# Agent Skills, OlmoEarth v1.3, ForTy, DGX Spark, Granite ensemble > Agent Skills moves to the Linux Foundation, Ai2 and DeepMind release planet-scale maps, DGX Spark reaches eight nodes, IBM ships a forecasting ensemble. Published 22 Oct 2026. Coverage: 22 Oct 2026 to 22 Oct 2026. By Ali Asif. ## Items 1. **Donating Agent Skills to the Agentic AI Foundation** (Anthropic, Agents and coding): The Agent Skills standard moves to the AAIF under the Linux Foundation, with more than 70 platforms integrated, 1 million skills published and nearly 280 million installs. [Primary source](https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation) 2. **OlmoEarth v1.3** (Ai2, Open weights): An open satellite embedding model with precomputed global 10 m embeddings, leading AlphaEarth Foundations and Tessera v2 on 20 of 27 evaluations at 128 dimensions. [Primary source](https://allenai.org/blog/olmoearth-v1-3) 3. **Beyond tree cover: mapping the world's forest types** (Google DeepMind, Science): ForTy 2020 classifies every 10 m patch into six forest and land-use types, validated at 90% forest-type accuracy on more than 9,000 sites, plus an early 2017 to 2025 time series. [Primary source](https://deepmind.google/blog/beyond-tree-cover-mapping-the-worlds-forest-types) 4. **Running Frontier-Class AI Models Across 8 NVIDIA DGX Spark Nodes** (NVIDIA, Agents and coding): Official eight-node DGX Spark clusters give 1 TB of unified memory, enough to serve the full GLM 5.3 locally, with decode on a 27B model scaling from about 12 to 52 tokens per second. [Primary source](https://developer.nvidia.com/blog/running-frontier-class-ai-models-across-8-nvidia-dgx-spark-nodes) 5. **Ensemble forecasting with IBM Granite time-series models** (IBM Research, Open weights): An open recipe that pools four Granite forecasters with no retraining and no fitted weights, scoring a MASE of 0.683 on 97 GIFT-Eval configurations against 0.685 for the best single member. [Primary source](https://huggingface.co/blog/ibm-research/granite-timeseries-ensemble-r1) Today's log has one standards story and a cluster of releases about reading the planet and the desk. I read the primary pages and ran none of the code, so the figures are what each publisher reports. ## Agent Skills goes to the AAIF Anthropic is donating the Agent Skills standard to the Agentic AI Foundation, under the Linux Foundation. A skill is a folder with a SKILL.md file that teaches an agent a task, plus any scripts or references it needs. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, have integrated the standard since its December 2025 publication, with 1 million skills published and nearly 280 million installed. The donation covers the name, logo, GitHub organization and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and others will maintain it as individuals, and the proposed roadmap lists versioned releases, a formal proposal process and security guidance for people who write and install skills. That last item matters to me most, since installing a skill means running someone else's instructions and scripts. ## Two planet-scale maps Ai2 released OlmoEarth v1.3, a satellite embedding model that reads a year of radar and multispectral data and produces one 128-dimensional vector per 10 meter pixel on Earth. The embeddings are precomputed for the whole planet, and a browser tool lets you find similar places or train a small classifier on a few labeled points. Ai2 reports the 128-dimension embeddings score highest on 20 of 27 evaluations against AlphaEarth Foundations and Tessera v2. Weights, pretraining data and code are open. Google DeepMind, with WRI and IIASA, released Forest Typology (ForTy) 2020. It sorts every 10 meter patch into six types, from primary forest to plantation and tree crops, and builds on AlphaEarth Foundations embeddings. Validated against more than 9,000 sites, it picked the right forest type 90% of the time. The early time series from 2017 to 2025 shows what replaced a forest, which tree cover maps miss. Both are free to use. ## DGX Spark reaches eight nodes NVIDIA announced official eight-node DGX Spark clusters with 1 TB of unified memory. The full GLM 5.3 checkpoint, about 470 GB in NVFP4, runs across the cluster at 31 decode tokens per second for one user. On a dense 27B model, decode scales from about 12 tokens per second on one node to 52 on eight with a switch and the RoCEnante backend. NVIDIA notes the backend suits small clusters and low concurrency, and that NCCL remains the better default at scale. ## Granite Timeseries Ensemble IBM Research published an ensemble recipe that pools four open Granite forecasters, returning median and quantile forecasts with no retraining. On 97 GIFT-Eval configurations the Probabilistic Ensemble reports a MASE of 0.683 against 0.685 for the best single member. The gap is small, and the value is mostly not having to pick a model. ## What I'd watch next Whether the AAIF roadmap ships security guidance for skills soon, how fast the ForTy time series moves past its early release, and whether the eight-node Spark support lands in NVIDIA Sync by the end of the year as promised. ## My take Agent Skills moving to a neutral foundation is the item that changes how I build, because a skill I write now has a home that no single vendor controls. The two Earth observation releases are the ones I would open in a browser first. I read each primary page and ran none of it, so every number here is the publisher's own. ## Sources - https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation - https://allenai.org/blog/olmoearth-v1-3 - https://deepmind.google/blog/beyond-tree-cover-mapping-the-worlds-forest-types - https://developer.nvidia.com/blog/running-frontier-class-ai-models-across-8-nvidia-dgx-spark-nodes - https://huggingface.co/blog/ibm-research/granite-timeseries-ensemble-r1 Canonical: https://heyaliasif.com/ai/2026-10-22-agent-skills-olmoearth-forty-dgx-spark-granite