Agent Skills are an open format for teaching AI agents a specific task. A skill is a folder containing a SKILL.md file, with a name, a description and step-by-step instructions, plus any scripts, templates or reference files the task needs. Anthropic created the format, and on 22 October 2026 it donated Agent Skills to the Agentic AI Foundation under the Linux Foundation, so it's now a vendor-neutral standard used by more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code.
If you've used Claude or ChatGPT for real work, you'll know the problem skills fix. The AI knows a lot in general, but it doesn't know how your business does things: your proposal layout, your brand voice, the checks you run before a report goes out. You end up explaining it again every time.
A skill writes that know-how down once, in a format any compatible agent can pick up. Here's what that means in practice, and why this week's news matters if you run a business rather than a dev team.
What changed on 22 October 2026
Anthropic announced it's donating Agent Skills to the Agentic AI Foundation (AAIF), the body under the Linux Foundation that it co-founded with Block and OpenAI, and the same home it gave the Model Context Protocol. It's handing over the Agent Skills name and logo, the GitHub organisation and the agentskills.io domain.
| Fact | What Anthropic says |
|---|---|
| Who made it | Anthropic, while working on Claude Code |
| First launch | October 2025, published as an open standard in December 2025 |
| Adoption | More than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code |
| Scale | About 1 million skills published and nearly 280 million installed |
| Who maintains it now | Engineers from Anthropic, Google, OpenAI and other companies, as individuals |
| Do you need to join? | No. You can use or contribute to the standard without joining the AAIF |
The proposed next steps are versioned releases, a formal proposal process and security guidance for people who write and install skills. That last one matters, and I'll come back to it.
How a skill works inside an agent
The clever part is that an agent doesn't read every skill all the time. The standard calls this progressive disclosure, and it happens in three stages:
- Discovery. When the agent starts, it loads only the name and description of each skill, roughly a hundred tokens each, just enough to know when one might help.
- Activation. When your request matches a skill's description, the agent reads the full SKILL.md instructions.
- Execution. It follows the instructions, running any bundled scripts or opening reference files only if the task needs them.
That's why you can keep lots of skills installed without slowing anything down, and why the description is the most important line you'll write. If it's vague, the agent won't know when to use the skill.
The rules for a SKILL.md file
The specification is short. These are the parts that trip people up:
- Name: up to 64 characters, lowercase letters, numbers and hyphens only, no hyphen at the start or end, no double hyphens, and it must match the folder name.
- Description: up to 1,024 characters, saying what the skill does and when to use it, with the words someone would naturally type.
- Instructions: any Markdown you like after the frontmatter. The spec recommends keeping the main file under 500 lines and moving long reference material into separate files.
- Optional folders: scripts for code the agent can run, references for extra documents, assets for templates, images and data.
A first skill a business owner could write
You don't need to code to get value from this. Here's the shape of a simple skill for writing client proposals in your house style. Save it as SKILL.md inside a folder called proposal-writer:
Example SKILL.md
---
name: proposal-writer
description: Writes client proposals in our house style. Use when asked to draft, rewrite or check a proposal, pitch or scope document.
---
1. Ask for the client name, their goal and the scope if they're missing.
2. Open with the client's goal in their own words, then our recommendation.
3. Use our section order: Goal, Approach, Timeline, What we need from you, Next step.
4. Plain British English, short paragraphs, no jargon.
5. Finish by listing anything you weren't sure about instead of guessing.
Notice the description does the heavy lifting. It tells the agent exactly which requests should trigger the skill. The numbered steps then capture the bit of your process that usually lives in your head.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Where you can use skills
Because it's an open standard, the same folder can work across tools that support it. Claude, where the format started, uses them, and in ChatGPT skills live in the Plugins area for eligible workspace plans. Gemini has its own skills feature too, which I cover in my guide to Gemini skills.
If you're building something more ambitious, skills pair naturally with agents. My beginner's guide to building AI agents with Claude shows where they fit.
The catch: a skill can run code
Skills aren't just text. They can bundle scripts that an agent runs on your behalf, which is exactly why the new roadmap includes security guidance. Treat a skill from a stranger like any other software: read it before you install it, check where it came from, and be wary of anything that asks for more access than its job needs.
My verdict: this is the most useful AI standard most business owners haven't heard of. Pick one task you explain to AI every week, write it down as a skill, and test it. If you'd like help turning your processes into skills your team can share, that's exactly what I do in my one-to-one AI sessions.
Frequently asked questions
What are Agent Skills?
Agent Skills are an open format for giving AI agents specialised knowledge. Each skill is a folder with a SKILL.md file containing a name, a description and instructions, plus optional scripts and reference files.
Who owns Agent Skills now?
Anthropic donated Agent Skills to the Agentic AI Foundation under the Linux Foundation on 22 October 2026, including the name, logo, GitHub organisation and the agentskills.io domain.
Which tools support Agent Skills?
Anthropic says more than 70 agent platforms have integrated the standard, including ChatGPT, Cursor, GitHub Copilot and VS Code.
What goes in a SKILL.md file?
YAML frontmatter with at least a name and a description, followed by Markdown instructions. The spec suggests keeping the main file under 500 lines.
Are Agent Skills safe to install?
Skills can include code, so review any skill from an outside source before installing it. The proposed roadmap includes security guidance for people who write and install skills.
