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 train-policygit 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/train-policy)<a href="https://agentmods.dev/skills/nebius/nebius-physical-ai/train-policy"><img src="https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/train-policy/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/train-policy"><img src="https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/train-policy.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00029 | $0.00573 |
| Opus 5 | $0.00015 | $0.00287 |
| Sonnet 5 | $0.00006 | $0.00115 |
| Haiku 4.5 | $0.00003 | $0.00057 |
Grade A, and why
train-policy 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 4d 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 — 58 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Train Policy
When To Use
Use this skill when a task asks how to train, fine-tune, evaluate, or export a robot policy through NPA workbench tools. It is the workflow-level entry point before choosing LeRobot, Isaac Lab, SONIC, or GR00T-specific skills.
Procedure
- Identify the policy family and data contract: LeRobotDataset for LeRobot, Isaac Lab task config for RL, retargeted motion artifacts for SONIC, or model-specific inputs for GR00T.
- Select the GPU target with
skills/atomic/gpu-selection/SKILL.md. - Configure input and output S3 prefixes. Checkpoints and evaluation artifacts must be run-scoped.
- Choose the executable path: direct CLI for a single tool, SDK for application code, or SkyPilot YAML for composed training workflows.
- Verify command help and YAML parsing locally before live GPU submission.
Three-Tier Contract
- CLI:
npa workbench lerobot train,npa workbench isaac-lab train,npa workbench sonic train, and relatedeval,export,serve, orinfercommands. - SDK: use the workbench SDK modules for application code and shared helper functions for request construction.
- YAML:
isaac-lab-rl-train.yaml,sonic-train-standalone.yaml, and sim-to-real workflow YAMLs are executable references. The parallel sweep is now thenpa.workflowspecworkflows/testing/isaac-lab-rl-sweep.yaml(--runtime); its raw template is retired. GR00T N1.7 training uses the realworkflows/testing/groot-1-7-finetune.yamltoolRef path, withgpu_countpropagated into both H100 resources and the upstream trainer world size.
Gotchas
- Do not route RT-core-dependent training or render validation to H100/H200.
- Do not substitute repository-local output directories for S3 artifact paths in public examples.
- Treat tiny smoke trainers as verification substitutes only when the prompt explicitly allows minimal production-input substitution.
- Keep W&B, Hugging Face, NGC, and S3 credentials redacted.
Verify
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.
- 4d ago Changed a2023b52f565
- 9d ago First seen · 58 lines · 29 tokens per session scan A 511a2661f335
train-policy is a skill published in the GitHub repository nebius/nebius-physical-ai (29 stars, last pushed today), licensed Apache-2.0. It adds 29 tokens to every session and 573 once invoked, about $0.0001 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 2026-09-03.
Other skills, from other repositories
spark-environment-setup
Set up a working ML training/inference environment on NVIDIA DGX Spark (GB10, aarch64, CUDA 13). Use when installing PyTorch/Unsloth/TRL/vLLM on DGX Spark, hitting libcudart or wheel-ABI errors on aarch64, or choosing between NGC containers and bare pip installs.
spark-memory-thermal-ops
Manage unified memory and thermals during long-running ML jobs on NVIDIA DGX Spark. Use when planning memory headroom for a training run on GB10, when a job OOMs on unified memory, or when monitoring temperature and power during multi-hour training.
spark-training-gotchas
Preflight and diagnose the ten known failure modes for ML training on NVIDIA DGX Spark. Use when a training run on DGX Spark fails to start, OOMs below the 128GB limit, slows down mid-run, or before any multi-hour training job on GB10.
llama-cpp
Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.
minicpm5-deploy-vllm-ascend
Deploy MiniCPM5-2B with vLLM on Huawei Ascend NPU using vLLM-Ascend. Use when the user mentions vLLM-Ascend, Ascend NPU, Huawei Ascend, CANN, torchnpu, davinci devices, or wants an OpenAI-compatible MiniCPM5 server on Ascend hardware.
minicpm5-deploy-litert
Run MiniCPM5-2B or MiniCPM5-1B on-device with Google's LiteRT-LM runtime — the litert-lm CLI or its OpenAI-compatible server on a desktop, the Kotlin API or the AI Edge Gallery app on Android, the same .litertlm bundle on CPU or GPU. Use when the user says "LiteRT", "LiteRT-LM", "litertlm", ".litertlm", "Android"…