In this article
  1. What happened
  2. What a skill is
  3. Why it matters
  4. What to do
  5. What we don’t know yet
  6. Updates

TL;DR: On October 22, 2026, Anthropic donated Agent Skills, the open format for packaging instructions and scripts that AI agents load on demand, to the Agentic AI Foundation (AAIF) under the Linux Foundation. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, already support it. Nothing breaks and nothing needs updating today. The practical point for small teams is the one on the new roadmap: a skill can include code your agent runs, so treat installing one like installing software.

What happened

Anthropic said it is handing AAIF the Agent Skills name and logo, the GitHub organization and the agentskills.io domain. AAIF is the Linux Foundation group Anthropic co-founded with Block and OpenAI, and it already hosts the Model Context Protocol (MCP). Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as lead maintainers. Anthropic says you can keep using or contributing to the standard without joining AAIF.

The same day, at AGNTCon + MCPCon North America in San Jose, AAIF announced four new projects: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba. That takes the foundation from three projects at its launch a year ago to 10.

Anthropic’s own figures: since the format launched in October 2025, “1 million skills have been published and nearly 280 million installed.”

What a skill is

According to the specification at agentskills.io, a skill is a folder with a required SKILL.md file: a short metadata block (a name and a description of when to use it) followed by plain instructions. The folder can also hold a scripts/ directory, which the spec labels “executable code,” plus reference docs and assets. Agents load only the name and description at startup and pull in the rest when a task needs it.

Why it matters

  1. Your skills stay portable. A skill you write for one tool should keep working in others, and the format is no longer controlled by one company.
  2. Changes get a process. Anthropic’s proposed roadmap covers versioned releases and a formal proposal process, so changes should arrive as numbered versions rather than surprises.
  3. Security guidance is proposed, not published. The same roadmap lists “security guidance for people who write and install skills.” Neither Anthropic nor AAIF has said when it will land. Until then, checking what you install is on you.

What to do

These steps are our own advice, not part of the spec.

  • Inventory what’s installed. List the skill folders in each agent tool your team uses, and delete anything nobody recognizes or uses.
  • Read the whole folder before installing a new skill. Open every file in scripts/, not just SKILL.md. If a script downloads more code, sends data to an outside address, or reads keys or .env files, don’t install it.
  • Prefer skills you can trace. A known company’s or colleague’s repository with a history beats a copy-paste from a stranger’s post. Pin to a specific commit or release rather than “latest,” so it can’t change under you.
  • Check allowed-tools. The spec has an experimental allowed-tools field listing tools a skill wants pre-approved. Support varies by tool, but if a skill asks for broad shell access, ask why.
  • Keep secrets out of reach. Give agents the fewest keys they need, and keep production credentials in your cloud provider’s secret store or your CI’s encrypted secrets, not in files the agent can read. The LiteLLM credential-stealer incident shows what a bad package does with keys it can see.
  • Write your own for repeat work. For your own processes, such as a refund policy or a weekly report format, a home-made skill is a Markdown file you control. The spec’s skills-ref validate tool checks the format.
  • Sandbox agents that touch code or data that matters. Running agent scripts in a container or a throwaway machine limits the damage from a bad skill. For more on keeping agents from acting on bad inputs, see our guide to stopping AI agents from trusting the wrong tool outputs.

What we don’t know yet

Neither Anthropic nor AAIF has given a date for the first versioned release or the security guidance, or said whether published skills will be vetted.

Updates

  • Oct 23, 2026, 10:40 a.m. Manila: First published. We’ll add a dated note here when the first versioned release or the security guidance appears.

Sources & method

Sources

Method. Last checked October 23, 2026, at 10:40 a.m. Manila time (02:40 UTC). The donation details, maintainers, roadmap and usage figures come from Anthropic’s post, and the usage figures are Anthropic’s own, which we haven’t verified. The project list and count come from AAIF’s announcement. The description of a skill comes from the specification. The steps under “What to do” are our own judgment. No affiliate links on this page. How we work.

Spotted an error? See our corrections policy and tell us.