Agent Skills joins the Agentic AI Foundation: what changes
2026-10-22 · 4 min read
On October 22, Anthropic handed Agent Skills, the SKILL.md format that teaches an AI agent a specific job, to the Agentic AI Foundation (AAIF), which sits under the Linux Foundation. For anyone already using skills, nothing breaks. The move doesn't change the spec, and Anthropic says you can keep using or contributing to the standard without joining the foundation.
The foundation announced it on X the same morning. @AgenticAIFdn posted "Welcome to AAIF, Agent Skills, Open Responses, OpenEnv, and OpenSandbox!" and said the four additions bring it to 10 hosted projects in its first year. Anthropic's own announcement has the details.
What actually moved
Anthropic is transferring the specification, the Agent Skills name and logo, the GitHub organization and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals. David Soria Parra and Mahesh Murag are the lead maintainers. The roadmap Anthropic proposes is mostly housekeeping: versioned releases, a formal proposal process, and security guidance for people who write and install skills.
A neutral home makes sense because of how far the format has spread. Anthropic says more than 70 agent platforms support Agent Skills, including ChatGPT, Cursor, GitHub Copilot, VS Code, Codex and OpenCode, with about 1 million skills published and nearly 280 million installs. A format that size is shared infrastructure now, and AAIF already holds MCP, the connector protocol Anthropic donated last December.
This also settles something we covered in August. When OpenAI, Google, AWS and Microsoft agreed on a common plugin format that leaned on Agent Skills, Anthropic wasn't part of that group, and that looked awkward. Putting skills under a neutral foundation, with Google and OpenAI engineers as maintainers, clears up who controls it.
Simon Willison called it back in December 2025, when he wrote that he wouldn't be surprised to see skills end up next to MCP at AAIF.
The four new AAIF projects
| Project | Contributed by | What it does |
|---|---|---|
| Agent Skills | Anthropic | Packages instructions, scripts and files an agent loads on demand |
| Open Responses | OpenAI | One request format that works across model providers |
| OpenEnv | Hugging Face | Shareable environments for training agents with reinforcement learning |
| OpenSandbox | Alibaba | Isolated sandboxes for running agent tasks, on Docker or Kubernetes |
Per the AAIF's blog post, as reported by Unite.AI, the spec is licensed under Apache 2.0 and CC BY 4.0.
How to write an Agent Skill
The format is small. A skill is a folder whose name matches the skill, with a SKILL.md file inside. Optional scripts/, references/ and assets/ folders can sit next to it. The specification requires two frontmatter fields:
name: up to 64 characters, lowercase letters, numbers and hyphens, matching the folder name.description: up to 1,024 characters, saying what the skill does and when to use it. The agent reads this to decide whether to load the skill, so put real keywords in it.
Here is a skill a service business could actually use:
---
name: quote-follow-up
description: Drafts a follow-up email for an unanswered quote. Use when the user mentions a quote, estimate or proposal that has had no reply.
---
1. Ask for the customer name, job, quote amount and date sent.
2. Write a short email under 120 words that restates the job and the price.
3. Offer one specific next step, such as a time to talk this week.
4. Never invent discounts or deadlines.
Only the name and description load at startup, which the spec puts at about 100 tokens per skill. The full body loads when a task calls for it. The spec suggests keeping SKILL.md under 500 lines and moving long material into references/. To check your frontmatter, the official skills-ref tool runs skills-ref validate ./quote-follow-up.
In Claude Code, personal skills go in ~/.claude/skills/ and project skills in .claude/skills/. Other agents that support the format look in their own folders, but the skill itself copies over unchanged. If you've seen what people build with Claude Code mods, skills are the simpler piece, since a useful one can be nothing more than a page of written instructions.
What this means for a small business
A skill that writes down one of your processes, say how you quote or how you onboard a new client, no longer ties you to one vendor. The format belongs to a neutral group now, with engineers from the big labs maintaining it together. That makes it a reasonable place to keep the way your business works, alongside the connectors in Claude's marketplace that hook an agent up to your apps.
Picking which process to write down first is usually the hard part. If you want help with that, our free process audit goes through your week and points to the tasks where a skill would save the most time.