nebius-physical-ai: Instructions file for Codex

Nebius Physical AI

Nebius Physical AI provides containerized workbench tools and SkyPilot workflows for robotics, simulation, perception, and synthetic-data workloads on Nebius infrastructure. Codex should use this file as a lightweight index, scan skills/index.yaml, and load the relevant root skills/ entry before changing behavior or operations.

Key Conventions

• Use npa/.venv/bin/python; never use bare python for repo validation.
• Use RELAXED_DIRTY_TREE_MODE: dirty files outside the run's target paths are not blockers.
• Do not add time, cost, or job-count limits unless the operator explicitly asks for them.

Codex Skills

The source of truth is skills/index.yaml. The tree is organized as:

• skills/workflows/: workflow-level procedures such as sim-to-real, policy training, Cosmos3 inference, and reference SkyPilot workflows.
• skills/atomic/: reusable actions and review conventions such as GPU selection, workflow submission, testing conventions, image build/push, Cosmos3 setup/troubleshooting, and agent visual feedback (Describe this).
• skills/atomic/contributions/SKILL.md: required readability, documentation, and anti-pattern rules when writing or reviewing contributions, including workbench agent changes.
• skills/atomic/agent-visual-feedback/SKILL.md: Describe-this multimodal feedback for Rerun / video / image / data viewers.
• skills/workflows/add-workbench-tool/SKILL.md: add a new workbench tool end to end — implementation, CLI, SDK, toolRef catalog, container, tests, docs, and skill, in the order that keeps every CI gate green.
• skills/atomic/npa-cli-conventions/SKILL.md: write or modify an npa CLI command or SDK function — registration, the --input-path/--output-path contract, output format, error handling and exit codes, the resolve_typer_defaults/json_stdout_contract/intent_boundary decorators, and config/credential access.
• skills/atomic/toolref-argv-contract/SKILL.md: the contract that stops a workflow stage from rendering cleanly and then crashing in the pod — real flag names, literal-value rules, reachability, image routing.
• skills/atomic/pre-pr-validation/SKILL.md: which PR gates apply to a change, and the exact local command for each, cheapest first.
• skills/atomic/audit-container-docs/SKILL.md: required catalog reconciliation when a workbench image or solution is added, removed, retagged, reclassified, published, or otherwise changed; it keeps the public-image table aligned with repository intent and anonymous GHCR state.
• skills/atomic/guardrail-failures/SKILL.md: map a failing guardrail or CI gate to its cause and fix.
• skills/tools/: concrete workbench and platform tools such as LeRobot, FiftyOne, Genesis, Isaac Lab, Cosmos, LanceDB, GR00T, SONIC, MJLab, Retargeting, SkyPilot, Scenario Gen, Dataset-of-record, Fleet, and Nebius infra.
• skills/workflows/first-run-setup/SKILL.md: zero to a first verified result on a fresh machine or new project — an ordered, gated path (install → configure → credential preflight → cheapest-proof workload → validate spec → provision → image pullability → submit) with a stop condition at every step.
• skills/atomic/gpu-first-guidance/SKILL.md: recommending a path, writing user-facing docs, or answering "what should I run first?" — lead with the Nebius GPU path, describe hosted inference by what it does, and stay warm and genuinely helpful when someone asks about cost or has no cluster yet.
• skills/atomic/health-preflight/SKILL.md: there is no npa doctor. Prove service credentials with npa workbench health preflight, request --checks nebius before provisioning, and verify gated-model access with npa workbench health access before spending GPU time.
• skills/atomic/debug-failed-run/SKILL.md: triage a run that failed, hung, or produced no artifacts — status and pod-level reason, stage logs, S3 evidence, image pullability, scheduling, and the resume-vs-cancel decision.
• skills/atomic/teardown-and-cost/SKILL.md: stop spend safely — cancel-before-destroy ordering, cloud versus local state, and the orphan audit for leaked clusters, agent VMs, controllers, buckets, and cross-project fleets.
• skills/tools/token-factory/SKILL.md: zero-GPU hosted inference (captioning, batch generation, Cosmos physical-AI reasoning) — the cheapest tier that produces a real artifact, with no cluster and no provisioning.
• skills/tools/vlm-eval/SKILL.md: score rollouts with a VLM and turn the score into a gate — run vs loop, rubric/threshold benchmark sweeps, backend selection, and judging against a plan an earlier stage wrote.
• skills/tools/golden-eval/SKILL.md: prove a container image actually works — per-container hello-world manifest, dry-run/local/serverless tiers, batch runs, and the offline manifest validation that gates CI.
• skills/tools/alpamayo2-super/SKILL.md: real Alpamayo 2 Super VLA inference, separate OpenMDW model and gated PhysicalAI-AV dataset terms, runtime-fetch packaging, and B200/RTX PRO 6000 validation.
• skills/tools/cosmos3-ray-serve/SKILL.md: persistent Cosmos3-Nano serving through cosmos-framework's native Ray Serve batching path, authenticated readiness, S3 provenance, and independent B200/RTX PRO 6000 validation.
• skills/tools/cosmos3-super-benchmark/SKILL.md: reproduce the fixed Cosmos3-Super vLLM-Omni serving sweep on one eight-GPU B200 or H200 node, or validate the isolated one-H200 TP-1 path, with strict MP4 validity and scope-correct throughput accounting.
• skills/tools/burst/SKILL.md: one gang-scheduled multi-node GPU job with torchrun rendezvous, deliberately not a workflow surface.
• skills/tools/gpu-cluster-provisioning/SKILL.md: managed-image vs GPU-Operator driver strategy (operator mode is unsafe on NVSwitch), the post-apply health gates (fabric, CUDA vectorAdd, stability window), accelerator-name discovery, and triage for nodes whose GPUs do not work.
• skills/tools/detection-training/SKILL.md: Faster R-CNN detectors trained from LanceDB materialized views (BDD100K failure-mode slices).
• skills/tools/artifact-viz-share/SKILL.md: sim demos → LeRobotDataset → .rrd/MP4, and time-boxed presigned Rerun share links.
• skills/workflows/emit-reviewable-rrd/SKILL.md: derive factual Rerun recordings from actual workflow outputs, declare run-scoped .rrd artifacts, and validate their decoded timelines, entities, and provenance before handoff.
• skills/tools/fleet/SKILL.md: deploy a fleet of Nebius Managed Kubernetes (k8s-training) clusters across one or many projects in a tenant from an npa.fleet/v0.0.1 spec — identical and/or custom clusters, create-on-demand projects, and a k8s-training source that can consume the latest upstream recipe.
• skills/tools/scenario-gen/SKILL.md: adversarial scenario generation — an RL adversary that maximizes failures of a policy-under-test, scenario ranking, and the adversarial-scenario-hardening workflow.
• skills/tools/dataset/SKILL.md: dataset-of-record — ingest, validate, curate, and query production sensor data as a versioned, lineage-tracked dataset (FiftyOne curation + LanceDB query index).
• skills/tools/foxglove/SKILL.md: Foxglove embedded viewer — the @foxglove/embed TypeScript SDK in the agent UI, MCAP recordings (convert/inspect/publish), and the npa-foxglove-embed container.
• skills/tools/insights/SKILL.md: lineage graph + common metrics store over workflow-run artifacts — non-invasive ingest-run, query, compare, lineage traversal, and dashboard (CPU-only, append-only S3 JSONL, LanceDB-optional).
• skills/workflows/sim2real-operate/SKILL.md: operate the compositional Sim2Real npa.workflow through the standard SkyPilot runtime — validate/plan/submit, durable S3 resume, preflight health checks, storage secret sync, and job monitoring.
• skills/workflows/agent-fresh-operate/SKILL.md: npa-driven agent teardown, fresh-setup, tiered verify gates, and deploy failure recovery on the operator/dev VM.
• skills/workflows/author-npa-workflow/SKILL.md: author and validate declarative npa.workflow/v0.0.1 specs (validate-spec, plan-spec, toolRef catalog).
• skills/workflows/byof-onboard/SKILL.md: BYOF OSS repo onboarding (Ubuntu/Isaac base, container-verify, agent onboard_solution).
• skills/workflows/contribute-workbench-image/SKILL.md: external fork PR through licensing review, trusted image build, registry-byte validation, and incremental GHCR publication.
• skills/workflows/onboard-world-model/SKILL.md: generic playbook for onboarding and containerizing a world model (learned action-conditioned simulator) as a multi-GPU BYOF registry candidate — containerize, stage a real dataset, encode the train→tokenize→dynamics→dream→visualize loop as capability smokes, validate on real GPUs (Open Dreamer is the reference example).
• skills/workflows/generate-npa-workflow/SKILL.md: design new creative npa.workflow pipelines from the catalog (loops, gates, reference YAML).
• skills/workflows/diagram-to-npa-workflow/SKILL.md: turn an architecture diagram + step write-up into a working npa.workflow/v0.0.1 YAML (boxes/arrows/diamonds/back-edges → states, loops, gates, catalog toolRefs); generalizes across sim2real, AV, RL, and Cosmos pipelines.
• skills/workflows/physical-ai-data-factory/SKILL.md: author/run/view the NVIDIA Physical AI Data Factory blueprint on Nebius + SkyPilot (no OSMO): annotate → Cosmos Transfer augment → Cosmos Evaluator gate → re-label → Cosmos Curator + FiftyOne curate → Rerun visualize. The evaluator and curator are the real Apache-2.0 NVIDIA projects (npa workbench cosmos-evaluator / cosmos-curate); see skills/NOTICE-NVIDIA-COSMOS-OSS.
• skills/workflows/neural-reconstruction/SKILL.md: NuRec/NRE neural reconstruction on Nebius — NCore V4 capture (incl. deriving the rig → world pose edge NRE requires) → 3DGUT Gaussian training → renderable USDZ → rig-offset novel views → reports/sim2real.rrd. RT-core GPU only (L40S / RTX PRO 6000, never H100/H200).
• skills/workflows/content-agents/SKILL.md: public zero-vendor-payload NVIDIA Content Agents packaging with runtime-fetched OVRTX and the RTX-only rigid-object USD workflow — real Material Agent, Physics Agent, OVRTX, and Validation Agent stages → self-contained USDZ plus a narrow Isaac Stage 2 adapter.
• skills/atomic/real-components/SKILL.md: ensure every advertised workbench pipeline stage invokes the real component (Cosmos Transfer, FiftyOne, VLM), not an echo/manifest stub.
• skills/atomic/solution-licensing/SKILL.md: when adding a solution, tool, image, model, or dataset — classify what the artifact actually bakes, decide whether it may be redistributed (public vs restricted), and record it in the packaging contract where the guards enforce it. Verify the claim against the BUILT image with npa/scripts/scan_image_omniverse_payload.py, not by reading the Dockerfile.
• skills/atomic/protect-nebius-infra-details/SKILL.md: sanitize commits, docs, reports, examples, tests, PR/issue text, and live-validation handoffs so concrete live Nebius infrastructure details remain only in access-controlled external evidence.
• skills/workbench/sim2real-engine/SKILL.md: canonical 14-stage Sim2Real graph, stateless adapters, parallel lane joins, ComponentRecords, and durable standard-runtime resume. Retired controller entrypoints remain finite legacy compatibility only.