In this post 9 sections
On 22 October 2026, Anthropic donated the Agent Skills standard to the Agentic AI Foundation (AAIF), a Linux Foundation project, handing over the name, logo, GitHub organization and the agentskills.io domain. Nothing changes in how a skill is built: it is still a folder with a SKILL.md file. What changes is who maintains the standard, and that makes it a reasonable moment for marketing teams to set rules for the skills they share.
Key Takeaways
- Anthropic's announcement says the donation covers the Agent Skills name and logo, the GitHub organization and the agentskills.io domain.
- Anthropic reports about 1 million published skills, nearly 280 million installs and more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code. These are Anthropic's figures; I found no independent count.
- Individual engineers from Anthropic, Google, OpenAI and other companies will maintain the standard, with David Soria Parra and Mahesh Murag as lead maintainers, per the same post.
- The proposed roadmap includes versioned releases, a formal proposal process and security guidance for skill authors and installers. That guidance does not exist yet.
- On Claude Team and Enterprise plans, admins can provision skills centrally; provisioned skills are on by default and users can turn them off.
- A skill can bundle scripts, so a shared skill is closer to installed software than to a saved prompt, and deserves a review step.
What an Agent Skill is
The agentskills.io overview defines a skill as a folder containing a SKILL.md file with at minimum a name and description, plus instructions. It can also hold scripts/, references/ and assets/ directories. Agents load skills in three stages: discovery (only name and description at startup), activation (the full instructions when a task matches) and execution (running bundled code or loading referenced files). Because full instructions load only when needed, an agent can keep many skills available without filling its context.
For a marketer, the useful way to think about it is a runbook the agent reads on demand. Examples: a technical SEO audit procedure, a brand-voice checklist, a weekly reporting routine. I wrote a worked example in Claude Code skills for SEO audits.
What the donation actually changes
SD Times confirms the announcement came on 22 October 2026 with three other projects (Open Responses from OpenAI, OpenEnv from Hugging Face, OpenSandbox from Alibaba), bringing AAIF to ten projects. The foundation also hosts MCP, which I cover in MCP in October 2026.
| Question | Before 22 October | After |
|---|---|---|
| Who owns the name, logo, repo and domain? | Anthropic | AAIF, under the Linux Foundation |
| Who decides spec changes? | Anthropic, with community input | Maintainers acting as individuals, with a proposed formal proposal process |
| Do existing SKILL.md files still work? | Yes | Yes; Anthropic says the build format is unchanged |
| Is there security guidance for skill authors? | Not in a standard form | Proposed on the roadmap, not published |
The vendor-neutral framing matters for one practical reason. If you build a skill library, you are betting that the format stays readable by more than one tool. Anthropic's list of more than 70 platforms is the evidence for that bet, and the move to neutral stewardship lowers the risk that one company changes the format on its own. It does not remove it: maintainers still come largely from a few large companies.
How Claude lets organizations share skills
Anthropic's organization skills article, dated 18 December 2025, says admins on Team and Enterprise plans can provision skills from admin settings. Provisioned skills are enabled by default for all users, and any user can switch one off for themselves. It also describes a partner skills directory at claude.com/connectors with skills from companies such as Notion, Canva, Figma and Atlassian, which admins can provision organization-wide.
Two things I could not verify from Anthropic's pages. Secondary guides claim that once organization-wide sharing is on, member-published skills need no approval step, and one July 2026 hands-on write-up reported that member-published skills did not appear in the admin console section. Those come from vendor blogs and a single tester. Check your own admin settings before you rely on either statement.
A governance checklist for a marketing team
I have not run a skills program at scale, so this is a proposal from the way the format works, not a report of results. A skill folder is small enough to review in minutes. Use that.
- One named owner per skill. Put the owner and a "last reviewed" date in the skill's description or a README in the folder.
- Keep skills in a repository, not a chat. Version control gives you history, review and rollback. A skill that exists only in one person's account cannot be audited.
- Review anything with a `scripts/` folder like code. A script runs with whatever access the agent has. Read it, and list what files and networks it touches.
- Write narrow descriptions. The description decides when the skill loads. A vague one fires on the wrong tasks. "Audit title tags and meta descriptions on a supplied URL list" beats "SEO helper".
- Separate drafting skills from publishing skills. A skill that writes copy is low risk. One that pushes to your CMS or ad account should need a human approval step.
- Set a retirement rule. Review each skill quarterly and delete the ones nobody used. Stale instructions produce confident, outdated work, which is the same decay problem I describe in governing a team prompt library.
- Pin facts that change. Do not hard-code prices, rankings or policy dates in a skill. Point it to a source file that someone updates.
The threat that makes steps 3 and 5 matter is already documented for agents in general. If a skill or a page it reads can carry instructions, a permissive agent may follow them; see AI agent security for marketing teams.
What this means for marketers and founders
- If you already use skills: do nothing urgent. Anthropic says the format is unchanged. Spend the week on the checklist above, starting with an inventory.
- If you have not started: begin with one skill for the task you repeat most, such as a monthly content audit. Test it on three real inputs and compare against your manual output before sharing it.
- If you lead a team: decide who may provision skills for the whole organization before anyone does it casually. Default-on provisioning means one bad skill reaches everyone.
- If you sell tools or content: a public skill is a distribution channel. Watch whether the foundation's security guidance arrives; it will probably set the bar for what directories accept. My notes on directory submission are in the plugin directory guide.
- Leave alone: switching tools because of the donation. It is a governance change, not a capability change.
What we do not know yet
- When the roadmap items (versioned releases, security guidance) will ship. No dates were given in the announcement I read.
- Whether the 1 million skills and 280 million installs figures count distinct, active use. They are Anthropic's numbers, with no definition I could find.
- How admin approval and visibility work today for member-shared skills. Anthropic's own article does not say who can share.
- Whether other large vendors will change their skill support as maintainers. The post says engineers from several companies will maintain it, which is a statement of intent.
FAQ
What did Anthropic announce about Agent Skills on 22 October 2026?
Anthropic donated the Agent Skills open standard to the Agentic AI Foundation, which sits under the Linux Foundation. The transfer covers the name, logo, GitHub organization and the agentskills.io domain. The skill format itself is unchanged.
Do my existing Claude skills still work?
Yes. Anthropic's announcement says nothing changes in how a skill is built: a folder with a SKILL.md file and any supporting scripts or reference material. There is no migration step in the post.
What is the Agentic AI Foundation?
It is a Linux Foundation project co-founded by Anthropic, Block and OpenAI that also hosts the Model Context Protocol. SD Times reports it grew to ten projects with the 22 October additions.
Who can share skills across a company on Claude?
Admins on Team and Enterprise plans can provision skills centrally, and provisioned skills are on by default for users. Anthropic's article does not state who else can share. Check your admin settings.
Are shared skills a security risk?
They can be. A skill may bundle scripts, and a script runs with the agent's access. Treat skills with scripts like code: review them, keep them in version control and require approval before one can publish or spend money.
How many Agent Skills exist?
Anthropic reports about 1 million published skills and nearly 280 million installs across more than 70 platforms. I found no independent count, so treat the figures as the company's own.
Do I need Claude Code to use skills?
No. The standard is supported across many tools, and Claude Team and Enterprise plans can provision skills from admin settings. Claude Code is one place to build and test them.
Want help with this?
I'm Younus Fardeen, an organic growth, SEO and AEO strategist with 4+ years of marketing experience across edtech and startup teams. If you want help turning your repeatable SEO, content or reporting workflows into reviewed, shareable skills, see my work and reach me through the contact form at younusfardeen.com.
Found this useful? Pass it on.