NVIDIA
- 1.2k skills
- 0 followers
- 23k repo stars
- 561 verified
- 10 hours ago last updated
- ▌ Kubectl Nv · nvidia bundleDiagnose NVIDIA GPU nodes in a Kubernetes cluster with the kubectl-nv plugin. Use it when a GPU workload will not schedule, a node advertises no GPUs, the GPU Operator operands are unhealthy, a driver or DCGM pod is failing, or someone asks which node a job can run on. It reads a live cluster or an offline must-gather bundle, reports findings as stable NV-<AREA>-<NNN> codes with the evidence behind each one, and never changes a Kubernetes object.
- ▌ AI Factory Operations Agent Headless · nvidiaUse AI Factory Operations Agent without opening the browser UI. Invoke it through the headless HTTP API, packaged CLI, or stdio MCP server, and deploy it first when no service is available.
- ▌ Bcm · nvidiaInspect Base Command Manager (BCM) inventory and health evidence, and make explicitly requested changes when Base Command Manager edit mode is enabled.
- ▌ Iraop · nvidiaResearch NVIDIA DGX, SuperPOD, GB200, NVLink, BlueField, Base Command Manager (BCM), and on-prem infrastructure documentation through the IRA/Sequoia retrieval agent. Use when the user asks to research docs, deployment policy, operational guidance, or newly uploaded knowledge-base documents.
- ▌ Slurm · nvidiaInspect mounted Slurm accounting exports and job logs for failed workload root cause analysis.
- ▌ Grafana · nvidiaCreate and open Grafana dashboards for metrics, Loki logs, and mixed panels. Use when the user asks for a dashboard, visualization, chart, panel, or graph of GPU, Base Command Manager (BCM), node, network, InfiniBand, Prometheus metrics, or Loki logs.
- ▌ Observability · nvidiaAnalyze cluster observability data and create Grafana dashboards. Use when the user asks about metrics, logs, log labels, cluster health history, dashboards, charts, panels, or visualizations.
- ▌ Gpu Networking · nvidia bundleAll documentation paths below are relative to the agent workspace. These files
- ▌ Hardware Agent · nvidiaTriage hardware faults on NVIDIA GPU compute nodes through the Hardware Agent. Starting a triage is read-only log collection and hardware_analyze_dut is available in View mode; never require Edit or Auto. Before asking to start any fresh NVDebug collection, you MUST say: 'A fresh NVDebug collection can take 5-30 minutes. Do you want me to start it?' Stop and wait for the next user response. An ordinary affirmative such as yes, let's go, do it, or proceed is sufficient; start immediately without demanding formal confirmation or asking twice. Use recent Hardware Agent triages for quick answers when possible. The backend owns DUT lookup and credentials; do not prompt for BMC credentials. Not for ML training, CI, or unrelated application errors.
- ▌ Slurm Job Sizing · nvidiaSize Slurm allocations, warmup and traffic for short debugging jobs or full srt-slurm benchmarks.
- ▌ Using Holodeck · nvidiaUse when the user wants to provision, manage, or destroy GPU-enabled test environments via the holodeck CLI. Covers env.yaml config, create/dryrun/list/status/ssh/scp/delete/cleanup/get/os workflows, and common pitfalls.
- ▌ Flashdreams Run Model · nvidia bundleSelect, install, launch, and troubleshoot an existing FlashDreams video or world-model demo using the V2 CLI. Use for model/demo presets, browser or file output, runtime versus application arguments, and first-run failures. Model porting, new demo development, and offline executable packaging are separate workflows.
- ▌ Flashdreams Create Demo · nvidia bundleBuild or modify a model-independent FlashDreams V2 demo application with sessions, model/UI loops, interactive controls, and stub-based tests. Use for a new interaction or application lifecycle change. For a new model binding to an existing demo, use the model integration workflow instead.
- ▌ Flashdreams Integrate Model · nvidia bundleAdd or migrate a video or world model to FlashDreams V2, including recipe reuse, checkpoint conversion, pipeline configuration, and an adapter for an existing demo. Use when porting an external model or wiring its application entry point; creating a new model-independent demo and running an already supported preset are separate tasks.
- ▌ Flashdreams Package Offline · nvidia bundlePackage an installed FlashDreams V2 application as an offline executable folder and diagnose preload or bundled-cache failures. Use for native-window or WebRTC distribution without Python, uv, or network access on the target. Ordinary demo launching, generic Python packaging, and website builds are outside this workflow.
- ▌ Security Code Review · nvidiaReview security-sensitive FlashDreams changes or an explicitly requested security review, including untrusted inputs, model loading, serving endpoints, credentials and CI permissions. Produce findings and evidence for human review.
- ▌ Physicsnemo Cfd Create Custom Metric · nvidia bundleCreate a custom metric for the PhysicsNeMo CFD benchmarking workflow. Use when the user wants to add a new evaluation metric, implement a custom error measure, compute force coefficients, or extend the benchmark with domain-specific quantities.
- ▌ Physicsnemo Cfd Create Model Wrapper · nvidia bundleCreate a new model wrapper for the PhysicsNeMo CFD benchmarking workflow. Use when the user wants to add a new CFD model, write a CFDModel wrapper, integrate a new neural network architecture, or run a custom model through the benchmarking pipeline.
- ▌ Physicsnemo Cfd Create Dataset Adapter · nvidia bundleCreate a new dataset adapter for the PhysicsNeMo CFD benchmarking workflow. Use when the user wants to add a new CFD dataset, write a DatasetAdapter, integrate a new mesh format, or benchmark models on custom data.
- ▌ Tilegym Cutile Optimizing · nvidia bundleQuestion-gated router for the curated cuTile optimization wiki: kernel-family playbooks, measured techniques with caveats, performance patterns, and practical language knowledge for Blackwell and Hopper. Use when an agent has inspected the current kernel and has a concrete question that could change its plan, when measurements point to a specific bottleneck, or when a cuTile language or API detail is blocking implementation. Do not use as mandatory onboarding, as the source of the first design, or as a fixed checklist of optimizations.
- ▌ Newton Usage · nvidia bundleSet up, build, or debug physics simulations with the Newton Python package. Use when installing Newton, choosing simulation components, finding examples, or diagnosing simulation and runtime problems.
- ▌ Newton Performance · nvidia bundleMeasure or improve the runtime, startup time, memory use, or throughput of a Newton simulation that already behaves correctly, or review a Newton step loop for performance problems such as CUDA graph capture mistakes and host synchronization. Not for setting up simulations, general solver or API questions, debugging unstable or incorrect physics without a performance goal, or optimizing Warp code that does not use Newton.
- ▌ Compileiq Bootstrap 2 · nvidia bundleUse when starting a fresh CompileIQ project, hitting a hang on the first evaluation, or before running any other compileiq-* skill. Verifies CUDA 13.3+, ptxas, GPU access, that `from compileiq.ciq import Search` and friends resolve, and that `PtxasSearchSpace().retrieve()` returns a real path. Documents the env vars that control timeouts, caching, and search-space mirroring. Triggers on "set up compileiq", "compileiq doesn't work", "first eval hangs", "where do search spaces come from", "air-gapped compileiq".
- ▌ Ising Calibration Setup · nvidia bundleUse when connecting to the NVIDIA Ising Calibration model via the hosted API or a self-hosted NIM. Covers auth, keys and connectivity checks only.
- ▌ Triton Onnx Model Repository · nvidia bundleDeploy a user's ONNX file as a working Triton Inference Server endpoint, end to end. Use when the user has a .onnx file and wants it served, wants a config.pbtxt generated from the ONNX graph, or wants to debug a Triton load or inference failure on an ONNX model. Use triton-pytorch-model-repository for .pt files and triton-tensorrt-model-repository for .plan engines.
- ▌ Triton Pytorch Model Repository · nvidia bundleDeploy a user's PyTorch model as a working Triton Inference Server endpoint, end to end. Use when the user has a TorchScript .pt file or a PT2 archive (.pt2 from torch.export + AOT Inductor) and wants it served, wants a config.pbtxt written from scratch, or wants to debug a Triton load or inference failure on a PyTorch model. Use triton-onnx-model-repository for .onnx files and triton-tensorrt-model-repository for .plan engines.
- ▌ Simready Foundation Conform Fet 006 Mdl · nvidia bundleUse for repairing exact FET_006_MDL SimReady conformance for MDL material conformance. Use when a profile, validation report, or user request names FET_006_MDL; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 025 Ros · nvidia bundleUse for repairing exact FET_025_ROS SimReady conformance for Isaac ROS-Ready bridge-node presence. Use when a profile, validation report, or user request names FET_025_ROS; default to version `0.1.0`.
- ▌ Simready Foundation Conform Fet 000 Isaac · nvidia bundleUse for repairing exact FET_000_ISAAC SimReady conformance: Isaac packaging Core (clean folder, thumbnail, physics attribute/schema source layer). Use when a profile, validation report, or user request names FET_000_ISAAC; default to version `0.1.0`.
- ▌ Simready Foundation Conform Fet 000 Physx · nvidia bundleUse for repairing exact FET_000_PHYSX SimReady conformance: the PhysX physics runtime variant scaffolding (PhysX variant set, runnables/physics/physx payload, and SimReady variant metadata). Use when a profile, validation report, or user request names FET_000_PHYSX; default to version `0.1.0`.
- ▌ Simready Foundation Conform Fet 003 Physx · nvidia bundleUse for repairing exact FET_003_PHYSX SimReady conformance for rigid body physics (physx) conformance. Use when a profile, validation report, or user request names FET_003_PHYSX; default to version `0.4.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 004 Physx · nvidia bundleUse for repairing exact FET_004_PHYSX SimReady conformance for simulate multi-body physics (physx) conformance. Use when a profile, validation report, or user request names FET_004_PHYSX; default to version `0.4.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 021 Isaac · nvidia bundleUse for repairing exact FET_021_ISAAC SimReady conformance for robot-core identity. Use when a profile, validation report, or user request names FET_021_ISAAC; default to version `0.3.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 022 Isaac · nvidia bundleUse for repairing exact FET_022_ISAAC SimReady conformance for Isaac driven-joint conformance. Use when a profile, validation report, or user request names FET_022_ISAAC; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 022 Physx · nvidia bundleUse for repairing exact FET_022_PHYSX SimReady conformance for PhysX driven-joint conformance. Use when a profile, validation report, or user request names FET_022_PHYSX; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 023 Isaac · nvidia bundleUse for repairing exact FET_023_ISAAC SimReady conformance for robot material organization conformance. Use when a profile, validation report, or user request names FET_023_ISAAC; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 024 Physx · nvidia bundleUse for repairing exact FET_024_PHYSX SimReady conformance for PhysX base-articulation conformance. Use when a profile, validation report, or user request names FET_024_PHYSX; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 028 Isaac · nvidia bundleUse for repairing exact FET_028_ISAAC SimReady conformance for Isaac gripper-site conformance. Use when a profile, validation report, or user request names FET_028_ISAAC; default to version 0.1.0 unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 100 Isaac · nvidia bundleUse for repairing exact FET_100_ISAAC SimReady conformance for Isaac composition conformance. Use when a profile, validation report, or user request names FET_100_ISAAC; default to version `0.4.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 000 Mujoco · nvidia bundleUse for repairing exact FET_000_MUJOCO SimReady conformance: the MuJoCo physics runtime variant scaffolding (MuJoCo variant set, runnables/physics/mujoco payload, and SimReady variant metadata). Use when a profile, validation report, or user request names FET_000_MUJOCO; default to version `0.1.0`.
- ▌ Simready Foundation Conform Fet 000 Newton · nvidia bundleUse for repairing exact FET_000_NEWTON SimReady conformance: the Newton physics runtime variant scaffolding (Newton variant set, runnables/physics/newton payload, and SimReady variant metadata). Use when a profile, validation report, or user request names FET_000_NEWTON; default to version `0.1.0`.
- ▌ Simready Foundation Conform Fet 003 Mujoco · nvidia bundleUse for repairing exact FET_003_MUJOCO SimReady conformance for MuJoCo rigid-body physics, including MjcCollisionAPI and MjcMeshCollisionAPI. Use when a profile, validation report, or user request names FET_003_MUJOCO; default to version `0.1.0` unless pinned otherwise.
- ▌ Simready Foundation Conform Fet 003 Newton · nvidia bundleUse for repairing exact FET_003_NEWTON SimReady conformance for rigid body physics (newton) conformance. Use when a profile, validation report, or user request names FET_003_NEWTON; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 004 Mujoco · nvidia bundleUse for repairing exact FET_004_MUJOCO SimReady conformance for MuJoCo multibody physics without creating geometry. Use when a profile, validation report, or user request names FET_004_MUJOCO; default to version `0.1.0` unless pinned otherwise.
- ▌ Simready Foundation Conform Fet 004 Newton · nvidia bundleUse for repairing exact FET_004_NEWTON SimReady conformance for simulate multi-body physics (newton) conformance. Use when a profile, validation report, or user request names FET_004_NEWTON. FET_004_NEWTON has a single published version, 1.
- ▌ Simready Foundation Conform Fet 022 Mujoco · nvidia bundleUse for repairing exact FET_022_MUJOCO SimReady conformance for MuJoCo driven-joint conformance. Use when a profile, validation report, or user request names FET_022_MUJOCO; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 022 Newton · nvidia bundleUse for repairing exact FET_022_NEWTON SimReady conformance for Newton driven-joint conformance. Use when a profile, validation report, or user request names FET_022_NEWTON; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 024 Mujoco · nvidia bundleUse for repairing exact FET_024_MUJOCO SimReady conformance for MuJoCo-compatible base articulation. Use when a profile, validation report, or user request names FET_024_MUJOCO; default to version `0.1.0` unless pinned otherwise.
- ▌ Simready Foundation Conform Fet 024 Newton · nvidia bundleUse for repairing exact FET_024_NEWTON SimReady conformance for Newton base-articulation conformance. Use when a profile, validation report, or user request names FET_024_NEWTON; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 028 Mujoco · nvidia bundleUse for repairing exact FET_028_MUJOCO SimReady conformance for MuJoCo gripper-site authoring with MjcSiteAPI. Use when a profile, validation report, or user request names FET_028_MUJOCO; default to version 1 unless pinned otherwise.
- ▌ Simready Foundation Conform Fet 006 Openpbr · nvidia bundleUse for repairing exact FET_006_OPENPBR SimReady conformance for OpenPBR (MaterialX) material conformance. Use when a profile, validation report, or user request names FET_006_OPENPBR; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 000 Standard · nvidia bundleUse for repairing FET_000_STANDARD Core failures (default `0.2.0`; use `0.1.0` when a profile pins it): asset file naming, directory layout, relative and resolvable paths, SimReady metadata, and undefined prims.
- ▌ Simready Foundation Conform Fet 001 Standard · nvidia bundleUse for repairing SimReady Minimal assets: units, upAxis, defaultPrim, hierarchy, mesh quality, extents, and origin placement.
- ▌ Simready Foundation Conform Fet 002 Standard · nvidia bundleUse for repairing exact FET_002_STANDARD SimReady conformance for posable body conformance. Use when a profile, validation report, or user request names FET_002_STANDARD; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 003 Standard · nvidia bundleUse for repairing exact FET_003_STANDARD SimReady conformance for rigid body physics conformance. Use when a profile, validation report, or user request names FET_003_STANDARD; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 004 Standard · nvidia bundleUse for repairing exact FET_004_STANDARD SimReady conformance for simulate multi-body physics conformance. Use when a profile, validation report, or user request names FET_004_STANDARD; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 005 Standard · nvidia bundleUse for vision-guided SimReady grasp repair, grasp_identifier curves, and physics-material triage.
- ▌ Simready Foundation Conform Fet 006 Standard · nvidia bundleUse for repairing exact FET_006_STANDARD SimReady conformance for USDPreviewSurface material conformance. Use when a profile, validation report, or user request names FET_006_STANDARD; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 007 Standard · nvidia bundleUse for repairing SimReady nonvisual sensor material attributes on bound USD materials.
- ▌ Simready Foundation Conform Fet 010 Standard · nvidia bundleUse for repairing exact FET_010_STANDARD SimReady conformance for display colour (`primvars:displayColor`) on renderable geometry. Use when a profile, validation report, or user request names FET_010_STANDARD, DISP.001 or DISP.002; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 011 Standard · nvidia bundleUse for repairing exact FET_011_STANDARD SimReady conformance for semantic label conformance. Use when a profile, validation report, or user request names FET_011_STANDARD; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 022 Standard · nvidia bundleUse for repairing exact FET_022_STANDARD SimReady conformance for standard driven-joint conformance. Use when a profile, validation report, or user request names FET_022_STANDARD; default to version `0.2.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 024 Standard · nvidia bundleUse for repairing exact FET_024_STANDARD SimReady conformance for standard base-articulation conformance. Use when a profile, validation report, or user request names FET_024_STANDARD; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 028 Standard · nvidia bundleUse for repairing exact FET_028_STANDARD SimReady conformance for standard gripper-site conformance. Use when a profile, validation report, or user request names FET_028_STANDARD; default to version 1 unless a profile or report pins another integer version.
- ▌ Simready Foundation Conform Fet 030 Standard · nvidia bundleUse for repairing exact FET_030_STANDARD SimReady conformance for packaging core conformance. Use when a profile, validation report, or user request names FET_030_STANDARD; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 031 Standard · nvidia bundleUse for repairing exact FET_031_STANDARD SimReady conformance for self-contained package source conformance. Use when a profile, validation report, or user request names FET_031_STANDARD; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 032 Standard · nvidia bundleUse for repairing exact FET_032_STANDARD SimReady conformance for packaging introspection conformance. Use when a profile, validation report, or user request names FET_032_STANDARD; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 033 Standard · nvidia bundleUse for repairing exact FET_033_STANDARD Metadata conformance (thumbnail and provenance metadata). Use when a profile, validation report, or user request names FET_033_STANDARD; default to version `0.4.0` unless a profile or report pins another version.
- ▌ Simready Foundation Conform Fet 004 Robot Physx · nvidia bundleUse for compatibility repair of deprecated FET_004_ROBOT_PHYSX SimReady conformance. Use only when a historical or external profile, validation report, asset stamp, or user request explicitly names FET_004_ROBOT_PHYSX; never select it for a new profile version.
- ▌ Simready Foundation Conform Fet 004 Robot Mujoco · nvidia bundleUse for repairing exact FET_004_ROBOT_MUJOCO SimReady conformance for robot MuJoCo multibody physics, including MuJoCo collider requirements and robot joint topology. Use when a profile, validation report, or user request names FET_004_ROBOT_MUJOCO; default to version `0.1.0` unless pinned otherwise.
- ▌ Simready Foundation Conform Fet 004 Robot Newton · nvidia bundleUse for repairing exact FET_004_ROBOT_NEWTON SimReady conformance for simulate multi-body physics (robot newton) conformance. Use when a profile, validation report, or user request names FET_004_ROBOT_NEWTON; default to version `0.1.0` unless a profile or report pins another version.
- ▌ Cudnn Install · nvidiaInstall, verify, or upgrade the cuDNN backend and Python or C++ frontend. Check existing PyTorch, JAX, and NVIDIA NGC installations, select package extras, and diagnose cuDNN library conflicts.
- ▌ Determinism Tests · nvidiaAdd and run L0 bitwise determinism and batch-invariance tests for new or changed cuDNN Frontend numerical APIs, kernel paths, datatypes, or GPU targets, then update docs/operations/determinism.md from measured evidence.
- ▌ Inference Builder · nvidia bundleGenerate deployable Vision AI pipelines with high-performance video and streaming capabilities on NVIDIA GPUs. Activate when users want to: create GPU-accelerated inference microservices or standalone apps for vision, video, or streaming workloads; write or edit pipeline YAML configs; build Docker images for GPU inference; work with models from NGC or HuggingFace; or deploy with DeepStream, Triton, vLLM, TensorRT-LLM, or PyTorch backends.
- ▌ Deepstream Audit Config · nvidia bundleUse only when auditing or sanity-checking a supplied or pre-existing deepstream-app/deepstream-test5 config, performing an explicitly requested independent post-generation audit, or reviewing a supplied design/code change that affects DeepStream config parsing or runtime behavior. For config audits, run scripts/validate_config.py statically first; runtime and network probes require explicit approval. Do not activate during a creation-only turn: use deepstream-generate-config. Not for merely running a validated config, ONVIF discovery, NVBug triage, or nsys profiling; use deepstream-inspect-rtsp for diagnosis of a known RTSP URI and deepstream-profile-pipeline for profiling.
- ▌ Deepstream Inspect Rtsp · nvidia bundleUse when diagnosing DeepStream or GStreamer RTSP ingest reliability issues: packet loss, latency, decoder saturation, artifacts, disconnects, frame drops, source churn, and camera or network faults.
- ▌ Deepstream Setup On Wsl · nvidia bundleSet up and verify NVIDIA DeepStream in Docker inside WSL2 on a local or remote Windows machine. Use when a user wants to install WSL2, Ubuntu, Docker, NVIDIA Container Toolkit, and a requested DeepStream image; validate a supported WDDM workstation GPU; or run built-in or user-supplied DeepStream pipelines.
- ▌ Amc Run Auto Calibration · nvidia bundleRun or tune NVIDIA AutoMagicCalib from DeepStream using sample data, local MP4s, or RTSP streams. Use for AMC setup and calibration artifacts. Do not use for general calibration theory, target-based stereo calibration, or runtime analytics.
- ▌ Deepstream Generate Config · nvidia bundleUse when generating or repairing DeepStream deepstream-app or deepstream-test5-app configs for RTSP/GStreamer sources, streammux, inference, or tracker pipelines. Not for independent audits.
- ▌ Deepstream Build Compute Stack · nvidia bundleProvision Ubuntu 24.04 NVIDIA stacks with Ansible on x86_64 or ARM SBSA, with Xorg setup and optional DeepStream dependencies. Not for Jetson.
- ▌ Deepstream Import Vision Model 3 · nvidia bundleImport detection, classification, segmentation, pose, and depth models into DeepStream with task routing and evidence-gated validation.
- ▌ Edge LLM Serve · nvidiaLaunch and verify TensorRT Edge-LLM's experimental OpenAI-compatible server for one Hugging Face checkpoint or local checkpoint. Use for `trtedge serve`, OpenAI API requests, server sizing, streaming, tools, media, batching, speculative decoding, or a server startup failure. Do not use for ONNX export or C++ inference (edge-llm-convert-model), installation (edge-llm-install), or adding architecture support (edge-llm-add-model).
- ▌ Edge LLM Install · nvidia bundleSelect, install, verify, update, or remove TensorRT Edge-LLM, and install this repository's agent skills into Claude Code or Codex. Inspects revision, CPU architecture, GPU or SoC, JetPack/DriveOS, CUDA, TensorRT, and Python, rejects unsupported combinations before changing anything, then picks exactly one route - published wheel, local wheel, C++ source for the ONNX workflow, or Python source for the experimental checkpoint-direct builder and server - and proves it with a per-route gate. Use when asked to install, set up, build from source, cross-compile, pip install tensorrt-edgellm, upgrade the installed package or build in place, uninstall, fix a CMake or wheel-load failure, or install, update, or remove the edge-llm skills. Do not use for running a model (use edge-llm-serve or edge-llm-convert-model), writing or selecting tests (edge-llm-e2e-test, edge-llm-cpp-test), accuracy debugging (edge-llm-few-layer-test, edge-llm-accuracy-debugging), or release-to-release migration (edge-llm-version-migration).
- ▌ Edge LLM Cpp Test · nvidia bundleWrite or change a C++ test under unittests/. Covers the BDD comment contract, mutation testing, and the substitute engines that drive LLMInferenceRuntime without a serialized engine. Apply whenever adding, migrating, or repairing a GTest case in this repository.
- ▌ Edge LLM E2e Test · nvidia bundleWrite, select, run, and verify TensorRT Edge-LLM end-to-end model tests. Use when adding or changing a model case, running one model deployment instead of an entire CI job, interpreting model accuracy or baseline failures, or selecting sufficient tests for Python, C++, CUDA, CuTe DSL, plugin, export, builder, runtime, or server changes.
- ▌ Edge LLM Add Model · nvidia bundleImplement TensorRT Edge-LLM support for a new Hugging Face checkpoint or model family. Use for a new `model_type`, architecture, text/vision/audio/action component, checkpoint tensor layout, speculative draft, model-specific custom operation, or runtime artifact. Starts from checkpoint inspection and the provider reference, finds the closest existing family, chooses the smallest supported frontend surface, and requires reference-to-export-to-engine-to-inference evidence. Do not use merely to convert an already supported checkpoint (edge-llm-convert-model), serve it (edge-llm-serve), or develop a standalone plugin/kernel (edge-llm-plugin-extension, edge-llm-cutedsl-kernel).
- ▌ Edge LLM Convert Model · nvidiaConvert and run one supported Hugging Face checkpoint with TensorRT Edge-LLM. Use for quantization, checkpoint-to-ONNX export, `llm_build`/`visual_build`/`audio_build`, C++ `llm_inference`, the experimental checkpoint-direct `trtedge build`, engine profile sizing, or copying ONNX artifacts to an Edge target. Do not use for HTTP serving (edge-llm-serve), implementing a new architecture (edge-llm-add-model), installation (edge-llm-install), or accuracy triage (edge-llm-few-layer-test).
- ▌ Edge LLM Documentation · nvidia bundleWrite, edit, or review TensorRT Edge-LLM documentation using the Google developer documentation style guide. Apply to READMEs, tutorials, how-to guides, conceptual and design documents, API and CLI references, and release notes; not ordinary code changes or chat replies.
- ▌ Edge LLM Cutedsl Kernel · nvidia bundleAdd or change a TensorRT Edge-LLM CuTe DSL kernel end to end, from the Python kernel source through the KERNEL_VARIANTS registry in kernelSrcs/build_cutedsl.py, GPU-free AOT artifact generation (native or AArch64 cross build), the metadata-driven CUTE_DSL_*_ENABLED defines in cmake/CuteDsl.cmake, and the C++ runner and SM dispatch, to numerical tests on every claimed SM and prebuilt tarball packaging. Use when adding a CuTe DSL kernel group or variant, changing supported_sms, regenerating cpp/kernels/cuteDSLArtifact, debugging a missing CuTe DSL define, symbol, or module-load failure, or preparing cutedsl tarballs. Do not use it to create or register a TensorRT plugin (use edge-llm-plugin-extension), for GTest authoring rules (use edge-llm-cpp-test), or for model deployment tests (use edge-llm-e2e-test).
- ▌ Edge LLM Few Layer Test · nvidia bundleAdd or repair a few-layer numeric validation case, which compares the first N decoder layers of a checkpoint against a PyTorch golden. Covers when a change needs one, choosing the checkpoint, setting a cosine gate that is measured rather than tuned, and triaging a failure. Apply when adding a model architecture, layer type, quantization recipe, or kernel, or when a few-layer case fails.
- ▌ Edge LLM Plugin Extension · nvidia bundleAdd or change a TensorRT Edge-LLM operation (TensorRT IPluginV3 plugin, native TensorRT layer composition, or runtime-only operation) so one tensor, dtype, shape, layout, serialization, and weight-layout contract holds across the ONNX export route, the experimental checkpoint-direct builder, the C++ engine builders, libNvInfer_edgellm_plugin, and the runtime. Use when adding a custom op, plugin creator field, plugin input or output, dtype, weight repacking, or external-weight recipe; when a plugin is not found or rejects a configuration at engine build; or when choosing between a plugin and native TensorRT layers. Do not use for writing CuTe DSL kernels (use edge-llm-cutedsl-kernel), C++ test discipline (edge-llm-cpp-test), model CI cases (edge-llm-e2e-test), or diagnosing wrong model output (edge-llm-few-layer-test, then edge-llm-accuracy-debugging).
- ▌ Edge LLM Version Migration · nvidia bundleMigrate an existing TensorRT Edge-LLM deployment, integration, or downstream patch between releases, branches, TensorRT/CUDA/JetPack/DriveOS stacks, or checkpoint-family versions; also backport fixes to a release branch. Use when commands, package extras, CMake options, plugins, ONNX, engines, server requests, or downstream code must be mapped old-to-new. Inventories both revisions, generates a source-backed surface diff, rebuilds derived artifacts, and proves the migrated workload. Do not use for a first install (edge-llm-install), a supported-model conversion (edge-llm-convert-model), or new architecture support (edge-llm-add-model).
- ▌ Edge LLM Accuracy Debugging · nvidia bundleLocalize a TensorRT Edge-LLM accuracy mismatch against the model provider's PyTorch reference after few-layer validation identifies a failing component or layer.
- ▌ Edge LLM Benchmark Accuracy · nvidiaMeasure dataset accuracy (MMLU and other NeMo Evaluator benchmarks) of a TensorRT Edge-LLM deployment through its OpenAI-compatible chat endpoint with NeMo Evaluator. Use for "benchmark accuracy", "run MMLU/GSM8K/IFEval on this checkpoint", comparing a quantization, speculative-decoding, or engine-profile change against a baseline score, or gating a change on a minimum score. Do not use to launch a server only (edge-llm-serve), to measure latency or throughput (edge-llm-benchmark-performance), or to locate which layer diverges after a mismatch (edge-llm-few-layer-test, edge-llm-accuracy-debugging).
- ▌ Edge LLM Feature Combinations · nvidia bundleList the supported combinations a change reaches and verify them before the change is submitted. Use before coding a change to quantization or checkpoint parsing, checkpoint loading, the lm_head, KV cache, a speculative decoding method or draft, shared linear or attention code, either engine frontend (ONNX exporter or checkpoint-direct builder), a kernel or TensorRT plugin, tensor parallelism, or a C++ feature guard. Skip it for documentation, CI, test-only, and server-protocol changes, and for a fix that stays inside one model family on one frontend and does not touch quantized weight loading.
- ▌ Edge LLM Benchmark Performance · nvidiaBenchmark TensorRT Edge-LLM serving performance (TTFT, inter-token latency, request latency, output throughput, concurrency scaling) through the OpenAI-compatible server with AIPerf. Use for "benchmark performance", "measure TTFT/ITL/tokens per second", prefill or decode sweeps over input and output lengths, in-flight-batching concurrency scaling, or comparing two deployments (quantization, speculative decoding, engine profile, another runtime). Do not use to launch a server only (edge-llm-serve), to score dataset accuracy (edge-llm-benchmark-accuracy), or to profile kernels.
- ▌ Serve Model · nvidiaUse when serving a cached model on one GPU with vLLM, checking a local completion endpoint, or diagnosing startup and inference failures in the single-node quickstart.
- ▌ Finetune Model · nvidiaUse when checking whether one DeepOps Slurm GPU node is prepared for fine-tuning, diagnosing missing cached training inputs or container prerequisites, or distinguishing readiness from actual training success.
- ▌ Model Workflows · nvidiaUse when choosing a DeepOps model workflow, combining shared model caching with single-node serving, or determining whether a fine-tuning or multi-node request has a supported procedure.
- ▌ Deploy Airgapped · nvidiaPrepare mirrors and transfer artifacts, configure DeepOps, deploy Slurm or Kubernetes GPU clusters without Internet access, and validate them with machine-readable gates. Use for disconnected, restricted-egress, offline, or air-gapped DeepOps installations and for diagnosing missing package, file, chart, or container artifacts.