Home Services Work About Blog FAQ Contact
← Back to blog

Anthropic gave away Agent Skills. Here's what changes for your automation stack.

On October 22, Anthropic handed Agent Skills — the format that gives Claude, ChatGPT, Cursor, and dozens of other AI tools their task-specific know-how — to an independent foundation. If you're already running an AI agent built on skills, or thinking about building one, the ownership change matters more than the press release lets on.

Here's the short version: Anthropic doesn't own Agent Skills anymore. It gave away the name, the logo, the GitHub organization, and the agentskills.io domain to the Agentic AI Foundation, a neutral group that operates under the Linux Foundation. Anthropic co-founded that foundation last year with Block and OpenAI, and it's already home to the Model Context Protocol, which we wrote about back in June.

What a skill actually is

A skill is a folder with one required file: SKILL.md. That file tells an agent how to do one specific task or work inside one specific domain, and it can carry along scripts or reference material the agent needs to finish the job. Anthropic built the format while working on Claude Code, because the agent was already good at general knowledge work but kept stumbling on the procedural stuff — the way your business actually handles a refund, or the exact steps your team follows when onboarding a new client.

Write that procedure once as a skill, and any agent that reads SKILL.md picks it up. No retraining. No rewriting the same instructions into a new prompt every time you switch tools.

A skill folder is deliberately boring: a short description at the top, plain instructions below, and whatever supporting files the task needs — a script, a template, a reference doc. Anthropic kept the format text-first on purpose, which means a skill is something a non-engineer can read, write, and hand off. You don't need to know how to code to write one. You need to know your own business process well enough to write it down clearly.

This matters most once you're running more than one AI tool. One person using Claude for drafting emails doesn't need any of this. A business running Claude for proposals, an n8n agent for lead intake, and a Copilot instance for code review does — because without a shared format, the same SOP gets written three times, in three incompatible prompt styles, and drifts out of sync the first time someone updates it in only one place.

The adoption numbers are the real story

Skills launched in October 2025. Developers started publishing them on GitHub within days, and people wrote loaders so Codex, Cursor, and OpenCode could use them too — tools Anthropic never built support for directly. By December, Anthropic formalized the format as an open standard, with its own spec, GitHub org, and domain.

A year later: more than 70 agent platforms have integrated it, ChatGPT, Cursor, GitHub Copilot, and VS Code among them. Over a million skills have been published. Close to 280 million have been installed.

A format that goes from an internal tool for Claude Code to 280 million installs across 70 platforms in a year isn't a feature anymore. It's infrastructure.

Why handing over control is good news for you

We've watched this play out before. MCP followed almost the same path — Anthropic built it, open-sourced it, watched it get adopted everywhere, then handed governance to the same foundation. Within about a year of that handoff, n8n shipped native MCP support directly in its nodes panel. Tools don't wait for permission to adopt a standard nobody can yank out from under them. The pattern works because a standard one company controls is a standard everyone else is renting. The moment that company's roadmap stops lining up with yours, you're stuck rebuilding.

Under the Agentic AI Foundation, Agent Skills gets maintained by engineers from Anthropic, Google, OpenAI, and other companies, acting as individuals rather than company reps. David Soria Parra and Mahesh Murag are the lead maintainers. Nothing requires you to join the foundation to use or contribute to the standard — you just get the insurance of a spec that doesn't move every time one vendor changes strategy.

The foundation's stated roadmap adds three things worth watching: versioned releases, so a skill written against today's spec doesn't quietly break tomorrow; a formal proposal process for changes; and security guidance for people writing and installing skills. That last one is overdue. A skill is code and instructions you're letting an agent run, and we've already written about what happens when an agent trusts the wrong input.

What to actually do about this

If you're not running any AI agents yet, nothing changes for you today. The announcement doesn't touch Claude's pricing, access, or features.

If you are, here's where five minutes buys you something:

  1. Write down one procedure your team repeats constantly — how you qualify a lead, how you handle a refund, how you draft a proposal — as a SKILL.md file. It's plain markdown. You don't need an engineer to do this.
  2. Check whether the AI tools you already pay for support skills. If you're on Claude, ChatGPT, Cursor, or GitHub Copilot, the answer is probably yes.
  3. Before you install someone else's skill from GitHub, read it the way you'd read a contractor's invoice before paying it. The security guidance the foundation promised isn't live yet, which means you're the security guidance for now.

If you want help figuring out which of your workflows are worth turning into a skill-driven agent instead of a one-off automation, that's exactly the kind of audit we do.

— Cole

Sources

Already running AI agents and wondering what's worth standardizing?

A 30-minute discovery call covers what's worth turning into a reusable skill, what isn't, and what the realistic build looks like.

Book a Discovery Call →