Anthropic cède Agent Skills à l'Agentic AI Foundation
TL;DR
- All four donations address distinct stack layers: Agent Skills (skill packaging), Open Responses (model API portability), OpenEnv (RL training), OpenSandbox (safe execution).
- AAIF grew from 3 to 10 hosted projects in 12 months, surpassing 300 members with RBC, Vercel, and Ant Group joining this quarter.
- Agent Skills already has 280M installs across 70+ platforms, giving donated governance real installed-base weight from day one.
Anthropic donne Agent Skills. Le nom, le logo, l'organisation GitHub et le domaine agentskills.io partent à l'Agentic AI Foundation (AAIF), hébergée par la Linux Foundation, annonce la société.
La fondation n'est pas nouvelle pour Anthropic : « the Agentic AI Foundation (AAIF) under the Linux Foundation, which we co-founded with Block and OpenAI last year, and to which we donated the Model Context Protocol (MCP) », écrit l'entreprise. Agent Skills prend donc le même chemin que MCP, un an plus tard.
Les chiffres cités justifient la bascule. « More than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated Agent Skills », indique l'annonce, qui ajoute que « 1 million skills have been published and nearly 280 million installed ». Des loaders ont été écrits par la communauté pour Codex, Cursor et OpenCode.
La gouvernance technique est confiée à David Soria Parra et Mahesh Murag comme Lead Maintainers. Anthropic précise que « Engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals » — soit à titre personnel, pas au nom de leurs employeurs.
La posture est explicite : « We're committed to keeping Agent Skills open, community-driven, and vendor-neutral. » Le texte ne détaille pas la répartition des droits de marque ni le processus de décision interne à l'AAIF.
C'est notre énième alerte Anthropic en trois mois, et la deuxième fois que l'entreprise cède un standard clé à la même fondation.
Ce qu'en disent les autres médias
-
Linux Foundation Lire →
Official press release naming all four contributing companies with direct executive quotes from Anthropic, OpenAI, Hugging Face, and Alibaba.
A year ago, AAIF launched with three projects and a bet that agentic AI needed a neutral home to thrive. Today we host 10.
-
The Register Lire →
Only outlet to flag that banks are raising concerns about AI agents accessing customer accounts without authorization, tempering the foundation's growth narrative.
We're committed to keeping Agent Skills open, community-driven, and vendor-neutral.
-
SiliconANGLE Lire →
Clearest technical breakdown of what each donated tool solves: provider portability, skill consistency, RL environment interop, and safe sandboxing.
The AAIF has done important work in giving the Model Context Protocol a neutral home since we donated it less than a year ago.
-
Simon Willison's Weblog Lire →
Developer commentator who publicly predicted this outcome in December 2025; demonstrates Agent Skills in practice using Claude and datasette-mcp.
-
SD Times Lire →
Developer-trade framing: maps each project to the specific agentic stack layer it addresses and tracks the 3-to-10 project growth arc as governance validation.
A year ago, AAIF launched with three projects and a bet that agentic AI needed a neutral home to thrive. Today we host 10.
-
Unite.AI Lire →
Emphasizes cross-company collaboration with adoption metrics (1M+ published skills, 70+ platform support) that put substance behind the vendor-neutrality claim.
Article original publié par anthropic.com
Lire l'article original →Titre original : Anthropic cède Agent Skills à l'Agentic AI Foundation, 70+ plateformes et 280 M d'installations