What the new agent skills standard means for your business

Agent Skills, the format Anthropic built so an AI agent can follow a written procedure instead of guessing, just left Anthropic's control. On October 22 Anthropic handed the standard to the Agentic AI Foundation, the Linux Foundation group that already stewards the Model Context Protocol, putting engineers from Google, OpenAI, and outside companies in charge of where the format goes next.

If you have not run into the term, a skill is a folder of instructions, scripts, and reference material that tells an AI tool exactly how your business wants a task done: how you format a client proposal, how you triage a support ticket, how you close out a job in your CRM. Claude Code popularized the idea this year, and it caught on because you write the procedure once instead of re-explaining it in every new chat.

What actually changed on October 22

Anthropic gave up the Agent Skills name, logo, GitHub organization, and the agentskills.io domain, and says it is committed to keeping the standard open, community-driven, and vendor-neutral. David Soria Parra and Mahesh Murag are the new lead maintainers, and engineers from Anthropic, Google, OpenAI, and other companies will maintain the spec as individuals, not as representatives of their employers. Anthropic reports the format is already used by more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, with over 1 million skills published and close to 280 million installs.

Why agent skills matter for a small operation

Most owners who have written an SOP, a script for the phone, or a checklist for onboarding already have the raw material for a skill. The difference is that a skill is written so an AI agent can read it and carry out the steps, not just a person. Until this week, writing that procedure for one assistant meant redoing it if you tried a different tool next quarter. A shared, open format means the same skill file should keep working if you move from Claude to ChatGPT, or add Copilot for a different team, because the platforms agreed to read the same thing.

What to do with this in the next week

1. 1

   Find one procedure you already repeat

   Pick something you explain the same way every time: quoting a job, writing a weekly report, answering a common support request.

2. 2

   Write it down as steps, not prose

   List the inputs, the order of operations, and the exact output format you expect. That structure is what makes it readable by an agent.

3. 3

   Test it in whatever tool you already pay for

   Point Claude, ChatGPT, or Copilot at the folder and run the task. Note where it guesses wrong, then tighten the instructions there.

4. 4

   Keep the file portable

   Avoid tool-specific syntax where you can. The point of an open standard is that the same file travels with you to the next tool.

None of this requires picking a side in the AI platform fight. It just means the procedure you write down this month has a better shot at still being useful next year, in whatever tool you end up running it in. If there is a process you are still explaining by hand every time someone new joins, that is a good one to bring, and we can turn it into a portable skill together.