Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Skills · The Agent Skills standard

Agent Skills has a neutral home — and still no security story

Written October 31, 2026

Anthropic handed the Agent Skills standard to the Agentic AI Foundation on 22 October 2026. Your existing skills keep working, the format is unchanged, and the security guidance is still only proposed — so a skill written by a stranger is no safer than it was on 21 October.

What changed

On 22 October 2026, at AGNTCon + MCPCon North America in San Jose, Anthropic donated the Agent Skills standard to the Agentic AI Foundation (AAIF), a Linux Foundation project it co-founded with Block and OpenAI. The donation covers the Agent Skills name and logo, the GitHub organisation, and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers. The same day, the Linux Foundation announced four new AAIF projects — Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba — bringing the foundation from three launch projects to ten hosted projects in its first year. What changed is who governs the standard and who decides its next version. Anthropic's announcement puts it plainly: the AAIF will provide a neutral home, and the community has a say in where the standard goes next.

The adoption figures in Anthropic's announcement are Anthropic's own, with no counting method stated: more than 70 agent platforms have integrated Agent Skills since the standard was published in December 2025, and 1 million skills have been published with nearly 280 million installed. Those figures describe reach, not review. Nothing in either announcement says any of those skills were checked.

What did not change

The specification did not change. A skill is still a directory holding a SKILL.md file, with YAML frontmatter followed by Markdown instructions. The specification lists six frontmatter fields and requires two of them: name (at most 64 characters) and description (at most 1024 characters, describing what the skill does and when to use it). The optional fields are license, compatibility, metadata, and allowed-tools, the last marked experimental. The specification page carries no version number, and its last update predates the handover, dated 4 August 2026. Existing SKILL.md files stay valid, and the products that load them keep loading them from the same places, because the handover moved governance, not the file format.

None of the six fields records who published a skill, whether anyone reviewed it, or whether it was signed. A skill directory may also hold a scripts/ folder with executable code the agent can run, alongside reference material and assets. The format therefore still lets a skill carry code that runs with the agent's permissions, with nothing in the file itself to say whether that code is safe. That is the gap the handover does not close.

What is only proposed

Anthropic proposed starting with a roadmap covering versioned releases, a formal proposal process, and security guidance for people who write and install skills. Proposed is the operative word: none of the three has shipped. There is no versioned release of the specification to pin a skill to, no published proposal process to watch for changes, and no published security guidance from the foundation to follow. Until one of those appears, the neutral home is a governance fact, not a safety signal. A skill that claims foundation approval, a version number, or a security review is claiming something that does not exist yet.

What to do about it

Limits

Related playbooks

Sources

Back to: Skills index · Also: Vet a skill before you install it · Make sure your Claude Code skill actually loads when it should: test it in each agent before you rely on it · NVIDIA SkillSpector — scan a skill before you install it · Coding playbooks hub