MeanCEO: Tech Startups and Startup Ideas

Anthropic Gives Agent Skills to the Linux Foundation: The SKILL.md Format Behind 280 Million Installs Is Now Neutral. What Founders Should Build, and Check

TL;DR: What happened to Agent Skills?

On 22 October 2026, Anthropic handed Agent Skills, the open format for teaching AI agents a task with a SKILL.md file, to the Agentic AI Foundation, part of the Linux Foundation. The name, logo, GitHub organization and the agentskills.io domain now sit with a neutral foundation that also holds MCP. More than 70 agent platforms support the format, including ChatGPT, Cursor, GitHub Copilot and VS Code, and the foundation says skills on the skills.sh registry have been installed nearly 280 million times. For founders, this makes skills a safer thing to build and sell across tools. It doesn't make them safe to install: studies found serious flaws in more than one in ten public skills.
If you have written a prompt file that tells Claude Code how to write your release notes, or a folder of instructions that teaches Cursor your brand voice, you have already built an agent skill. Until last week, the format behind it belonged to one company. Now it doesn't.
That sounds like boring governance news. It isn't, if you build on AI agents. It decides whether the workflows you package today will still work when you switch from Claude to ChatGPT, Gemini CLI or Mistral's Vibe next year, and whether you can sell them to clients who use different tools. Here's what changed, what didn't, and what I'd do this week.

What happened on 22 October

In its announcement, Anthropic says it is donating Agent Skills, "the open standard that gives AI agents specialized knowledge and workflows", to the Agentic AI Foundation (AAIF). In Anthropic's words: "We are donating the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain."
  • Who maintains it: "Engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers." The foundation adds that a multi-company Technical Steering Committee will guide the project.
  • Who uses it: more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have added support since Anthropic published the standard in December 2025.
  • What's next: Anthropic has proposed a roadmap with "versioned releases, a formal proposal process, and security guidance for people who write and install skills." It is a proposal for now, not a published plan.
  • What you need to do: nothing to keep using it. "You can continue to use the standard or contribute to it without joining the AAIF."
Agent Skills was one of four projects the foundation took on that day. The Linux Foundation's press release, issued at AGNTCon + MCPCon North America in San Jose, lists OpenAI's Open Responses, Hugging Face's OpenEnv and Alibaba's OpenSandbox as the other three. The foundation launched in December 2025 with three projects and now has ten. David Soria Parra, who co-created MCP at Anthropic, said: "Without the right context, even the most intelligent agents fall short."
ProjectDonated byWhat it does
MCP (Model Context Protocol)Anthropic, Dec 2025Connects agents to tools and data
AGENTS.mdOpenAI, Dec 2025A file that tells coding agents how a repository works
gooseBlock, Dec 2025An open-source agent
Agent SkillsAnthropic, Oct 2026Packages instructions, scripts and files that teach an agent a task
Open ResponsesOpenAI, Oct 2026An open API format that works across model providers
OpenEnvHugging Face, Oct 2026Containers for building reinforcement learning environments
OpenSandboxAlibaba, Oct 2026Tools to limit and sandbox what agents can do
Sources: Linux Foundation press releases of 9 Dec 2025 and 22 Oct 2026; The Register, 23 Oct 2026. The foundation's other projects are A2A, Agent Router and agentgateway.

Where the 280 million installs come from

The big numbers need one caveat. Anthropic writes that "1 million skills have been published and nearly 280 million installed" without naming a source. The foundation's own blog post is more precise: "In only seven months, a million skills have been listed on the popular skills.sh registry, and these skills have been installed nearly 280 million times." skills.sh is a registry run by Vercel. So these are registry listings and installs, not active users, and many skills will be tiny tests or copies.
The same post shows how teams use skills in practice. At Block, engineers share an internal marketplace of "hundreds of skills" for incident response, code review and experiment setup, and they use them with Goose, Codex, Claude and Amp. At Brainlabs, staff in any role submit skills through Notion, where a human reviews them before they go out. That second example is the one to copy if you run a small team: the person who knows the job writes the skill, someone else checks it.

What a skill is, in plain words

A skill is a folder. According to the Agent Skills specification, it must contain one file, SKILL.md, with a short header and plain Markdown instructions below it. It can also hold scripts, reference documents and assets such as templates. The header needs only two fields: a name of up to 64 characters (lowercase letters, numbers and hyphens, matching the folder name) and a description of up to 1,024 characters that tells the agent when to use it.
The clever part is how little it costs to keep many skills around. At start-up, the agent loads only each skill's name and description, about 100 tokens each. It reads the full instructions, under 5,000 tokens is recommended, only when a task matches, and opens the extra files only if it needs them. The spec suggests keeping SKILL.md under 500 lines. So you can give an agent fifty skills without filling its context window.
That is why skills suit bootstrapped teams. You don't need a developer to write one. If you can explain your invoicing routine, your onboarding email sequence or your SEO brief to a new hire, you can write it as a skill, and any compatible agent can follow it.
Type of toolExamples that support Agent Skills
Big-lab agentsClaude Code, ChatGPT and Codex, Gemini CLI
Code editorsCursor, GitHub Copilot, VS Code, JetBrains Junie, Kiro
Open-source agentsGoose, OpenCode, OpenHands, OpenClaw
EuropeanMistral AI Vibe
Data platformsDatabricks Genie Code, Snowflake Cortex Code
FrameworksSpring AI, Laravel Boost
Source: agentskills.io client showcase (46 tools listed on 26 Oct 2026; Anthropic counts more than 70 platforms).
There's a European option too. Mistral's documentation for Vibe Code skills says Vibe follows the Agent Skills specification, so skills made there work with other agents that use the same standard. If you need to keep your agent stack with a European provider, your skill library can move with you.

The catch: skills are a new supply chain

Here's the part the press releases skip. A skill runs with whatever access your agent has: your files, your terminal, your API keys. Installing one from a stranger is like running a stranger's script, except that the dangerous part can be a sentence in plain English.
Security firms have been measuring this all year. In its ToxicSkills study of 3,984 skills from ClawHub and skills.sh, Snyk found that 13.4% (534 skills) had at least one critical security issue and 36.82% had at least one flaw of any severity. Its researchers confirmed 76 skills with malicious payloads in their instructions. Snyk also counted daily skill submissions rising from under 50 in mid-January to over 500 by early February. A January study on arXiv, "Agent Skills in the Wild", scanned 31,132 skills and found at least one vulnerability in 26.1% of them, with 5.2% showing high-severity patterns that suggest malicious intent. Skills that bundle scripts were 2.12 times more likely to be vulnerable than instruction-only ones.
And it hasn't stopped. On 21 October, the day before the donation, JFrog said its researchers had tracked 969 AI agent skills "carrying high-impact payloads". We've seen the same pattern this month with tools founders trust: Rank Math PRO shipped malware to WordPress sites, and the Railway and LiteLLM breaches put AI API keys at risk. Anthropic's own advice is to install skills only from trusted sources and to audit anything else before use.
This is why the roadmap item on "security guidance for people who write and install skills" matters more than the logo transfer. Today the spec page shows no version number, and, as Snyk points out, skills come with no code signing and no security review by default. A neutral foundation with Anthropic, Google and OpenAI engineers at the table is the most likely place for those to come from. Until they do, checking skills is your job.

Is this real independence?

Not everyone is impressed. The Register titled its report "Vendor-affiliated AI agent tech gets adopted by friendly, funded foundation". It's a fair jab. The foundation's top members include Anthropic, OpenAI, Google, Microsoft and AWS, and on 26 October the Agent Skills GitHub organization still described the format as maintained by Anthropic. The foundation says it has 305 members, and its Silver dues for companies under 100 staff are listed at $5,000 to $10,000 a year, on top of Linux Foundation membership.
My read: for a founder, that doesn't matter much. MCP went the same way in December 2025 and kept growing across every major tool. What you want from a standard is that no single company can change it overnight or lock it to its own product, and moving the name, domain and repository to a foundation makes that much harder. You don't have to pay to use it, and you don't have to pay to contribute. European companies are inside, too: the foundation's launch release named SAP, Adyen, Ericsson and JetBrains among its Gold members, and Mistral AI among its members.

What this means for you

  1. Turn your repeat work into skills. Pick three tasks you explain to an AI agent every week, such as weekly reporting, client onboarding or a content brief, and write each one as a SKILL.md. Our guide to AI tools for solo founders is a good list of where those tasks hide.
  2. Write them once, for every tool. Stick to the spec's basics (name, description, Markdown body) and avoid tool-specific tricks, so the same skill works in Claude Code, Codex, Cursor or Vibe. That's your protection against the next pricing change or model switch, like Replit's move from Claude to OpenAI as its default.
  3. Audit before you install. Use the checklist above for anything from a public registry. Be extra careful with skills that bundle scripts.
  4. Give agents less access. Run new skills without production keys, and keep separate API keys for your own experiments and your customers, as we suggested after Anthropic's per-customer bans.
  5. Think about selling skills. If you're an agency or a consultant, a tested skill library for one niche (say, Shopify product pages in five EU languages) is now a product that works across your clients' tools. Keep the valuable parts private and share only what you're happy to see copied.
  6. Watch the roadmap. Versioned releases and security guidance will decide how skills get distributed and trusted. If you build agent tools, join the discussion on GitHub; it's free.
Skills are also how proactive agents learn your business. Anthropic's Claude Duet and OpenAI's always-on dots agents both point to a future where agents run your admin in the background. For every other agent launch this month, see our AI agents news for October 2026 and Claude Code news for October 2026.
Are you getting customers from AI search yet?

More and more traffic is coming from Perplexity, ChatGPT, Grok and other AI tools.

👉 Rank on ChatGPT

FAQ: Agent Skills and the Agentic AI Foundation

What are Agent Skills?

An open format for teaching AI agents a specific task. A skill is a folder with a SKILL.md file (a short header with a name and description, then Markdown instructions) and, optionally, scripts, reference files and assets. Agents load the full instructions only when a task needs them.

Who owns Agent Skills now?

Since 22 October 2026, the Agentic AI Foundation, part of the Linux Foundation. Anthropic donated the name, logo, GitHub organization and agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies maintain it as individuals, with David Soria Parra and Mahesh Murag as lead maintainers.

Which tools support Agent Skills?

Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code. The agentskills.io showcase also lists Claude Code, Codex, Gemini CLI, Goose, OpenCode, JetBrains Junie and Mistral AI Vibe.

Do I need to join the Agentic AI Foundation to use Agent Skills?

No. Anthropic says you can keep using the standard or contribute to it without joining. Membership is paid, with Silver dues for companies under 100 staff listed at $5,000 to $10,000 a year.

Are Agent Skills safe to install?

Not automatically. Snyk found that 13.4% of 3,984 public skills had at least one critical security issue and confirmed 76 with malicious payloads. Read every SKILL.md and script, test new skills without real keys, and install only from sources you trust.

What is the difference between MCP and Agent Skills?

MCP connects an agent to tools and data, such as your CRM or database. A skill teaches the agent how to do a task, step by step, using those tools. Both now sit in the same foundation, together with AGENTS.md and goose.

About the Author

Violetta Bonenkamp, also known as MeanCEO, is an experienced startup founder with an impressive educational background including an MBA and four other higher education degrees. She has over 20 years of work experience across multiple countries, including 5 years as a solopreneur and serial entrepreneur. Throughout her startup experience she has applied for multiple startup grants at the EU level, in the Netherlands and Malta, and her startups received quite a few of those. She’s been living, studying and working in many countries around the globe and her extensive multicultural experience has influenced her immensely.
2026-10-26 04:26 News and Trends