Agent Skills moves to a foundation. A skill is just a folder
Anthropic announced on October 22, 2026 that it is donating Agent Skills to the Agentic AI Foundation. The idea to keep: a skill is a folder with one required file, so the format is small enough to read in one sitting, whoever ends up governing it.
What was announced
The announcement is on Anthropic's own site, in its news section. As far as I can tell from the sources, that page is the only place the donation is described, so everything below about the donation is "Anthropic says", not a confirmed decision by the foundation.
The foundation itself is not new. Anthropic's earlier post says the Agentic AI Foundation was set up on December 9, 2025 under the Linux Foundation, with Anthropic, Block and OpenAI as founding members. At the same time Anthropic donated MCP, the Model Context Protocol, which it describes as an open standard for connecting AI applications to external systems.
Agent Skills has a different job. The project site describes it as an open standard for giving agents specialized knowledge and workflows, originally developed by Anthropic and open to contributions. My reading, not Anthropic's wording: MCP is how an agent reaches a system, and a skill is what the agent knows about doing a particular job. Two standards, two jobs, one foundation.
A skill is a folder
The specification defines a skill as a folder containing a SKILL.md file. Everything else is optional:
release-notes-writer/
├── SKILL.md required
├── scripts/ optional: code the agent can run
├── references/ optional: documentation
└── assets/ optional: templates, images, data
SKILL.md has two required fields at the top, name and description. The name is 1 to 64 characters, uses only lowercase letters, digits and hyphens, and has to match the folder name. The description is 1 to 1024 characters. Here is a small example I made up to show the shape:
---
name: release-notes-writer
description: Drafts App Store release notes from a git log. Use when asked for "what's new" text for an app update.
---
Instructions for the agent go here, in plain Markdown.
That is a complete, valid-looking skill. The part below the two fields is whatever you want the agent to know once the skill is in use; the spec leaves the content to you.
How an agent loads a skill
The spec describes three stages. In discovery, the agent loads only the name and the description. In activation, it reads the whole SKILL.md. In execution, it opens scripts, references and assets only when the task needs them.
This explains why the description matters more than it looks. In the first stage it is the only thing the agent has to decide whether a skill fits the task. That is my inference from the three stages, and the advice that follows is mine too: write the description like a label that says what the skill does and when to use it, not like a title.
What changes, and what does not
Nothing in the file format changes with this announcement, as far as the sources show. What changes is who holds the project: Anthropic says it is moving to a neutral foundation, the same one that now holds MCP.
Whether that matters in practice depends on things I cannot check from these sources, such as how many tools will keep supporting the format and how the foundation will run it. Anthropic's page also gives adoption figures. They are Anthropic's own counts with no independent tally, so I left them out. If you build skills today, the safe reading is narrow: the spec you can read on agentskills.io is the thing you are writing to.
A ten-minute experiment
Pick one task you explain to an AI agent again and again, such as writing release notes, and make it a skill. Create the folder, write the two fields, and check the two rules the spec is strict about:
dir=release-notes-writer
grep -m1 '^name:' "$dir/SKILL.md"
echo -n "$dir" | wc -c
The first line should print the same name as the folder. The second prints the folder name length, which has to be 64 or less. Then point an agent tool that supports skills at the folder and ask for the task in your own words, without naming the skill. If the agent picks it up, your description works. If it does not, rewrite the description first, before touching the instructions below it.
Where this leaves things
A skill is a folder with a SKILL.md, two required fields and three optional directories. Anthropic says the standard is moving to a foundation, and that is the news. The format you can read today is the same one. If you already keep notes for your agent in a long prompt, splitting one repeated task into a skill is a cheap way to see what the format feels like.
Sources
- Donating Agent Skills to the Agentic AI Foundation — Anthropic — the October 22, 2026 announcement that Anthropic is donating Agent Skills to AAIF (single source for the donation)
- Agent Skills — agentskills.io — a skill is a folder containing a SKILL.md file; open standard for specialized knowledge and workflows, originally developed by Anthropic; the three loading stages
- Agent Skills specification — agentskills.io — required
name(1 to 64 characters, lowercase letters, digits, hyphens, matches the folder name) anddescription(1 to 1024 characters), optional scripts/, references/ and assets/ - Donating the Model Context Protocol and establishing the Agentic AI Foundation — Anthropic — AAIF founded on December 9, 2025 under the Linux Foundation by Anthropic, Block and OpenAI; MCP donated at the same time; MCP described as an open standard for connecting AI applications to external systems