The Pulse
Anthropic Donates Agent Skills Standard to Linux Foundation’s AAIF
Anthropic is donating its Agent Skills standard to the Linux Foundation’s Agentic AI Foundation. The company says more than 70 platforms support the format, which packages task instructions and resources in folders built around a SKILL.md f
AI.info Team · Written with AI · fact-checked ·
“We’re committed to keeping Agent Skills open, community-driven, and vendor-neutral.”
Anthropic, in its October 22 announcement
Anthropic moves Agent Skills into AAIF
Anthropic announced on October 22, 2026, that it is donating Agent Skills to the Agentic AI Foundation (AAIF), a Linux Foundation initiative. The move places the specification under the foundation’s shared stewardship rather than leaving it solely with the company that created it. Anthropic says the transfer gives the standard a neutral home and gives the community a say in its direction.
The donation follows Anthropic’s contribution of the Model Context Protocol (MCP), which connects AI systems with tools and data. The Linux Foundation launched AAIF in December 2025 with MCP, Block’s goose agent framework and OpenAI’s AGENTS.md as founding projects. Agent Skills adds a format for packaging reusable task knowledge alongside those projects’ tools, frameworks and coding instructions.
A folder gives agents task-specific procedures
Agent Skills packages instructions and supporting resources in a folder built around a SKILL.md file. A skill can also include scripts and reference material; an agent loads the relevant content when a task calls for it. Anthropic developed the approach while working on Claude Code, where the company says general-purpose ability did not always supply the procedural knowledge needed for business tasks.
Anthropic first introduced Skills in October 2025 and published the format as an open standard in December. The company says more than 70 agent platforms have integrated it, naming ChatGPT, Cursor, GitHub Copilot and VS Code. The format’s appeal is its portability: developers can write task instructions once and use them across compatible agents, rather than rebuilding the same workflow for each product.
Registry numbers show reach, not quality
Anthropic says one million skills have been published and nearly 280 million installed. Those figures match Vercel’s September report on its skills.sh registry, which reached one million listings and nearly 280 million recorded installs in seven months. They describe activity on that registry, not a count of unique users or proof that the listed skills work well.
Vercel’s report says nearly half of registry skills were installed once, while 375 listings accounted for 62% of installs. That concentration suggests a large catalog can coexist with a small set of heavily used entries. It also makes the distinction between a portable format and a trustworthy skill important for teams choosing what to install.
Shared governance does not vet every skill
A neutral foundation can steward the specification, but that does not by itself review every folder published in the ecosystem. Skills may contain executable code as well as instructions. Anthropic’s earlier technical guide warns that malicious skills can exploit the environment in which they run or direct an agent to expose data, and advises users to inspect material from unfamiliar sources.
The standard’s transfer therefore addresses who helps shape the format, not whether each skill is safe or effective. Anthropic’s own account points to broad adoption; the registry data shows that use is uneven and that install counts alone offer limited evidence about quality. For organizations, the practical test remains the contents of each skill and the permissions its scripts receive.