nebius-physical-ai

Use when ingesting, validating, curating, or querying production sensor data as a versioned dataset-of-record, or wiring the dataset-ingest-curate workflow.

Use when generating adversarial scenarios via RL, ranking mined failures of a policy-under-test, or wiring the adversarial-scenario-hardening workflow.

Use when navigating the Cosmos3 integration in NPA or locating upstream Cosmos3 framework files, defaults, scripts, configs, recipes, and docs.

Use when Cosmos3 setup, fetch, inference, CUDA, uv, Docker, Hugging Face, GitHub, NGC, or checkpoint staging fails in NPA or in an upstream Cosmos framework checkout.

Use when setting up Cosmos3 access through NPA, checking source or Hugging Face reachability, staging the public Cosmos3 framework and checkpoint cache, or deciding which NPA workflow to use before inference.

Use for Claude Code reviews that need robotics, simulation, GPU-routing, sim-to-real, or BDD100K pipeline domain context.

Use when submitting, validating, or debugging NPA SkyPilot workflow YAMLs and workflow runner paths.

Use when working on Isaac Lab RL simulation, deployment, SkyPilot workflows, or customer custom-fork support.

Use when working on MJLab locomotion evaluation, SONIC checkpoint scoring, SkyPilot MJLab YAMLs, or Workbench MJLab CLI behavior.

Use when working on Workbench motion retargeting, SONIC retargeted motion artifacts, SkyPilot retargeting YAMLs, or retargeting CLI behavior.

Use when authoring, reviewing, running, or debugging NPA SkyPilot workflows and runner scripts.

Use when working on SONIC whole-body-control training, export, evaluation, serving, GPU routing, validation, or CUDA alignment.

Use when running or modifying Cosmos3 inference through NPA, especially the public text-to-image SkyPilot workflow, guardrails behavior, prompt/input handling, or upstream Cosmos inference arguments.

Use when planning, reviewing, or explaining Cosmos3 supervised fine-tuning and post-training in NPA, including upstream recipes, dataset/checkpoint preparation, and why NPA does not expose post-training as a fake skill command.

Use when turning an architecture diagram plus a step-by-step write-up into a working npa.workflow/v0.0.1 YAML — parse boxes/arrows/decision-diamonds and numbered steps into states, loops, gates, and catalog toolRefs, then validate/plan until green. Generalizes across sim2real, AV perception, RL, and Cosmos pipelines.

Use when evaluating and onboarding an open-source Physical AI solution into the NPA registry/catalog with documented capabilities, BYOF packaging, smoke tests, and live Nebius validation.

Use when working on NPA reference SkyPilot YAMLs, runner scripts, cookbooks, or customer-adaptable pipeline implementations.

Use when building, enhancing, or testing the NPA chat agent backend — grounded-first routing, cost-aware Token Factory model selection, the embedded-backend mechanism, and cheap-token test tiers.

Use when the NPA agent should describe or critique the current viewer (Rerun, video, image, or data) via the Describe this control or multimodal chat.

Use when operating the NPA agent VM, chat UX, API grounding, bootstrap deployment, or verify-live checks.

Use for Claude Code architectural review of the Nebius Physical AI platform, workbench layer, orchestrator choices, and partner model.

Use when building, tagging, validating, or pushing NPA workbench container images for Nebius registry-backed workflows.

Use when onboarding an OSS repo via BYOF — containerize on Ubuntu or Isaac Lab, push to Nebius registry, and smoke on live Kubernetes.

Use when working on LeRobot workbench training, evaluation, serving, inference, dataset conversion, or robot policy workflows.

Use when authoring, validating, or reviewing NPA workflow specs (apiVersion npa.workflow/v0.0.1) — declarative state machines that invoke workbench tools via SkyPilot.

Use when discovering or loading run artifacts in npa agent without workflow/type/path allowlists.

Use when deploying, tearing down, or reproducing a fresh NPA agent VM from scratch — npa-driven destroy/fresh-setup, profile selection, tiered verify gates, and teardown failure recovery.

Use when verifying that a bootstrapped NPA agent VM can create, validate, plan, provision for, or run npa.workflow YAMLs.

Use for Nebius runtime configuration, provision-if-absent setup, cluster, registry, storage, GPU routing, and credential assumptions that affect NPA runs.

Use to deploy or operate a Nebius soperator (Slurm-on-Kubernetes) cluster from npa — the npa.soperator/v0.0.1 spec, multi-preset worker pools, per-pool Docker/Enroot image cache, quota preflight, and post-deploy fixes.

Use before running or interpreting NPA tests, lint checks, or validation reports.

Use when inventing a new npa.workflow/v0.0.1 pipeline from the tool catalog — creative stage graphs, loops, gates, and golden YAML output.

Use when running, monitoring, or debugging the staged Sim2Real pipeline on a Kubernetes GPU cluster — the runbook, the direct-K8s submit path, preflight health checks, cluster storage secrets, and job monitoring.

Use when working on Cosmos world model serving, inference, serverless training smoke validation, backend selection, or rendering limitations.

Use when navigating, reviewing, or changing the Sim2Real staged pipeline engine — stage map, preamble/inner/outer/finalize entrypoints, K8s sibling jobs, and S3 artifact contracts.

Use when choosing or reviewing GPU targets for NPA workbench tools, training, rendering, inference, or workflow YAML resources.

Use during Claude Code reviews to classify API, IAM, cleanup, exception, concurrency, config, temp-file, and version-pin risks.

Use when drafting, executing, or reviewing Codex super-prompts for this repository.

Use when deploying, launching, loading data into, or reviewing the FiftyOne workbench dataset curation and visualization tool.

Use when working on Genesis simulation, RL teacher training, visual demo generation, or related serverless/EGL behavior.