NVIDIA

Migrate applications, examples, integrations, documentation, package manifests, and repository code from NeMo Flow naming and packages to NeMo Relay across Python, Rust, Node.js, Go, WebAssembly, C FFI, CLI, config, and observability surfaces; use when a user asks to rename nemo_flow/nemo-flow/NeMo Flow APIs, automate a migration, update imports or dependencies, or validate a Flow-to-Relay conversion

Add or change a public NeMo Relay API surface across the core runtime and every affected binding

Add a new third-party framework integration maintained as a NeMo Relay patch set

Add a new guardrail or intercept type to the NeMo Relay middleware pipeline

Contribute a new NeMo Relay public API surface safely, with binding parity and docs in mind

Contribute documentation or example changes that stay aligned with NeMo Relay public behavior

Contribute a new or updated third-party framework integration for NeMo Relay

Maintain and review NeMo Relay GitHub Actions workflows with explicit per-job permissions, pinned action SHAs, deterministic caching, reusable workflow permission boundaries, and local validation

Onboard a HuggingFace multimodal model (vision/audio/video + text) to the TensorRT-LLM PyTorch backend. Use when writing a new `tensorrt_llm/_torch/models/modeling_<vlm>.py` plus its input processor and weight mapper, or extending an existing VLM. Not for AutoDeploy — use `ad-model-onboard` for that path.

Debug AutoDeploy accuracy regressions vs a reference score (PyTorch backend or published baseline). Use when an AutoDeploy model's eval score is significantly below the reference and the root cause is unknown.

Claude Code skill (trtllm-agent-toolkit): implement or extend TensorRT-LLM AutoDeploy fusion transforms under transform/library/ in a TensorRT-LLM checkout. Prefer existing kernels and custom ops; use Triton only when no viable existing-kernel path exists. Use ad-graph-dump for AD_DUMP_GRAPHS_DIR workflows. Covers TRT-LLM paths, registry, default.yaml registration, graph validation, tests, and a review checklist — without prescribing profiling tools or throughput targets.

Enable and interpret TensorRT-LLM AutoDeploy FX graph text dumps via AD_DUMP_GRAPHS_DIR. Use when you need before/after graphs per transform, to locate subgraphs, or to confirm a rewrite ran. Paths and behavior are grounded in tensorrt_llm/_torch/auto_deploy (GraphWriter, BaseTransform). Complements ad-add-fusion-transformation.

Visualize a specific transformer decoder layer from an AutoDeploy FX graph text dump as a hierarchical DOT/PNG diagram. Optionally annotate nodes with actual GPU kernel names and durations from an nsys trace. Use when the user wants to visualize, inspect, or debug a layer in an AutoDeploy model graph dump. Triggers on: "visualize layer", "show layer", "graph of layer", "layer visualization", "dump graph layer". Assumes graph dumps already exist in a directory (produced by AD_DUMP_GRAPHS_DIR).

Write and implement GPU kernels using NVIDIA CuTe DSL (CUTLASS 4.x Python API) — NOT for Triton, CUDA C++, or conceptual explanations. Trigger only when the user wants to write or implement a kernel, not when asking questions about CuTe DSL concepts or layouts. CuTe DSL uses cute.jit/cute.kernel decorators and cutlass.cute imports. Covers element-wise kernels, GEMM patterns, reductions, memory hierarchy (global/shared/register/TMA), MMA tensor core operations, software pipelining, and framework integration.

Optimize existing Triton kernels for NVIDIA TileIR backend on Blackwell GPUs (sm_100+). Adds TileIR-specific autotune configs: occupancy, num_ctas, TMA descriptors. Covers kernel classification (dot-related, norm-like, elementwise, reduction), type-specific transformations, and PTX-vs-TileIR benchmarking. Triggered by: "optimize for TileIR", "add TileIR configs", "Blackwell optimization", "TMA descriptors", "2CTA mode", "occupancy tuning". Kernels use standard `import triton`; TileIR activates via ENABLE_TILE=1 when nvtriton is installed.

ONLY for OpenAI Triton (@triton.jit) kernel development. NEVER use for CUDA C++ kernels, TileIR, or profiling tools (ncu, nsys). The user's request must involve Triton explicitly. Covers Triton-specific patterns: fused elementwise, reductions (softmax, LayerNorm, RMSNorm), tiled GEMM with triton.autotune, and flash attention. Workflow: design, write, verify (with fast-path for explicit requests).

Connects NemoClaw to a local inference server. Use when setting up Ollama, vLLM, TensorRT-LLM, NIM, or any OpenAI-compatible local model server with NemoClaw. Trigger keywords - nemoclaw local inference, ollama nemoclaw, vllm nemoclaw, local model server, openai compatible endpoint, switch nemoclaw inference model, change inference runtime, nemoclaw additional model, nemoclaw sub-agent model, openclaw sub-agent, agents.list, sessions_spawn, vlm-demo, nemoclaw tool calling, ollama tool calls, vllm tool-call-parser, raw json in tui, nemoclaw inference options, nemoclaw onboarding providers, nemoclaw inference routing.

Installs NemoClaw, launches a sandbox, and runs the first agent prompt. Use when onboarding, installing, or launching a NemoClaw sandbox for the first time. Trigger keywords - nemoclaw quickstart, install nemoclaw openclaw sandbox, nemohermes quickstart, hermes agent nemoclaw, run hermes openshell sandbox, nemoclaw prerequisites, nemoclaw supported platforms, nemoclaw hardware software, nemoclaw windows wsl2 setup, nemoclaw install windows docker desktop.

Explains operational tasks after the quickstart: listing sandboxes, status and health checks, logs, diagnostics, port forwards, multiple sandboxes, credential reset, rebuilds, network presets, upgrades, and uninstall. Trigger keywords - manage nemoclaw sandboxes, nemoclaw status, nemoclaw list, nemoclaw dashboard port, nemoclaw rebuild, nemoclaw upgrade sandboxes, nemoclaw uninstall, sandbox mutability, sandbox runtime configuration, sandbox rebuild, nemoclaw backup, nemoclaw restore, workspace backup, openshell sandbox download upload, nemoclaw messaging channels, nemoclaw telegram, nemoclaw discord, nemoclaw slack, nemoclaw whatsapp, openshell channel messaging, nemoclaw workspace files, soul.md, user.md, identity.md, agents.md, sandbox persistence.

Explains how OpenClaw, OpenShell, and NemoClaw form the ecosystem, NemoClaw's position in the stack, what NemoClaw adds beyond the community sandbox, and when to prefer NemoClaw versus integrating OpenShell and OpenClaw directly. Use when users ask about the relationship between OpenClaw, OpenShell, and NemoClaw, or when to use NemoClaw versus OpenShell. Trigger keywords - nemoclaw ecosystem, openclaw openshell, nemoclaw vs openshell, sandboxed openclaw, how nemoclaw works, nemoclaw sandbox lifecycle blueprint, nemoclaw overview, openclaw always-on assistants, nvidia openshell, nvidia nemotron, nemoclaw release notes, nemoclaw changelog.

Describes the NemoClaw plugin and blueprint architecture and how they orchestrate the OpenClaw sandbox. Use when looking up architecture, plugin structure, or blueprint design. Trigger keywords - nemoclaw architecture, nemoclaw plugin blueprint structure, nemoclaw vs openshell, which cli, nemoclaw cli, openshell cli, sandbox commands, nemoclaw cli commands, nemoclaw command reference, nemoclaw network policy, sandbox egress control operator approval, nemoclaw troubleshooting, nemoclaw debug sandbox issues.

Presents a risk framework for every configurable security control in NemoClaw. Use when evaluating security posture, reviewing sandbox security defaults, or assessing control trade-offs. Trigger keywords - nemoclaw security best practices, sandbox security controls risk framework, nemoclaw credential storage, openshell provider, api key security, openclaw security controls, nemoclaw security boundary, prompt injection, tool access control.

Adds, removes, or modifies allowed endpoints in the sandbox policy. Use when customizing network policy, changing egress rules, or configuring sandbox endpoint access. Trigger keywords - customize nemoclaw network policy, sandbox egress policy configuration, nemoclaw integration policy examples, post-install policy setup, openshell approval workflow, policy preset, nemoclaw approve network requests, sandbox egress approval tui.

Scan recent git commits for changes that affect user-facing behavior, then draft or update the corresponding documentation pages and refresh generated user skills for release prep. Use when docs have fallen behind code changes, after a batch of features lands, during daily release prep, or when preparing a release. Trigger keywords - update docs, draft docs, docs from commits, sync docs, catch up docs, doc debt, docs behind, docs drift, release prep docs, refresh user skills.

Manually dispatch and iterate on the Release Canary workflow that smoke-tests published OpenShell artifacts (install.sh on macOS/Ubuntu/Fedora, Helm chart on kind) after each Release Dev publish. Use when changing `.github/workflows/release-canary.yml`, validating a release before tagging, debugging a canary failure, or reproducing a canary job locally. Trigger keywords - release canary, release-canary, canary failed, canary dispatch, test release canary, post-release smoke, install.sh canary, helm chart canary, kind canary, dispatch canary.

Debug why an OpenShell gateway deployment is unhealthy, unreachable, or unable to create sandboxes. Use when the user has a gateway health failure, Docker/Podman runtime issue, Helm install failure, Kubernetes scheduling issue, TLS secret issue, VM driver issue, or sandbox startup problem. Trigger keywords - debug gateway, gateway failing, deployment failing, helm install failing, cluster health, gateway health, gateway not starting, health check failed, sandbox pending, docker driver, podman driver, vm driver.

Start up, tear down, and configure the local Kubernetes development environment for OpenShell. Uses k3d (Docker-backed k3s) + Skaffold + Helm. Covers cluster lifecycle, optional add-ons (Keycloak OIDC, Envoy Gateway), and port mappings. Trigger keywords - local k8s, local cluster, k3d, skaffold, helm dev, start cluster, stop cluster, tear down cluster, delete cluster, create cluster, helm:k3s, helm:skaffold, local dev environment, dev cluster, k8s dev, envoy gateway local, keycloak local.

Given a GitHub issue number, plan and implement the work described in the issue. Operates iteratively - creates an implementation plan, responds to feedback, and only builds when the 'state:agent-ready' label is applied. Includes tests, documentation updates, and PR creation. Trigger keywords - build from issue, implement issue, work on issue, build issue, start issue.

Investigate a plain-language problem description by deeply exploring the codebase, then create a structured GitHub issue with technical findings. Prequel to build-from-issue — maps vague ideas to concrete, buildable issues. Trigger keywords - spike, investigate, explore, research issue, technical investigation, create spike, new spike, feasibility, codebase exploration.

Guide agents through using the OpenShell CLI (openshell) for sandbox management, gateway registration, provider configuration, policy iteration, BYOC workflows, and inference routing. Covers basic through advanced multi-step workflows. Trigger keywords - openshell, sandbox create, sandbox connect, logs, provider create, policy set, policy get, image push, forward, port forward, BYOC, bring your own container, use openshell, run openshell, CLI usage, manage sandbox, manage provider, gateway add, gateway select.

Create GitHub pull requests using the gh CLI. Use when the user wants to create a new PR, submit code for review, or open a pull request. Trigger keywords - create PR, pull request, new PR, submit for review, code review.

Implement a fix for a reviewed security issue. Takes an issue number or scans for issues labeled "topic:security" and "state:agent-ready". Reads the security review from the issue comments and implements the remediation plan. Trigger keywords - fix security issue, remediate security, implement security fix, patch vulnerability.

Use when investigating Kit app livestream performance bottlenecks, WebRTC/native StreamSDK lag, freezes, dropped frames, browser WebRTC stats, copy fence timeouts, NVST_R_BUSY, disconnects, or resolution mismatch warnings in omni.kit.livestream.

Profile Python functions with NVTX in non-Kit environments (Isaac Lab 3.0+ standalone, any Python app without Carbonite). Uses a bundled PYTHONPATH-scoped sitecustomize.py with sys.setprofile hook, NVTX push/pop ranges, module include/exclude filtering, and Nsight Systems integration. Use when CARB_PROFILING_PYTHON doesn't work (no Kit/Carbonite runtime), when profiling standalone Isaac Lab scripts, or when you need per-function Python tracing in nsys captures outside Kit.

Resolve common Kit/Isaac Sim/Isaac Lab performance issues using specific settings and configuration changes. Covers PresentFrame stalls, resolveSamplerFeedback, headless mode, multi-GPU tradeoffs, DLSS/DLSS-G, PhysX tuning, RTX presets (isaaclab_performance/balanced/quality), viewport gizmos, HydraEngine waitIdle, fsWatcher overhead, and CPU governor. Use when profiling data shows a specific bottleneck and you need the fix, when someone asks "why is it slow" and you have Tracy/nsys evidence, or when tuning RTX settings for GPU-bound workloads. NOT for: initial triage (use diagnose-perf), capturing profiles (use profiling), or analyzing traces (use nsys-analyze).

Install profiling tools for Isaac Sim / Isaac Lab / Kit-based applications. Covers Nsight Systems (`nsys` CLI), `sqlite3`, Tracy `csvexport`, canonical Tracy `capture`/`capture-release`, and `update` for memory strip tests, with optional `tracy-capture`/`tracy-update` aliases. Use when setting up a profiling environment, when nsys/sqlite3/csvexport/capture/update tools are missing, or before running profiling, nsys-analyze, or tracy-memory.

Analyze profiling data from Kit-based apps. Covers Omniverse-specific NVTX zone interpretation, phase detection using sqlite3, Tracy Statistics/Range Limit analysis, csvexport fallback queries, and two-version comparison methodology. Use after capturing profiles with the profiling skill. NOT for capturing traces (use profiling), adding zones to code (use profiling-api), or applying fixes (use perf-tuning).

Add profiling zones, metrics, and annotations to Kit-based C++ and Python code. Covers Carbonite macros (CARB_PROFILE_ZONE, CARB_PROFILE_FUNCTION, GPU zones), Python profiler API (decorators, begin/end), profiler masks, channels, Tracy plot data, event annotations, and automatic Kit Python function capture (CARB_PROFILING_PYTHON). Use when a developer asks how to add profiling spans to Kit/Carbonite code, configure masks/channels, record custom Tracy plots, or annotate traces with event markers. NOT for capturing traces (use profiling), analyzing traces (use nsys-analyze), or non-Kit Python function tracing (use nvtx-python).

Capture performance traces using CPU ChromeTrace, Tracy, and Nsight Systems/NVTX for Kit-based applications (Isaac Sim, Isaac Lab, Kit SDK). Covers COLD/WARM/TRACY measurement separation, canonical Tracy capture sequence, last-resort force-kill handling, nsys profile commands, Kit profiler args, and lightweight export handoff to nsys-analyze. Use when running profiling captures, setting up trace collection, or troubleshooting capture failures. NOT for adding profiling zones (use profiling-api), deep trace analysis (use nsys-analyze), memory allocation profiling (use tracy-memory), or applying performance fixes (use perf-tuning).

Profile CPU and GPU memory allocations using Tracy in Kit-based applications after Tracy capture tooling is installed. Covers LD_PRELOAD setup for liballocwrapper.so, Kit memory-channel flags, capture binary isolation (unset LD_PRELOAD), tracy-update strip-test verification, Tracy Memory tab analysis, and debug symbol requirements. Use when investigating memory leaks, allocation hotspots, or VRAM growth in Isaac Sim, Isaac Lab, or Kit apps. Requires profiling capture setup plus install-profilers. NOT for generic trace capture (use profiling) or non-memory trace analysis (use nsys-analyze).

Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.

Bump the NVIDIA PyTorch base image (`nvcr.io/nvidia/pytorch:<YY.MM>-py3`) used by Megatron-LM CI. Covers the two pin sites (GitHub CI in `docker/.ngc_version.dev` and GitLab CI in `.gitlab/stages/01.build.yml`), the post-bump CI loop (re-run functional tests, refresh golden values, mark broken tests), and the gotchas that bit PRs

Refresh golden values from a GitHub Actions workflow run (failing-only or all jobs), score the change with average normalized relative differences, and produce a PR-ready summary. Use when the user asks to update goldens for a CI run, refresh golden values from a workflow ID, or generate a golden-value diff summary for a PR description.

Domain knowledge for the nightly main-to-dev sync workflow. Covers merge strategy, CI architecture, failure investigation, and known issues.

Container-based dev environment setup and dependency management for Megatron-LM. Covers acquiring and launching the CI container, uv package management, and updating uv.lock.

CI/CD reference for Megatron-LM. Covers CI pipeline structure, PR scope labels, triggering internal GitLab CI, and CI failure investigation.

Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.

Onboard 1-node GitHub MR functional tests for GB200 from existing mr-scoped 2-node tests.

Modify, build, test, debug, and contribute to NVIDIA cuOpt (C++/CUDA, Python, server, CI). Use for solver internals, PRs, DCO, and code conventions.

After solving a non-trivial problem, detect generalizable learnings and propose skill updates so future interactions benefit automatically. Always active — applies to every interaction.

Install cuOpt for Python, C, or as a server (pip, conda, Docker) — system requirements, install commands, and verification. Use when the user wants to install or verify cuOpt for any user-facing interface. For building cuOpt from source or contributing to cuOpt, see cuopt-developer.

LP, MILP, and QP (beta) with cuOpt — C API only. Use when the user is embedding LP, MILP, or QP in C/C++.

LP, MILP, and QP (beta) with cuOpt — CLI only (MPS files, cuopt_cli). Use when the user is solving LP, MILP, or QP from MPS via command line.

Solve Linear Programming (LP), Mixed-Integer Linear Programming (MILP), and Quadratic Programming (QP, beta) with the Python API. Use when the user asks about optimization with linear or quadratic objectives, linear constraints, integer variables, scheduling, resource allocation, facility location, production planning, portfolio optimization, or least squares.

Vehicle routing (VRP, TSP, PDP) with cuOpt — Python API only. Use when the user is building or solving routing in Python.

cuOpt REST server — start server, endpoints, Python/curl client examples. Use when the user is deploying or calling the REST API.

Use before creating, editing, or deciding whether to update any AI coding agent skill in this repository, including corrections to existing skill behavior, references, or routing.

Use when selecting, configuring, composing, or troubleshooting NeMo Agent Toolkit agents and control-flow components, including ReAct, tool-calling, ReWOO, reasoning, router, sequential, parallel, and sub-agent patterns.

Use when designing, configuring, running, or troubleshooting NeMo Agent Toolkit evaluations, datasets, evaluator selection, ATIF surfaces, quality gates, custom evaluators, and `nat eval`.

Use when installing or configuring NVIDIA NeMo Agent Toolkit, verifying the `nat` CLI, setting up optional extras, or creating a first hello-world workflow.

Use when serving NeMo Agent Toolkit workflows, exposing workflows through FastAPI, configuring MCP clients or servers, or troubleshooting transport and server setup.

Use when configuring or running NeMo Agent Toolkit optimization with `nat optimize`, including Optuna parameter tuning, prompt evolution, optimizer sizing, output interpretation, and optimizer datasets.

Use when fixing NeMo Agent Toolkit documentation path-check failures, especially failed `ci/scripts/path_checks.py` output, slash-delimited text mistaken for paths, relative path references, Markdown code escaping, and path-check allowlist decisions.

Use when adding, configuring, or troubleshooting NeMo Agent Toolkit logging, tracing, telemetry exporters, OpenTelemetry, Langfuse, LangSmith, Weave, Phoenix, profiling, or observability provider integrations.

Unified accessibility auditing workflow across static analysis, runtime testing, ARIA patterns, keyboard navigation, and color contrast. Use this skill whenever the user mentions accessibility, a11y, WCAG, ARIA roles, axe tests, screen readers, focus management, keyboard navigation, color contrast, or wants to audit, verify, or fix accessibility on any component. Also use when writing or debugging .test.axe.ts files, checking tabindex management, or reviewing focus trapping behavior.

Component API design patterns following Elements conventions for properties, attributes, CSS custom properties, slots, and events. Use this skill whenever the user is designing or deciding on a component API, choosing between properties vs attributes vs slots, naming CSS custom properties, designing events, avoiding impossible states, deciding whether to reflect to an attribute, using CSS Parts, implementing the DataElement interface, or applying the internal-host pattern. Also trigger when the user asks about shorthand vs granular CSS properties or event naming conventions.

Wireit orchestration, build optimization, caching strategies, and build troubleshooting for the Elements monorepo. Use this skill whenever the user asks about Wireit configuration, build tasks, build dependencies, cache invalidation, build performance, adding new build tasks to package.json, cross-package dependency patterns, sideEffects declarations, CI pipeline configuration, or diagnosing slow or failing builds. Also trigger when the user mentions wireit, vite build issues, pnpm run ci, or build output problems.

Comprehensive code review process for Elements monorepo changes. Provides structured feedback on type safety, testing, documentation, and adherence to project guidelines. Use this skill whenever the user asks you to review code, check staged changes, look at a diff, give feedback before committing, review a PR or merge request, or evaluate code quality. Trigger on phrases like "review my changes," "check my code," "give me feedback," "look at my staged files," "review this PR," or "before I commit."

Guide for creating new Elements components with all required files, base classes, metadata patterns, and test boilerplate. Use this skill whenever the user wants to create, scaffold, or set up a new component from scratch, needs to understand the required 10-file structure, asks about base classes (LitElement vs BaseButton), define.ts vs index.ts patterns, static metadata, component registration, or sub-component parent relationships. Also trigger when the user mentions creating test boilerplate for all 5 test types.

Guidelines for writing documentation files including 11ty templates, Eleventy shortcodes, JSDoc annotations, and markdown content. Use this skill whenever the user works with documentation markdown files, Vale prose linting errors, Eleventy shortcodes (dodont, example), frontmatter, JSDoc annotations that must pass Vale, or the documentation site. Also trigger when the user mentions Vale errors, adding terms to the vocabulary, suppressing Vale rules, or writing and fixing prose in .md or .ts files.

Guidelines for writing example template files (*.examples.ts) including naming conventions, @summary JSDoc comments, and stateless HTML patterns. Use this skill whenever the user creates, modifies, or reviews *.examples.ts files, asks about example naming rules (PascalCase, 3-word max, no component prefix), needs to write or improve @summary JSDoc comments, encounters example-naming ESLint errors, or wants to add stateless or interactive examples to a component's example file.

Convert validated playground templates or HTML compositions into reusable pattern files (*.examples.ts) in the pattern library. Use this skill whenever the user wants to save, persist, store, or catalog a template, prototype, composition, or playground result as a reusable pattern. Trigger on phrases like "save as pattern," "create pattern," "add to pattern library," "persist this template," "convert to examples.ts," or when the user has validated HTML and wants it stored in projects/internals/patterns/src/.

Use this skill when the user wants to generate Kubernetes YAML manifests for NVIDIA networking deployment using k8s-launch-kit (l8k). Activate for: manifest generation, profile selection, choosing between SR-IOV/host-device/RDMA-shared/IPoIB/MacVLAN/Spectrum-X, creating deployment files, or when the user asks 'which profile should I use' or needs help choosing a network configuration.

Use this skill when the user needs help understanding, creating, or editing a k8s-launch-kit (l8k) configuration file (l8k-config.yaml or cluster-config.yaml). Activate for: config file questions, parameter tuning, subnet configuration, NV-IPAM setup, DOCA driver settings, NIC configuration operator settings, changing MTU, VFs, resource names, or understanding what any config field does.

Use this skill when the user wants to discover their Kubernetes cluster's network hardware capabilities using k8s-launch-kit (l8k). Activate for: cluster discovery, hardware detection, NIC detection, finding what GPUs or NICs are in a cluster, creating a cluster config file, or when the user says 'discover' in the context of l8k or NVIDIA networking.

Use this skill when the user wants to verify that an NVIDIA networking deployment matches the configuration that produced it. Activate for: 'is my deployment correct', 'are all the manifests applied', 'does the network operator version match', 'verify deployment', 'check cluster state against config', or any question about whether the cluster reflects what l8k generated. Wraps the `l8k validate` subcommand.

Embody a senior NVIDIA Networking Engineer who is an expert on deploying cloud-native networking on Kubernetes with k8s-launch-kit (l8k). Activate whenever the user mentions NVIDIA network profiles, SR-IOV, RDMA, Spectrum-X, BlueField, ConnectX, NIC configuration, Network Operator, DOCA drivers, multirail networking, l8k, k8s-launch-kit, or any Kubernetes networking topic involving NVIDIA hardware. Also activate when the user asks general questions about high-performance networking, GPU interconnect, or RDMA configuration.

Use this skill when the user wants to deploy generated NVIDIA networking manifests to a Kubernetes cluster using k8s-launch-kit (l8k). Activate for: applying manifests, deploying to cluster, the `l8k deploy` subcommand or the legacy --deploy flag on `l8k generate`, applying generated files, or any mention of pushing l8k output to a live cluster. Even if the user just says 'apply these' or 'push to cluster' after generating manifests, use this skill.

Use this skill when the user wants to run the full k8s-launch-kit (l8k) pipeline end-to-end: discover cluster hardware, select a profile, generate manifests, and deploy them all in one command. Also activate for CI/CD integration, automation pipelines, 'one-liner', 'complete workflow', or end-to-end NVIDIA networking deployment.

k8s-launch-kit (l8k) CLI: Shared patterns for binary location, global flags, output formatting, exit codes, and error handling. Read this before using any other k8s-launch-kit skill.

Evaluates accuracy of quantized or unquantized LLMs using NeMo Evaluator Launcher (NEL). Triggers on "evaluate model", "benchmark accuracy", "run MMLU", "evaluate quantized model", "run nel". Handles deployment, config generation, and evaluation execution. Not for quantizing models (use ptq), deploying/serving models (use deployment), or comparing completed baseline-vs-quantized results (use compare-results).

Establish baseline-vs-candidate evaluation plans, delegate missing evaluations, compare validated results, and decide quantization feasibility. Use when the user asks to compare baseline vs quantized runs, explain an accuracy drop/regression, verify whether a quantized checkpoint is acceptable, or compare NEL/MLflow evaluation outputs. Do NOT use for generic single-model evaluation without comparison intent (use evaluation), live NEL status/debugging (use launching-evals), or generic MLflow browsing without a comparison goal (use accessing-mlflow).

Serve a quantized or unquantized LLM checkpoint as an OpenAI-compatible API endpoint using vLLM, SGLang, or TRT-LLM. Use when user says "deploy model", "serve model", "start vLLM server", "launch SGLang", "TRT-LLM deploy", "AutoDeploy", "benchmark throughput", "serve checkpoint", or needs an inference endpoint from a HuggingFace or ModelOpt-quantized checkpoint. Do NOT use for quantizing models (use ptq) or evaluating accuracy (use evaluation).

Run, monitor, analyze, and debug LLM evaluations via nemo-evaluator-launcher. Covers running evaluations, checking status and live progress, debugging failed runs, exporting artifacts and logs, and analyzing results. ALWAYS triggers on mentions of running evaluations, checking progress, debugging failed evals, analyzing or analysing runs or results, run directories or artifact paths on clusters, Slurm job issues, invocation IDs, or inspecting logs (client logs, server logs, SSH to cluster, tail logs, grep logs). Do NOT use for creating or modifying evaluation configs.

Monitor submitted jobs (PTQ, evaluation, deployment) on SLURM clusters. Use when the user asks "check job status", "is my job done", "monitor my evaluation", "what's the status of the PTQ", "check on job <slurm_job_id>", or after any skill submits a long-running job. Also triggers on "nel status", "squeue", or any request to check progress of a previously submitted job.

This skill should be used when the user asks to "quantize a model", "run PTQ", "post-training quantization", "NVFP4 quantization", "FP8 quantization", "INT8 quantization", "INT4 AWQ", "quantize LLM", "quantize MoE", "quantize VLM", or needs to produce a quantized HuggingFace or TensorRT-LLM checkpoint from a pretrained model using ModelOpt.

Cherry-pick merged PRs labeled for a release branch into that branch, then open a PR and apply the cherry-pick-done label. Use when asked to "cherry-pick PRs for release/X.Y.Z", "pick PRs to release branch", or "cherry-pick labeled PRs".

Query and browse evaluation results stored in MLflow. Use when the user wants to look up runs by invocation ID, compare metrics across models, fetch artifacts (configs, logs, results), or set up the MLflow MCP server. ALWAYS triggers on mentions of MLflow, experiment results, run comparison, invocation IDs in the context of results, or MLflow MCP setup.

Append a learning to agent-learnings.md when the agent or user identifies a mistake, non-obvious pattern, or useful convention worth remembering. Use when the user says "capture this", "remember this", or "add to learnings".

Set up MSC credentials and MCP server integration for cloud storage access. Use when the user wants to interact with cloud storage (S3, GCS, Azure, OCI), asks to set up credentials, connect to a bucket, list objects, move files, or sync data. Checks for existing config first — only runs interactive setup if credentials are missing.

Fix CVE vulnerabilities by upgrading affected dependencies across Python (uv), Go, and Rust layers. Use when the user mentions CVE numbers, asks to fix security vulnerabilities, or asks to patch dependencies for CVEs.

Produce a feature specification covering API changes, backward compatibility, cross-language impact, performance, and a file-level change plan. Use when designing a new feature, planning a significant change, or the user asks to spec something out.

Produce a test plan with specific test names, assertions, and coverage across Python, Rust, and Go. Use when planning tests for a feature, after generating a spec, or the user asks for a test plan.

Stage files, commit with correct format, push, and generate an MR description. Manual invocation only — has side effects (git operations). Use when the user explicitly asks to prepare, create, or submit an MR.

Prepares and automates the release process for multi-storage-client (version bump, release notes, checklist). Use when the user wants to cut a release, bump version, prepare release, publish a new version, or automate release steps.

Before designing a feature, research how comparable libraries solve the problem, check best practices, known pitfalls, and upstream changes. Use when starting a new feature, investigating an unfamiliar area, or the user asks to research something.