A standard born inside Claude Code no longer belongs to one company
Anthropic announced it is donating Agent Skills, the format that teaches AI agents specific tasks and domain knowledge, to the Agentic AI Foundation (AAIF), a Linux Foundation project that already hosts MCP (Model Context Protocol). The donation takes governance of the standard out of a single company's hands and hands future decisions to a group with multiple sponsors, including Anthropic, Block, OpenAI, Google, Microsoft, AWS, Cloudflare and Bloomberg.
From internal experiment to mass adoption
Agent Skills was born inside Claude Code, Anthropic's command line assistant, as an answer to a practical problem: the model handled general knowledge work well, but lacked the procedural know how real business tasks need. A skill is a folder with a SKILL.md file that teaches an agent how to perform a task, along with scripts and reference material it can consult when needed. Published as an open standard in October 2025, the format already counts 1 million published skills and nearly 280 million installs, according to Anthropic. More than 70 agent platforms have adopted the format, including ChatGPT, Cursor, GitHub Copilot and VS Code.
Why give away something that already worked
Donating a successful standard to a neutral foundation follows the same playbook Anthropic used with MCP in December 2025: avoiding an ecosystem that depends on one company's goodwill to keep compatibility and evolve the protocol. AAIF now manages the Agent Skills name, the GitHub organization and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies keep maintaining the standard as individuals, with David Soria Parra and Mahesh Murag as lead maintainers.
What changes in practice for current skill users
For anyone building or using skills today, nothing breaks overnight: the standard keeps working as is, and contributing doesn't require joining the foundation. What changes is the trajectory: future decisions about the format no longer depend only on Anthropic and now have representatives from other companies and the community, which lowers the risk of unilateral changes that break third party integrations. For teams deciding which agent stack to run in production, it's a signal that Agent Skills isn't a single company bet: the foundation already hosts ten projects, including recent contributions from Alibaba with OpenSandbox, Hugging Face with OpenEnv and OpenAI with Open Responses, announced the same day during the AGNTCon plus MCPCon North America conference in San Jose.
The limit of this reading
The donation doesn't mean Anthropic loses real influence: it remains one of the largest code contributors and a co-founder of the foundation, and two of the lead maintainers come directly from the company. Shared governance lowers capture risk, but doesn't erase the weight of whoever built the standard and still holds much of today's user base.
MaxAssistant's take
The main value here is practical, not symbolic: teams betting on Agent Skills gain an extra layer of protection against vendor lock-in, because the standard now answers to a structure with multiple stakeholders, not a single company that could decide alone when to discontinue or rework the format.
Sources
Anthropic: Donating Agent Skills to the Agentic AI Foundation; Linux Foundation: AAIF Announces Four New Projects; Anthropic: Donating MCP and establishing the Agentic AI Foundation.
