Anthropic donates Agent Skills to the AAIF and launches Claude on Google Cloud

Agent Skills moves to neutral ground

Anthropic announced on October 22 that it is donating Agent Skills to the Agentic AI Foundation (AAIF), part of the Linux Foundation. Anthropic co-founded the AAIF with Block and OpenAI last year and had already donated the Model Context Protocol to it. A skill, as Anthropic describes it, is a folder containing a SKILL.md file that teaches an agent a specific task or domain, with any scripts or reference material it needs. Anthropic says it developed skills while working on Claude Code, which handled general knowledge work well but often lacked the procedural knowledge that real business tasks need.

The donation covers the Agent Skills name and logo, the GitHub organization and the agentskills.io domain. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have integrated the standard since its publication in December 2025, and that 1 million skills have been published and nearly 280 million installed. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers. Anthropic proposes starting with a roadmap of versioned releases, a formal proposal process and security guidance for people who write and install skills. It adds that the standard can be used or contributed to without joining the AAIF.

Sources: Donating Agent Skills to the Agentic AI Foundation — Anthropic

Research publications and a call to life scientists

Anthropic also published an alignment assessment of what it calls the UK AISI cybertesting incident. The material provided for this edition carries only the title of that piece. It does not say what happened, what caused it or what the assessment concludes, so this edition does not characterize the incident. The original should be read directly.

Another Anthropic item, titled "LLMs are making formal software verification feasible at a much larger scale", was published the same day. It is likewise available here only by its title. Its headline thesis is that language models are widening the scale at which software can be formally verified.

On the science side, Anthropic opened a call for guest investigators in the life sciences on October 22. It invites principal investigators and their groups at academic institutions, nonprofits or government labs to propose work in causal human biology, structural and computational molecular biology, medicinal chemistry, drug discovery and clinical science. Selected teams would work with Anthropic researchers for 8 to 12 weeks, with Anthropic running its models and harnesses at no cost to them. The outputs are to be published, and the accompanying data and code open-sourced. Guest investigators keep the rights to their data and to independent follow-on research. Anthropic says it is not asking for their data and wants only publicly available information in applications. Applications stay open for about 14 days, and Anthropic aims to notify selected participants about 14 days after they close.

Sources: An alignment assessment of the UK AISI cybertesting incident — Anthropic · LLMs are making formal software verification feasible at a much larger scale — Anthropic · Open call for guest investigators in the life sciences — Anthropic

A new route to Claude through Google Cloud, and a week of Claude Code releases

Anthropic's API release notes for October 22 record the launch of Claude Models on Google Cloud (Anthropic managed). It is the Claude API operated by Anthropic and reached through a customer's Google Cloud project, with Google Cloud IAM controlling access and Google Cloud Marketplace handling billing. The same notes say fast mode is available for Claude Opus 5.5 on this offering.

Claude Code shipped three versions in quick succession. Version 2.1.307 adds a /claude-api api-version command to move Claude API code to API version 2026-10-21. It also adds a claude_code.safety_event OpenTelemetry event for each tool call that a permission rule, auto mode, a safety check, a hook or the sandbox stopped, prompted or sent back. Version 2.1.309 is largely a sandbox-hardening release. It fixes sandboxed Bash commands on macOS and Linux running the shell's startup files outside the sandbox before each command, and macOS denyWrite glob patterns not applying at a path's real location when reached through a symlink. It also fixes sandboxed commands and auto-accepted edits being able to modify the MCP config file passed with --mcp-config.

Version 2.1.310 adds the Monitor tool and plugin monitors on Bedrock, Vertex and Foundry, and when telemetry or nonessential traffic is disabled. Long foreground sleeps are now blocked in those cases too. It also makes Claude Code retry without the effort-2025-11-24 beta value or mid-conversation system messages when some gateways and compatible servers reject them. Finally, it fixes background cleanup deleting some recent files, such as pasted images, when cleanupPeriodDays is set to a very large value.

Sources: Anthropic API release notes, October 22, 2026 — Anthropic (release notes) · Claude Code 2.1.307 — Claude Code (GitHub) · Claude Code 2.1.309 — Claude Code (GitHub) · Claude Code 2.1.310 — Claude Code (GitHub)

Two brief error episodes, both resolved

The official status page recorded two incidents in the early hours of October 24 UTC, both now resolved. Elevated errors affected Claude Opus 5.5 from 17:23 to 17:34 PT on October 23 (00:23 to 00:34 UTC on October 24), about eleven minutes. Elevated errors for Claude Fable 5 and Claude Fable 5.1 followed from 22:00 to 22:35 PT on October 23 (05:00 to 05:35 UTC on October 24), about thirty-five minutes.

The status entries provided give the windows of impact and their resolution. They state no cause and no scale, and this edition does not add one.

Sources: Elevated errors for Claude Opus 5.5 — Claude Status · Elevated errors for Claude Fable 5 and Claude Fable 5.1 — Claude Status