Agent Skills standard donated to Agentic AI Foundation

Anthropic has transferred Agent Skills, an open standard for building modular AI agent capabilities, to the Agentic AI Foundation, a Linux Foundation initiative. The move represents a step toward vendor-neutral, interoperable agent infrastructure across the AI development landscape.

What changed

Anthropic developed Agent Skills while working on Claude Code and published the standard in December 2025. The company created the format to address a specific gap: Claude Code handled general knowledge tasks well but lacked procedural knowledge required for real business workflows. Within weeks of Claude Code’s October 2025 launch, developers published thousands of skills on GitHub and created loaders for Codex, Cursor, and OpenCode.

Anthropic has now transferred stewardship of Agent Skills to the Agentic AI Foundation, donating the standard name, logo, GitHub organization, and agentskills.io domain. This follows Anthropic’s earlier donation of the Model Context Protocol (MCP) to the same foundation.

The adoption has been rapid. More than 70 agent platforms have integrated Agent Skills, including ChatGPT, Cursor, GitHub Copilot, and VS Code. Since publication, one million skills have been published and deployed nearly 280 million times.

The technical structure is straightforward: a skill is a folder containing a SKILL.md file that teaches an agent a specific task or domain, along with any scripts or reference material needed. This simplicity appears to have driven uptake, making Agent Skills an accessible format for developers publishing capabilities across platforms.

Engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag designated as Lead Maintainers. Anthropic has proposed an initial roadmap including versioned releases, a formal proposal process, and security guidance for skill authors and installers.

Why it matters

The transfer to AAIF signals a shift toward standardization in the agent ecosystem. As AI agents become more widely deployed for professional workflows, the ability to share capabilities across platforms reduces vendor lock-in and lowers friction for teams choosing or switching between tools.

For professionals evaluating AI agent tools, the Agent Skills open standard addresses a real bottleneck: the need to rewrite or adapt procedural knowledge for each platform. By establishing a portable format, Agent Skills lets organizations build skills once and deploy them across multiple agent environments, much like containerization in infrastructure or API standardization in software integration.

The scale of adoption matters. With 280 million installations in roughly two months, Agent Skills already has meaningful traction. This level of adoption suggests that the format addresses genuine developer demand and may become a de facto standard for agent capabilities.

The governance structure is significant. Anthropic, Block, and OpenAI co-founded AAIF, suggesting that competing AI companies believe a neutral foundation serves their collective interests. This resembles how competitors collaborate on standards through organizations like the Linux Foundation. Anthropic’s willingness to donate both MCP and Agent Skills to neutral governance frames both as strategic bets on openness rather than proprietary advantage.

However, this also means the Agent Skills standard will evolve through community processes rather than under Anthropic’s direct control. For teams that have built skills using the current specification, compatibility across future versions becomes a consideration.

What to test

Before adopting Agent Skills as part of your agent infrastructure, consider these points:

• Verify that the platforms you currently use or plan to use actually support Agent Skills. The announcement lists 70+ platforms, but confirming your specific tools are on that list and understanding how deeply they have integrated the standard is important.
• Review the security guidance that will be published as part of the roadmap. Since skills can package scripts and procedural logic, security properties matter. Check whether the standard includes mechanisms for permissions, sandboxing, or auditing before installing unfamiliar skills.
• Test skill portability in your own environment. The intention is that a skill written for one platform works across many, but real-world performance and behavior may vary. Run the same skill on multiple platforms to confirm it behaves as expected.
• Track the AAIF governance process. The foundation will establish a formal proposal process for future changes. If your team plans to publish skills, understand how that process works and whether upcoming changes will require rework.
• Assess the maturity of available skills. With one million skills published, quality and maintenance vary widely. Evaluate the skills you plan to use: are they actively maintained, well-documented, and published by reliable authors?

Common questions

Is there an open standard for agent skills?

Yes. Agent Skills, originally developed by Anthropic and published in December 2025, is an open standard for portable AI agent capabilities. The standard specifies a simple format: a folder with a SKILL.md file that documents a capability, along with optional scripts or reference materials. Anthropic has now donated it to the Agentic AI Foundation under the Linux Foundation.

What are examples of agent skills?

The announcement does not provide specific examples of individual skills. However, it notes that skills teach agents “a specific task or domain” and that within weeks of Claude Code’s launch, developers published thousands of skills on GitHub. The standard is designed to package procedural knowledge that agents lack, such as domain-specific workflows or business processes.

What is the specification for agent skills?

The complete specification lives at agentskills.io. Engineers from Anthropic, Google, OpenAI, and other companies maintain the standard, and Anthropic has proposed a roadmap including versioned releases and a formal proposal process. The specification is open for community review and contribution, with discussion available on GitHub and the project’s Discord.

The conclusion

Agent Skills moving to the Agentic AI Foundation represents a strategic commitment to interoperability. The rapid adoption, 70+ platforms, 280 million installations, suggests the format solves a genuine problem in the agent ecosystem.

For professionals choosing agent tools, understanding Agent Skills is practical rather than academic. It signals which platforms prioritize portability and which have the engineering alignment to collaborate on standards. When comparing platforms, you can now ask: Do you support Agent Skills? How deeply? What is your maintenance commitment?

The next phase to watch is implementation of the proposed roadmap, particularly security guidance and the formal proposal process. If those succeed, Agent Skills may become foundational to agent development similar to how Docker became foundational to infrastructure. If governance becomes contested or the standard fractures, the benefit of portability diminishes. The AAIF structure should prevent that, but outcomes depend on community participation over time.