Quick Answer
Anthropic Agent Skills is now governed through the Agentic AI Foundation, a Linux Foundation project intended to provide neutral oversight for a reusable AI agent workflow standard. Anthropic says more than 70 platforms have integrated the format. Developers should treat the move as a portability improvement, while reviewing every downloaded skill because bundled instructions and code can create security risks.
Key Takeaways
- Anthropic donated Agent Skills to the Agentic AI Foundation on October 22, 2026.
- Agent Skills packages center on a SKILL.md file with instructions and metadata.
- Anthropic says more than 70 agent platforms have integrated the standard.
- ChatGPT, Cursor, GitHub Copilot, and VS Code are among the named integrations.
- Users should inspect skills before installation because they can include scripts and files.
What did Anthropic donate to the Agentic AI Foundation?
Anthropic donated the Agent Skills standard to the Agentic AI Foundation, placing governance of the reusable AI agent workflow format under a Linux Foundation group. Anthropic announced the transfer on October 22, 2026, and said it is transferring the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain. Anthropic’s announcement describes the handoff as an effort to establish neutral stewardship for the standard.
Agent Skills began as work connected to Claude Code before Anthropic first published the format as an open standard in December 2025. The governance change does not turn Agent Skills into a Claude-only feature. The purpose is to give developers and AI platform providers a common way to package specialized instructions, tools, and supporting files for agents that need to perform repeatable work.
The most important practical point is that a shared format can reduce the need to rewrite the same workflow for each AI environment. That said, a common package format does not guarantee identical results across models or products. Developers still need to test a skill in the specific agent platform where they plan to use it.
What are Anthropic Agent Skills?
Anthropic Agent Skills are folders that give an AI agent specialized knowledge and workflows for a defined task. Agent Skills documentation says each skill centers on a SKILL.md file containing at least a name, a description, and instructions. A skill can also include scripts, reference material, templates, and other resources needed to carry out the workflow.
The folder-based structure matters because an agent often needs more than a prompt to complete work consistently. A software development skill, for example, can provide instructions about a repository’s conventions alongside templates or scripts that support the task. The standard organizes those materials in one package rather than requiring a user to manually paste the same context into every conversation.
Agent Skills are not the same as an AI model’s underlying training. A skill gives an existing agent task-specific guidance and resources, while the model still determines how it interprets the instructions and produces an answer. Users who are comparing agent features may also want to understand how proactive AI agents differ from reusable workflow packages.
How do Agent Skills load information into an AI agent?
Agent Skills use progressive disclosure, meaning an AI agent loads information in stages instead of reading every file at once. The Agent Skills documentation says an agent first loads a skill’s name and description, then reads the full SKILL.md instructions when the skill is relevant, and then accesses bundled code or files only when needed. The Agent Skills documentation presents that process as a way to keep specialized context available without loading unnecessary material for every request.
Progressive disclosure can make an agent workflow more manageable because a platform can identify potentially relevant skills before pulling in the complete instructions. The approach also helps skill authors separate a short description from detailed operational guidance. A short description tells the agent when to use the skill, while SKILL.md can provide the procedure, limitations, and expected output.
The limitation is that selective loading does not remove the need for careful design. A vague description can make a skill harder for an agent to select, and incomplete instructions can lead to inconsistent execution. Skill authors should state the intended task, required inputs, allowed tools, and stop conditions clearly in the package.
Which AI platforms support the Agent Skills standard?
More than 70 agent platforms have integrated Agent Skills, according to Anthropic. Anthropic specifically named ChatGPT, Cursor, GitHub Copilot, and VS Code among the platforms that support the standard. The company’s figure is self-reported, so it establishes the scope Anthropic describes rather than providing an independent compatibility test for every implementation.
Platform support matters because a reusable skill becomes more useful when developers can take the same workflow across multiple tools. An engineering team may use one product for coding assistance and another for agent-based automation, while keeping a similar skill structure. The potential benefit is less duplication of workflow documentation and supporting resources.
Compatibility still depends on how each platform implements the format. A platform may support skill discovery, instructions, or bundled resources differently, and a specific agent can have separate permissions or tool limits. Users should confirm what a given product supports before relying on a skill for production work, especially where the agent can access code, files, or external services.
| Agent Skills component | Documented role | Why it matters |
|---|---|---|
| SKILL.md | Provides a name, description, and instructions. | Gives the agent the core information needed to identify and follow a workflow. |
| Scripts | Can be included as bundled resources. | May support repeatable actions, but require security review before use. |
| References | Can be included as bundled resources. | Provides task-specific context without placing all information in one prompt. |
| Templates | Can be included as bundled resources. | Helps standardize outputs for recurring tasks. |
Why does neutral governance matter for AI agent workflows?
Neutral governance matters because Agent Skills is intended for use across competing AI products rather than one vendor’s product line. Anthropic said engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag named lead maintainers. A multi-company maintenance model can give users more confidence that future changes are discussed through a shared process rather than decided solely by one product vendor.
The Agentic AI Foundation also announced Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba. The foundation said those additions brought it to 10 hosted projects from three at launch in December 2025. The Agentic AI Foundation announcement describes the projects as part of its broader agentic AI work.
Neutral governance does not mean every participating company will build identical tools or share identical priorities. The practical value is a more open process for maintaining a basic interoperability layer. Developers should follow the project’s releases and proposal process instead of assuming that a skill package will remain unchanged across future versions.
What security risks come with installing Agent Skills?
Agent Skills can create security risks because a skill may include instructions, scripts, references, templates, and other bundled resources that influence an agent’s behavior. A skill package is not automatically safe merely because it uses an open format or comes from a public repository. The safe response is to inspect the SKILL.md file and every included resource before allowing a skill to access sensitive files, credentials, business systems, or external tools.
Anthropic proposed a roadmap that starts with versioned releases, a formal proposal process, and security guidance for people who create and install skills. That focus is important because reusable agent workflows can spread quickly between teams and platforms. Security guidance can establish better review practices, but it cannot verify that every community-created skill is trustworthy.
Developers should start with skills from known maintainers, review scripts for unexpected network access or credential handling, and test the workflow in a limited environment. The risk resembles the broader problem of granting AI agents access to commercial sites and tools, where agent security failures can affect systems beyond the original prompt. Stop and contact an organization’s security team before deploying an unreviewed skill in an environment containing customer data, production credentials, or financial systems.
What should developers do after the Agent Skills handoff?
Developers should evaluate Agent Skills as a portable workflow format, not as a reason to automatically install large numbers of packages. Anthropic says 1 million skills have been published and nearly 280 million have been installed, but those are company-reported figures. The figures indicate substantial claimed adoption, while they do not establish the quality, safety, or maintenance status of each individual skill.
Developers can take 4 practical steps before adopting a skill:
- Review the skill’s name and description to confirm that the stated purpose matches the intended task.
- Read the full SKILL.md file and identify the inputs, permissions, tools, and outputs the workflow expects.
- Inspect bundled scripts, templates, references, and files before running them in an agent environment.
- Test the skill with limited permissions before connecting it to production code, customer information, or external services.
Agent Skills can make repeatable agent work easier to share across tools, particularly for teams that use more than one AI platform. The limitation is that portability does not replace operational controls. The most sensible approach is to adopt the standard gradually, document approved skills, and treat each package as software that needs review and maintenance.
FAQ
What is Anthropic Agent Skills?
Anthropic Agent Skills is an open standard for packaging specialized AI agent knowledge and workflows in a folder centered on a SKILL.md file. A skill can also include scripts, references, templates, and other resources for a defined task.
Did Anthropic give up control of Agent Skills?
Anthropic transferred Agent Skills governance to the Agentic AI Foundation, including the name and logo, GitHub organization, and agentskills.io domain. Anthropic engineers remain involved as individual maintainers alongside people from other companies.
Which products support Agent Skills?
Anthropic says more than 70 agent platforms have integrated Agent Skills, including ChatGPT, Cursor, GitHub Copilot, and VS Code. Platform support can vary, so developers should confirm the features available in their specific product.
Are Agent Skills safe to install?
No, Agent Skills are not automatically safe to install because packages can include instructions, scripts, and other files. Review every component and test with limited permissions before allowing a skill to access sensitive data or production systems.
Why does the Linux Foundation group matter?
The Agentic AI Foundation provides a neutral governance structure for Agent Skills rather than leaving the standard solely under Anthropic’s control. The arrangement can support a shared maintenance process for companies building different AI agents.