🔍 Intelligence Synthesis JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes

AI Tech Flash ── 2026-10-23

Anthropic Donates AI Skills Standard to Foundation

Agent Skills moves to a Linux Foundation-hosted foundation

What changes

The standard for AI "skills" is leaving the control of a single company. Skills are reusable instruction sets that AI agents load to do a task. The new home is the Agentic AI Foundation, which sits under the Linux Foundation. According to MIXED, developers can use the standard and contribute without joining the foundation.

A shared format makes it easier to distribute the same skill across several products. MIXED also notes that individual skills can still depend on particular tools and permissions.

Why it matters

According to MIXED, more than 70 agent platforms support the format. They include ChatGPT, Cursor, GitHub Copilot and VS Code. The standard is no longer one company's own specification.

Security is also a concern. The security firm JFrog said in a press release that its 2026 report identified 969 malicious agent skills carrying high-impact payloads. As skills spread, checking them for safety becomes more important.

What it is

Agent Skills is a standard for giving AI agents reusable instructions and workflows. A skill combines instructions in a document called SKILL.md with any scripts and reference material the task needs.

The Agentic AI Foundation sits under the Linux Foundation. MIXED reports that it already houses the Model Context Protocol. That is another standard Anthropic helped establish.

Who and where

Anthropic developed the approach while working on Claude Code. It launched skills in October 2025 and published the open standard that December. MIXED reports that the donation was announced on October 22.

The donation covers the Agent Skills name and logo, its GitHub organisation and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies will maintain the project as individuals. David Soria Parra and Mahesh Murag will be lead maintainers.

Why it was built

Anthropic published the open standard after developers began adapting the idea for other products. According to MIXED, a shared format lets a team package a procedure once and use it across different agents.

The problem it addresses is repeated explanation. Without skills, a developer must describe a team's procedure in chat again and again. A packaged skill lets the agent load that procedure when needed.

How it works

A developer writes instructions in SKILL.md. Any scripts and reference material needed for the task go into the same skill. The agent loads the skill and follows the procedure it describes.

JFrog states in its press release that skills ship plain-English instructions that an agent reads and acts on. Older security tools were built to catch vulnerable code. JFrog announced products that read the intent of skills and enforce policy when an agent uses them.

Key points

  1. Anthropic is donating the Agent Skills name, logo, GitHub organisation and domain to the Agentic AI Foundation.
  2. Anthropic says over 70 platforms support it, with one million published skills and nearly 280 million installs.
  3. Proposed next steps are versioned releases, a change process and security guidance. These are proposals only.

Conclusion

Agent Skills moves from Anthropic's sole control to a foundation. According to MIXED, engineers from several companies will maintain it as individuals. The adoption figures are Anthropic's own reports, not independently verified user counts.

Quotes and sources

“A skill combines instructions in a SKILL.md document with any scripts and reference material needed for a particular task.”
── MIXED, 2026-10-23
“The foundation sits under the Linux Foundation and already houses the Model Context Protocol, another standard Anthropic helped establish.”
── MIXED, 2026-10-23
“Those figures describe adoption reported by Anthropic, rather than a count of independently verified, distinct users.”
── MIXED, 2026-10-23
“Agent skills, plugins, and APM packages ship something different: plain-English instructions that an agent reads and acts on.”
── source, 2026-10-21

Source articles

Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.