Anthropic donated Agent Skills, the open folder standard for agent workflows, to the Linux Foundation's Agentic AI Foundation. Google Cloud opened a generally available Anthropic-managed path for Claude. NVIDIA launched Minerva, an AIOps platform that watches GPUs, networks, and jobs inside AI factories. OpenAI made the Decisions API generally available on GPT-6 Luna.
In the news
Anthropic donates Agent Skills to the Agentic AI Foundation
On Oct 22 Anthropic donated Agent Skills, the open standard that packages specialized knowledge and workflows as a folder with a SKILL.md plus scripts and reference files, to the Linux Foundation's Agentic AI Foundation. Block and OpenAI co-founded that foundation. The gift covers the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain. Using or contributing to the spec does not require joining the foundation.1
Anthropic says more than 70 agent platforms have integrated the standard since it published it in December 2025, including ChatGPT, Cursor, GitHub Copilot, and VS Code. The company reports 1 million skills published and nearly 280 million installed. Engineers from Anthropic, Google, OpenAI, and other companies will maintain the spec as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers.2
Proposed next work includes versioned releases, a formal proposal process, and security guidance. The spec lives at agentskills.io, with discussion on GitHub. The same foundation already holds MCP, a different standard Anthropic donated last year.3
Anthropic-managed Claude on Google Cloud
Google Cloud said on Oct 23 that an Anthropic-managed deployment option for Claude is generally available, alongside the existing Google-managed option. Google says both paths deliver the same frontier models on day zero, the same pricing, and the same Google Cloud billing and eligible commitment drawdown. The split is who is the data processor and how fast Claude Platform features land.4
On the Anthropic-managed path, Anthropic operates the stack as data processor on Google Cloud infrastructure under Anthropic's Commercial Terms. Customers get day-zero access to Claude Managed Agents (still in beta), skills, and code execution. They still authenticate with Google Cloud IAM, keep Cloud Audit Logging, and consolidate spend under existing Google Cloud billing.5
On the Google-managed path, Google Cloud remains the data processor under existing customer agreements, with VPC Service Controls, customer-managed encryption keys, and data residency in the U.S., EU, and APAC. Platform features roll out after they are integrated into Google-managed infrastructure. Native integrations include BigQuery and Model Armor. The models already on the platform are Claude Fable 5.1, Sonnet 5.5, Haiku 5.5, and Opus 5.5. This is a deployment choice, not a new model release.6
NVIDIA Minerva for AI factories
NVIDIA launched Minerva on Oct 22, an AIOps platform for AI factories. Minerva joins telemetry from GPUs, networking, storage, schedulers, and workloads, then applies purpose-built models aimed at factory yield: GPU hours, goodput, cluster uptime, and token production. It builds a dynamic knowledge graph of GPUs, SuperNICs, switches, cables, and jobs. NVIDIA says the software analyzes billions of data points per hour, groups event storms into a focused incident with a likely root cause, and ships a native AI assistant plus hooks to external agentic tools.7
NVIDIA names IREN as using Minerva, with CTO Denis Skrinnikoff quoted. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla plan to deploy it. Dell, Lenovo, and Supermicro will offer Minerva as part of their AI factory operations portfolios.8 9
OpenAI Decisions API generally available
OpenAI made the Decisions API generally available on Oct 22, built on GPT-6 Luna, for bounded-answer questions (predicate, choice, or score) from text, images, and native JSON. OpenAI says it is up to 10 times faster than GPT-6 Luna through the Responses API. Since public beta the API has gained native GPT-Live integration, prompt caching, inference residency in the US and EU, images in question instructions, up to 400 questions per request (twice the previous limit), and extra Asia capacity.10 11
Pricing is $0.10 per million input tokens, $0.01 per million for cached input, and no output-token charges. Zero Data Retention is available to eligible customers. Docs and a playground are live.12 13
Named early users include Box, Datadog, Ramp, Sierra, and Speak. Box reports about five times faster route selection on 30 certificate-of-insurance cases. Sierra reports under 280 ms p95, about three times faster and 90 percent lower cost than its previous approach. Speak says GPT-Live time-to-next-action dropped from 1.4 seconds to 0.13 seconds.
