Gareth MorgansProduct
What Claude Agent Skills mean for teams
Claude Agent Skills began inside Claude Code. On 22 October 2026, Anthropic donated the Agent Skills standard to the Agentic AI Foundation, so a team can keep one skill and use it in the tools that support the format.
If you've already kept a Claude copy and a Cursor copy of the same guidance, you can now write it once.
Start free beta · Skills registry
Related: what is a skills registry · skill portability · manage skills across Claude and Cursor
What changed on 22 October
Anthropic handed the standard to the Agentic AI Foundation (AAIF), which sits under the Linux Foundation. It co-founded the AAIF with Block and OpenAI, and had already donated the Model Context Protocol (MCP) to it. The donation covers the Agent Skills name and logo, the GitHub organisation, and the agentskills.io domain.
The file format is unchanged. A skill is still a folder with a SKILL.md file, plus optional scripts, references and other supporting files. You can keep using the standard, or contribute to it, without joining the AAIF. The specification is published under the Apache 2.0 and CC BY 4.0 licences.
More than 70 agent platforms have integrated it since the standard was published in December 2025, including ChatGPT, Cursor, GitHub Copilot and VS Code. Anthropic also reports that about 1 million skills have been published and nearly 280 million have been installed. Those publication and install figures are Anthropic's own.
David Soria Parra and Mahesh Murag are the lead maintainers. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals. A technical steering committee at the AAIF will guide what comes next.
Anthropic has proposed a roadmap that covers versioned releases, a formal proposal process, and security guidance for people who write and install skills. That roadmap hasn't shipped.
Source: Anthropic's announcement, 22 October 2026 and the AAIF welcome post.
What Claude Agent Skills actually are
People search under that name because the format began in Claude Code. Anthropic launched skills there in October 2025. Within weeks, developers had published thousands of skills on GitHub and written loaders for Codex, Cursor and OpenCode.
Anthropic then published Agent Skills as an open standard in December 2025, and the standard is now vendor-neutral. If you use skills in Claude Code or Cursor, see managing skills in Claude Code and Cursor.
The Agent Skills specification is short. A skill is a folder containing a SKILL.md file. That file holds metadata, at minimum a name and a description, plus the instructions that teach an agent how to do a particular job. The folder can also bundle scripts, reference material, templates and other files the agent may need.
Compatible agents load skills in stages, which the specification calls progressive disclosure. When the agent starts, it reads only the name and description of each available skill, enough to know when one might be relevant. If a task matches, it loads the full SKILL.md instructions. It then follows those instructions, and may run bundled code or open a referenced file if the task needs it.
A Claude skill, in the sense people usually mean, is a named package of instructions, and any files that travel with it, written so a compatible agent can pick it up when the work calls for it. Once the format is open, the same skill package can travel with the team, rather than being rewritten for each tool.
Agent Skills are not the same as MCP
Skills package procedural knowledge: how your team does a job, what a good result looks like, and which steps to follow. MCP connects an agent to tools and data. One holds the method. The other provides the access. Our explainer on what an MCP server is covers that side of the stack.
The two can work together. There is now an official MCP Skills extension, known as SEP-2640, which lets MCP servers expose skills so a client can discover them and read their files. It defines methods such as skills/list and a skill:// address scheme. The extension is marked final, though the documentation notes that support in software development kits and host apps is still being implemented. Joyst doesn't serve that extension.
Approved MCP servers, meanwhile, sit on the team's MCP catalogue as a list, not as a proxy in front of those servers.
What teams should do now
A shared format makes it easier to move a skill from one tool to another. Your team still has to decide which skills to trust.
Keep one approved list of skills, and treat that list as the source of truth. If a skill isn't on it, it isn't in use. A skills registry is the usual place to hold that list.
Review third-party skills before anyone on the team relies on them, especially any that bundle scripts. A skill is more than a SKILL.md file. Bundled code runs with the access the agent already has, so you need to know what it does before it becomes part of everyday work.
Version and share from one place, rather than forking a copy for each tool. See version and share skills, skill portability and manage skills across AI tools. Otherwise one edit lands in Claude, another stays in Cursor, and nobody is sure which copy is current. We wrote about that earlier in manage skills across Claude and Cursor.
Someone on the team still needs to own each skill, including who may change it. We go into that in ownership of skills and prompts. If you assemble skills with prompts and MCP servers into a reusable setup, agent packages is the next step, and the agents page shows how those packages are put together.
How Joyst fits
Joyst is a shared workspace for the skills, prompts, MCP servers and secrets your team already uses. It doesn't invent a new skill format, and it doesn't connect to ChatGPT.
You can import a raw SKILL.md as a private draft. It stays private until someone reviews it and shares it.
Downloads go out through the Joyst MCP, and only for Claude Code, Cursor, Codex CLI, Copilot and Windsurf. Those are the tools Joyst can translate a skill into today.
Approved skills then sit in the team's skills registry beside prompts, MCP servers and secrets, so everyone works from the same list. You can read more in what is a skills registry.
The open standard decides how a skill is packaged. Your team still decides which ones are approved, who can change them, and which tools they are allowed to reach.
FAQ
What are Claude Agent Skills?
Claude Agent Skills are skills written in the Agent Skills format that began in Claude Code and is now an open standard. Each one is a folder with a SKILL.md file, and optional scripts or references, that teaches a compatible agent how to do a specific job.
Did Anthropic still own Agent Skills after 22 October 2026?
No. On that date Anthropic donated the Agent Skills name, logo, GitHub organisation and agentskills.io domain to the Agentic AI Foundation, under the Linux Foundation. Anthropic engineers still help maintain the standard, as individuals, alongside people from other companies.
Are Claude Agent Skills the same as Claude Code skills?
The names overlap because the format started in Claude Code. That name now refers to the open format. Skills you run inside Claude Code are one product's use of it. The same skill file can be used in other tools that have adopted the standard. If you use skills in Claude Code or Cursor, see managing skills in Claude Code and Cursor.
Do Agent Skills replace MCP?
No. Skills hold the instructions for a task. MCP connects the agent to tools and data. An official MCP extension can expose skills from a server, but that is a way to discover and load them, not a replacement for either standard.