Coding agents
Agent Skills moves to the Linux Foundation: Anthropic hands its standard to the AAIF
Anthropic donates Agent Skills, the SKILL.md format adopted by more than 70 agent platforms, to the Agentic AI Foundation. Open Responses (OpenAI), OpenEnv (Hugging Face) and OpenSandbox (Alibaba) join the foundation too, which now hosts 10 projects.
On October 22, 2026, Anthropic announced it is donating Agent Skills, the open standard that gives AI agents specialized knowledge and workflows, to the Agentic AI Foundation (AAIF), hosted by the Linux Foundation. According to Anthropic, more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have integrated Agent Skills since the standard was published in December 2025 (Anthropic).
The same day, at AGNTCon + MCPCon North America in San Jose, the AAIF took on three more projects: Open Responses (OpenAI), OpenEnv (Hugging Face) and OpenSandbox (Alibaba). That brings the foundation to 10 hosted projects in its first year, up from 3 at its launch in December 2025 (AAIF).
The short version for developers. Nothing changes in your files: a skill is still a folder with a
SKILL.md. What changes is who decides how the format evolves. Anthropic says you can keep using the standard or contribute to it without joining the AAIF (Anthropic).
What is Agent Skills?
A skill is a folder containing a SKILL.md file that teaches an agent a specific task or domain, along with any scripts or reference material it needs. Anthropic says it developed skills while working on Claude Code: the agent handled general knowledge well, but often lacked the procedural knowledge that real tasks need (Anthropic).
The specification published on agentskills.io sets a minimal format (Agent Skills):
- Folder:
SKILL.mdis required;scripts/,references/andassets/folders are optional. - YAML front matter: two required fields,
name(64 characters at most, lowercase letters, numbers and hyphens, matching the folder name) anddescription(1,024 characters at most, saying what the skill does and when to use it). Optional fields:license,compatibility,metadataandallowed-tools, the latter still experimental. - Progressive loading: at startup the agent only loads each skill's name and description (about 100 tokens), then the
SKILL.mdbody when it activates the skill (under 5,000 tokens recommended), and extra files only when needed. The specification advises keepingSKILL.mdunder 500 lines. - Validation: the
skills-refreference library checks the front matter and naming rules.
Minimal example from the specification:
---
name: skill-name
description: A description of what this skill does and when to use it.
---
And the validation command, as documented:
skills-ref validate ./my-skill
What the donation covers
- What is transferred: the Agent Skills name and logo, the GitHub organization and the agentskills.io domain (Anthropic).
- Who maintains it: engineers from Anthropic, Google, OpenAI and other companies, acting as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers. The AAIF describes a multi-company Technical Steering Committee (AAIF).
- Licences: the specification is open source under Apache 2.0 and CC BY 4.0, according to the AAIF.
- Proposed roadmap: versioned releases, a formal proposal process and security guidance for people who write and install skills. Feedback on specification changes goes through the project's Discord and GitHub (Anthropic).
This is the same path as the Model Context Protocol (MCP), which Anthropic had already given to the AAIF, a foundation it co-founded with Block and OpenAI. To plug an MCP server into your tools, see our MCP setup guide.
The adoption figures
Anthropic and the AAIF give the same orders of magnitude, but not quite in the same words:
- More than 70 agent platforms have integrated the standard (Anthropic).
- 1 million skills published and nearly 280 million installs, according to Anthropic. The AAIF specifies that this means a million skills listed on the skills.sh registry in seven months, and points to a Vercel study for the install count (AAIF).
These figures come from the standard's backers; we could not verify them independently.
The three other projects joining the AAIF
According to the AAIF's announcement, each project covers one layer of the agentic stack (AAIF):
| Project | Contributed by | Role |
|---|---|---|
| Agent Skills | Anthropic | Open format for giving agents new capabilities and domain expertise |
| Open Responses | OpenAI | Multi-provider LLM interface specification, based on the OpenAI Responses API |
| OpenEnv | Hugging Face | Framework and hub for building interoperable reinforcement learning environments |
| OpenSandbox | Alibaba | Sandbox platform for AI applications (multi-language SDKs, Docker and Kubernetes runtimes) |
For developers, Open Responses is the most concrete after Agent Skills: it defines a shared schema, consistent streaming events and extensible tooling, so you can describe requests and outputs once and run them across model providers with minimal translation (AAIF). SiliconANGLE notes that, being modeled on OpenAI's API, it should make it easier to move code from one provider to another (SiliconANGLE). OpenSandbox targets use cases including coding agents, agent evaluation and AI code execution.
The four projects join A2A, Agent Router, agentgateway, AGENTS.md, goose and MCP at the foundation.
What it means for your tools
- Portable skills: the same folder can work across several compatible agents. The AAIF cites Block, whose engineers share an internal marketplace of hundreds of skills (incident response, code review, experiment setup) used with Goose, Codex, Claude or Amp, without standardizing on a single agent (AAIF).
- Neutral governance: the format no longer evolves at Anthropic's sole discretion, which matters to the vendors that integrated it, such as Cursor, GitHub Copilot or Codex. On Google's side, Antigravity CLI also keeps Agent Skills support (see our Gemini CLI migration guide).
- Security still to come: security guidance for writing and installing skills is only on the proposed roadmap for now. A skill can ship scripts: review the ones you install.
Simon Willison points out that he predicted this move back in December 2025, when Anthropic first declared skills an "open standard" (Simon Willison). To pick the agent that will run your skills, see our comparison Claude Code, Codex or GitHub Copilot.
Sources
- Anthropic: Donating Agent Skills to the Agentic AI Foundation (Oct 22, 2026), verified 2026-10-23
- AAIF: Agentic AI Foundation Announces Four New Projects (Oct 22, 2026), verified 2026-10-23
- AAIF: Agent Skills, Sharing Team Expertise Across AI Agents (Oct 22, 2026), verified 2026-10-23
- Agent Skills: specification, verified 2026-10-23
- GitHub: agentskills/agentskills, verified 2026-10-23
- SiliconANGLE: Agentic AI Foundation adds four more open-source AI tools (Oct 22, 2026), verified 2026-10-23
- Simon Willison: Donating Agent Skills to the Agentic AI Foundation (Oct 22, 2026), verified 2026-10-23
Key takeaway: Agent Skills, the
SKILL.mdformat adopted by more than 70 agent platforms according to Anthropic, moves under the AAIF's neutral governance (Linux Foundation), as MCP did before it; Open Responses, OpenEnv and OpenSandbox arrive at the same time, and the foundation now hosts 10 projects.
Spotted an error in this article? Report it · Our corrections policy