Alibaba contributed OpenSandbox. The additions take the foundation from three launch projects in December 2025 to ten in its first year, joining the Agent2Agent protocol, Agent Router, agentgateway, AGENTS.md, goose, and MCP. Source

What Agent Skills actually is

A skill is a folder. It contains a SKILL.md file that teaches an agent a specific task or domain, plus any scripts or reference material the task needs. Anthropic developed the format while building Claude Code, which handled general knowledge work well but lacked the procedural knowledge that real business tasks demand. After an October 2025 launch inside Claude's tooling, Anthropic published Agent Skills as an open standard under its own GitHub organization and domain in December 2025 so that any agent could use it. Source

The AAIF's own framing shows where the format fits in the wider stack. The foundation describes its four new projects as covering four layers of agentic systems: how agents gain new skills, how applications talk to models across providers, how agents are trained and evaluated, and where agents run safely. Agent Skills owns the first of those layers, which makes it the piece most teams will touch directly. Source

The adoption numbers explain why the governance question came up at all. More than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated the standard. Developers have published 1 million skills and installed nearly 280 million of them. Within weeks of the original launch, developers had already published thousands of skills and written loaders for Codex, Cursor, and OpenCode, which is what pushed Anthropic to make it a standalone standard in the first place. Source

Why the governance move matters

A standard that 70 competing platforms depend on has a structural problem when one vendor controls it: every other adopter is betting on a competitor's roadmap. Moving Agent Skills to the AAIF resolves that the same way the industry resolved it for MCP, which Anthropic donated to the foundation earlier, after co-founding the AAIF with Block and OpenAI. AAIF CTO and co-founder Manik Surtani framed the pattern directly: four different companies chose to contribute their projects to neutral governance, and the foundation's job is giving the community one place to build shared standards with no single company controlling the outcome. Source

The MCP precedent is instructive here. After that protocol moved to the foundation, it kept its momentum and became the default way agents connect to tools and data, with competing vendors implementing it on equal footing. Agent Skills now gets the same structure at a similar moment in its adoption curve: proven enough to be worth standardizing, young enough that governance choices still shape the ecosystem. Source

The practical roadmap Anthropic proposed for the project under the AAIF includes versioned releases, a formal proposal process for specification changes, and security guidance for people who write and install skills. The project will run community input through a Discord and transparent decision-making in the open, following the MCP model. Source

The timing also completes a picture that started forming two days earlier. AGNTCon + MCPCon is the same conference where Google introduced Agent Compute Engine, its purpose-built infrastructure for running agents. One vendor after another is moving agent infrastructure, protocols, and now capability formats into shared, neutral, or open governance. The agent stack is standardizing in public. Source

What changes for teams building agents

Your skills investment is safer. If you have written skills for one platform, the standard's new home makes it more likely they keep working across the 70-plus platforms that integrated it. A SKILL.md folder is becoming the closest thing agents have to a portable plugin format, and neutral governance lowers the risk of building on it. Source

Treat skills like dependencies, because that is what they are. A skill folder can carry scripts and reference material that an agent will execute or follow. Installing a third-party skill is closer to adding a package to your project than to reading a blog post, and it deserves the same review: read the SKILL.md, inspect any scripts, and pin versions once versioned releases land. Our AI agent security checklist covers the broader review habit.

Skills are becoming shared team knowledge. Because a skill folder captures how a task should be done, teams are starting to use the format to onboard agents the way they onboard people: here is how we write release notes, here is how we triage this queue, here is our deployment checklist. Neutral governance makes that knowledge portable when you switch or add platforms, which is exactly when documented process usually gets lost. Source

Watch the security guidance work. The proposed roadmap explicitly includes security guidance for skill authors and installers. With 280 million installs already behind the format, that guidance will likely become the baseline auditors ask about. Teams that adopt the review discipline early will not have to retrofit it. Source

The standards around your agent are consolidating. Between MCP for tool connections, AGENTS.md for repo instructions, A2A for agent-to-agent messaging, and now Agent Skills for capabilities, the AAIF portfolio covers most of the integration surface a production agent touches. If you already follow our MCP server security best practices, the skills layer is the next place to apply the same rigor.

What to watch next

Three signals will show whether the donation changes behavior rather than just letterhead. First, whether the versioned release process actually ships, since pinning to versions is what makes enterprise adoption possible. Second, who participates in the formal proposal process: the maintainers are individuals from competing companies, and the balance of that group will decide how neutral the standard feels in practice. Third, what the other three new AAIF projects, Open Responses, OpenEnv, and OpenSandbox, do to the interoperability picture across model providers, training environments, and execution sandboxes over the next two quarters. Source