Playbooks · Coding tasks
Agent Skills joins the Agentic AI Foundation
Written November 9, 2026
Agent Skills — the SKILL.md standard — moved to the Agentic AI Foundation on Oct 22. What moved, who joined, and the missing security baseline.
What moved on October 22
On October 22, 2026, Anthropic donated Agent Skills to the Agentic AI Foundation (AAIF), a Linux Foundation project it co-founded with Block and OpenAI. The donation covers the Agent Skills name and logo, the GitHub organization that holds the specification, and the agentskills.io domain. The foundation already hosts MCP, which Anthropic donated earlier, plus goose from Block and AGENTS.md from OpenAI.
The same day, the Linux Foundation announced four new AAIF projects — Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba — taking the foundation from three projects to ten.
What did not change
The format is untouched. A skill is still a folder with a SKILL.md file: YAML frontmatter, then Markdown instructions, with optional scripts/, references/, and assets/ folders beside it. The specification was last edited on August 4, 2026, and nothing you have installed stops working.
Adoption is broad: Anthropic counts more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, plus 1 million published skills and nearly 280 million installs. Those counts are Anthropic's figures, not independently checked.
Maintenance moves to engineers from Anthropic, Google, OpenAI, and other companies, acting as individuals rather than for their employers. David Soria Parra and Mahesh Murag are the lead maintainers. You do not have to join the foundation to use the standard or contribute to it.
The gap: security guidance is proposed, not published
Anthropic proposed three pieces of work as the foundation's first order of business: versioned releases, a formal proposal process, and security guidance for people who write and install skills. None of the three is published yet. The spec says almost nothing about security today, while skills can bundle scripts and, in Claude Code, pre-approve tools and register hooks. A shared skill is executable configuration, and the standard gives you no baseline to check it against — ask what a skill runs before you install it, and treat the foundation as a neutral home, not a certifier.
The practical catch: six fields, not twenty
The spec allows six frontmatter fields: name and description (required), plus license, compatibility, metadata, and allowed-tools. Claude Code accepts twenty — the other fourteen are its own, among them hooks, disable-model-invocation, user-invocable, model, and context. Those fields work in Claude Code, but a skill uploaded to claude.ai or the Skills API with any of them fails with a hard error.
Verdict: keep to the six spec fields if a skill should work beyond Claude Code. Put a version number and anything else of your own under metadata, where the spec's own example puts it, and say in the README that a skill using Claude Code-only fields is for Claude Code.
Limits
- The 1 million skills and 280 million installs are Anthropic-reported figures. How installs are counted is not explained in the announcement.
- Versioned releases, the proposal process, and security guidance are proposals. Until they ship, "follows the standard" cannot name which version of the standard.
- Neutral governance is a structure, not a track record. The maintainers act as individuals, and the first contested proposal has not happened yet.
Related playbooks
- Addy Osmani's 25 agent skills for coding agents
- Chrome's modern-web-guidance: skills for your agent
- Agent Skills has a neutral home — and still no security story
- GitHub Agentic Workflows are now GA
Sources
- Donating Agent Skills to the Agentic AI Foundation — Anthropic
- Agent Skills Moves to the Linux Foundation — AgenticSkills
Back to: Coding playbooks hub · Also: Agent Skills has a neutral home — and still no security story · Audit apps that installed agent plugins for you · Vet an Antigravity plugin before the one-click install: what the bundle can do, and what Google does not check