Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add nebius/nebius-physical-ai --skill ray-workbenchgit clone --depth 1 https://github.com/nebius/nebius-physical-aiWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/nebius/nebius-physical-ai/ray-workbench)<a href="https://agentmods.dev/skills/nebius/nebius-physical-ai/ray-workbench"><img src="https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/ray-workbench/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/nebius/nebius-physical-ai/ray-workbench"><img src="https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/ray-workbench.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00038 | $0.00737 |
| Opus 5 | $0.00019 | $0.00368 |
| Sonnet 5 | $0.00008 | $0.00147 |
| Haiku 4.5 | $0.00004 | $0.00074 |
Grade A, and why
ray-workbench scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 63 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Ray in Workbench
Start with the consolidated guide. Use this skill to route a Ray request to behavior that current code actually supports.
Route the request
- Native Jobs or Ray Core GPU development: use
docs/testing/fast-source-iteration.mdand the guarded CLIP application undernpa/workflows/workbench/ray-clip-development/. - Ray Train: load
skills/tools/ray-train-synthetic/SKILL.md. The shipped path is a synthetic Ray Train V2 application submitted through native Jobs, not an NPA Train service or workflow. - Ray Serve: load
skills/tools/cosmos3-ray-serve/SKILL.md. Only Cosmos3-Nano has a first-class native Ray Serve path. - KubeRay: load
skills/tools/fleet/SKILL.md. Fleet supports only the reviewed, fixed CPU RayCluster policy documented indocs/fleet-kuberay.md. - Ray Data or Ray Tune: state that no first-class Workbench recipe, CLI, service, workflow, or artifact contract exists. A trusted user may submit their own Ray application through a supported native Jobs environment; do not imply that this makes Data or Tune a supported Workbench capability.
There is no npa ray or NPA Jobs controller. Native ray job owns application
submission, list, status, logs, and stop. SkyPilot owns the CLIP and Train hosts
and long-running service task. Fleet owns KubeRay infrastructure. The Cosmos
client owns verified S3 batch publication. Keep these lifecycle identities
separate.
Operating rules
- Before provisioning, use
npa workbench health preflight --checks nebius --json; add the selected capability's storage and model-access checks. - For SkyPilot references, run
npa skypilot bootstrapand resolveNPA_SKYPILOT_BINfromnpa skypilot status --bin-path. Never use ambient management Ray discovery orskyfromPATH. - Keep Jobs and GCS on private networking behind an authenticated loopback tunnel. Ray accepts trusted code and a namespace is not a tenant boundary.
- Use unique native submission IDs.
ray job stopmust reach terminal status before hosting teardown. - Preserve the path-specific artifact boundary before cleanup: checksummed CLIP output, Train checkpoints/exports, verified Cosmos S3 publications, or explicitly copied KubeRay application output.
- Cancel exact application jobs before the exact SkyPilot service task or owned Fleet target. Preserve shared clusters, APIs, controllers, projects, and storage unless their ownership is separately established.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 63 lines · 38 tokens per session scan A d713521ec284
ray-workbench is a skill published in the GitHub repository nebius/nebius-physical-ai (29 stars, last pushed today), licensed Apache-2.0. It adds 38 tokens to every session and 737 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-10.
Other skills, from other repositories
tensorrt-llm
High-throughput LLM inference on NVIDIA GPUs.
google-cloud-solution-guided-gke-ai-migration
Guides the migration of existing AI workloads (Cloud Run, Gemini API, Gemini Enterprise Agent Platform) to self-hosted GKE inference using gcloud and kubectl. Use when the user has an existing AI inference workload (on Cloud Run, the Gemini API, Gemini Enterprise Agent Platform, or a custom VM) and wants to move it to…
agent-platform-tuning
Agent Platform Model Tuning. Use when you need to fine-tune open models or Gemini models using Agent Platform infrastructure. Don't use for model training outside Agent Platform, model deployment to endpoints (use agent-platform-deploy), or managing serving endpoints (use agent-platform-endpoint-management).
modal
Modal is a serverless cloud platform for running Python on demand, including on-demand GPUs. Use when deploying or serving AI/ML models, running GPU-accelerated workloads (training, fine-tuning, inference), serving web endpoints, scheduling batch jobs, or scaling Python code to cloud containers with the Modal SDK.
agent-platform-endpoint-management
Manages Agent Platform serving endpoints. Use when you need to create, list, describe, update, or delete serving endpoints for model deployment on Agent Platform. Also use when troubleshooting endpoint permission, quota, or resource busy errors. Don't use for deploying models to endpoints or for running model…
gke-inference
Deploys and optimizes AI/ML inference workloads on GKE, using GPUs, TPUs, and model servers. Use when deploying GKE inference servers, configuring GKE GPU resources for inference, or deploying LLMs on GKE. Don't use for generic batch jobs or HPC task queues (use gke-batch-hpc instead).