Agent Skills donated to AAIF: what's real?

Published 23 October 2026

Partly true. Anthropic really did hand Agent Skills to the Agentic AI Foundation on 22 October 2026, the licences stay the same, and ChatGPT, Cursor, GitHub Copilot and VS Code all document support for the format. The "70+ platforms" figure is another matter: the standard's own public list shows 46 products, and the spec still has no way to sign a skill or prove who published it. If your engineers already use coding agents, this makes skills a safer thing to invest in, and it makes vetting them part of the job.

What happened

On 22 October the Linux Foundation announced four new projects for the Agentic AI Foundation (AAIF), starting with "Agent Skills, contributed by Anthropic". AAIF already holds MCP, which Anthropic donated in December 2025.

Anthropic's post lists what moves: "the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain." It also claims "more than 70 agent platforms" have integrated the format since December 2025, plus 1 million published skills and nearly 280 million installs. The @ClaudeDevs thread added: "Your skills keep working, and the licenses don't change."

Is it true?

Anthropic and the Linux Foundation both confirm the transfer. The licence claim holds too. The agentskills/agentskills repo says its code is under Apache 2.0 and its documentation under CC-BY-4.0, and the spec lets each skill declare its own licence.

OpenAI's Codex docs say "Skills build on the open agent skills standard." VS Code's docs cover GitHub Copilot. Cursor calls it "an open standard".

We could not reproduce 70+. On 23 October we counted the entries in the repo's client showcase file: 46 products, last updated on 9 August. Nobody committed to the repo in the 11 weeks before the donation, so the list may lag, but Anthropic publishes no list or method of its own. We found no public dataset behind the skill and install totals either.

Is it useful for engineering teams?

Yes, if your team runs more than one coding agent. A skill is a folder with a SKILL.md file and optional scripts, so a runbook written once loads in Cursor, Copilot, Codex and Claude Code. Neutral ownership makes it harder for one vendor to change the format under you.

Portability still has limits. Datadog Security Labs describes a Claude Code feature where shell commands "run before the rendered skill content is sent to Claude". That feature is not in the open spec, so a skill that relies on it will behave differently in another agent.

What do the critics say?

The Register's headline calls it "vendor-affiliated AI agent tech" adopted by a "friendly, funded foundation". We think that is partly fair. Both lead maintainers named in Anthropic's post have worked at Anthropic: David Soria Parra, per the Linux Foundation release, and Mahesh Murag, an Anthropic product manager per VentureBeat in December 2025. On the other side, the @ClaudeDevs thread says engineers from Anthropic, Earendil (Pi), Google, Microsoft, OpenAI and OpenCode "will govern it as a Technical Steering Committee".

The stronger criticism, about security, predates the donation. Anthropic's own launch post warned that "malicious skills may introduce vulnerabilities". Three scans measured it, on very different samples:

We read the spec's frontmatter fields: name, description, licence, compatibility, metadata and an experimental allowed-tools list. None of them covers signing or publisher identity, and the donation does not change that.

What it means if you are hiring engineers

Do not make experience with one agent a hard requirement. The format is shared, so look for engineers who can write a precise, testable SKILL.md.

Treat skill hygiene the way you treat dependency hygiene. A good interview question is: "You find a useful skill on a public marketplace. What do you do before your agent loads it?" A strong answer includes reading the raw file to the end, pinning a version and keeping production credentials out of reach.

Use it when you hire AI developers or security engineers. More screening questions are in our CTO interview guide.

FAQ

What exactly did Anthropic hand over?

Anthropic transferred the name, the logo, the GitHub organization and the agentskills.io domain to AAIF. The code stays Apache 2.0 and the docs CC-BY-4.0, and you can use or contribute to the standard without joining the foundation.

Who controls the Agent Skills standard now?

The Agentic AI Foundation, part of the Linux Foundation, now holds the project. Per @ClaudeDevs, engineers from Anthropic, Earendil (Pi), Google, Microsoft, OpenAI and OpenCode govern it as a Technical Steering Committee.

Is it safe to install skills from the internet?

Not without a review. All three scans above found unsafe skills, and most malicious ones VirusTotal saw were plain instructions with no code, so read a skill like a script before running it.

How we checked

Wild.Codes wrote this. We help companies hire engineers. We pay Anthropic and OpenAI for their AI models as a customer and have no other relationship with them, the Linux Foundation or AAIF. On 23 October 2026 we read the announcements and vendor docs, counted the raw client list, checked the repo README and re-read the three security reports. We did not install or run any skill.

Sources

• Anthropic, "Donating Agent Skills to the Agentic AI Foundation", https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation, published 22 Oct 2026, read 23 Oct 2026.
• Linux Foundation, "Agentic AI Foundation announces four new projects", https://www.linuxfoundation.org/press/agentic-ai-foundation-announces-four-new-projects-agent-skills-open-responses-openenv-and-opensandbox, published 22 Oct 2026, read 23 Oct 2026.
• @ClaudeDevs on X, thread on the donation, https://x.com/ClaudeDevs/status/2113328528536305718, posted 22 Oct 2026, read 23 Oct 2026.
• Anthropic, "Donating the Model Context Protocol and establishing the Agentic AI Foundation", https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation, published 9 Dec 2025, read 23 Oct 2026.
• Anthropic, "Equipping agents for the real world with Agent Skills", https://www.anthropic.com/research/equipping-agents-for-the-real-world-with-agent-skills, updated 18 Dec 2025, read 23 Oct 2026.
• GitHub, agentskills/agentskills repository, README and specification, https://github.com/agentskills/agentskills, last commit 9 Aug 2026, read 23 Oct 2026.
• GitHub, agentskills/agentskills docs/snippets/clients.jsx, https://raw.githubusercontent.com/agentskills/agentskills/main/docs/snippets/clients.jsx, last changed 9 Aug 2026, counted 23 Oct 2026.
• OpenAI, Codex docs on skills, https://developers.openai.com/codex/skills, undated, read 23 Oct 2026.
• Microsoft, VS Code docs on Agent Skills, https://code.visualstudio.com/docs/agent-customization/agent-skills, undated, read 23 Oct 2026.
• VentureBeat, "Anthropic launches enterprise Agent Skills and opens the standard", https://venturebeat.com/technology/anthropic-launches-enterprise-agent-skills-and-opens-the-standard, published 18 Dec 2025, read 23 Oct 2026.
• Cursor, docs on skills, https://cursor.com/docs/skills, undated, read 23 Oct 2026.
• The Register, "Vendor-affiliated AI agent tech gets adopted by friendly, funded foundation", https://www.theregister.com/ai-and-ml/2026/10/23/vendor-affiliated-ai-agent-tech-gets-adopted-by-friendly-funded-foundation/5304912, published 23 Oct 2026, read 23 Oct 2026.
• VirusTotal, "From automation to infection, part III", https://blog.virustotal.com/2026/10/from-automation-to-infection-part-iii.html, published 8 Oct 2026, read 23 Oct 2026.
• arXiv 2601.10338, large-scale security analysis of agent skills, http://arxiv.org/abs/2601.10338, published 15 Jan 2026, read 23 Oct 2026.
• Snyk, "Agent skills threat landscape", https://research.snyk.io/blog/agent-skills-threat-landscape/, published 5 Feb 2026, read 23 Oct 2026.
• Datadog Security Labs, "Malicious skills: supply chain risks in coding agents with dynamic context", https://securitylabs.datadoghq.com/articles/malicious-skills-supply-chain-risks-in-coding-agents-with-dynamic-context/, published 11 May 2026, read 23 Oct 2026.