NeMo Relay Quick Start
Helps first-time NeMo Relay users prove observable execution value through the smallest suitable trial.
The skill explicitly requires least privilege, project-local scope, a read-only trial, credential non-disclosure, confirmation before configuration changes or model calls, and a recovery-file gate for persistent Codex Desktop changes. It does not fully document dependency and external-service data flows, configuration rollback, or default protection for sensitive telemetry payloads, so 4 points are deducted.
The routing, main workflow, diagnostics, and failure-boundary guidance are internally consistent, including configuration precedence, missing-output checks, and session-finalization timing. However, command availability is not statically proven, there is no committed test suite covering the key paths, and abnormal-input feedback remains limited; 2 points are deducted and the static ceiling applies.
The audience, CLI/integration/manual triggers, non-fit boundaries, and handoff paths are comparatively clear. Chinese-language interaction support and mainland-China reachability for NVIDIA documentation, package sources, or services are not addressed, and complex-environment boundaries have limited evidence, so 4 points are deducted.
The material has layered documentation, reference files, dependency notes, command examples, limitations, version information, and clear Apache licensing and NVIDIA attribution. The static report nevertheless identifies missing recommended Instructions and Examples sections and an invalid author format; changelog, maintenance ownership, and update procedure are also unclear, so 5 points are deducted.
The skill targets one minimal observable Relay trial and defines success through ATOF/ATIF output, tool and LLM lifecycle evidence, and parent-child relationships. The committed evaluation report shows positive task results, but this static review cannot verify directly usable outputs or comparative benefit, so the score is capped at 7 and 8 points are deducted.
The fixed revision, evaluation tasks, benchmark report, and auditable source files provide some traceability and fact/inference separation. There are no independently reproducible execution artifacts, real CI coverage for this skill, or corroborating sources available in the supplied material, so 6 points are deducted and the static ceiling applies.
- No commands or tests were executed; validate the CLI, package versions, configuration precedence, and output-directory permissions in an isolated, non-sensitive project before first use.
- The skill depends on external NVIDIA documentation, tooling, and potentially authenticated or networked services; mainland-China users should verify reachability and compliance in advance.
- ATOF/ATIF may contain prompts, tool arguments, or model-related data; verify redaction, access control, retention, and rollback requirements in the target project.
- Persistent Codex Desktop configuration must follow the recovery-file and confirmation gate; a temporary trial should not be promoted directly to global configuration.
What does this skill do, and when should you use it?
This skill guides first-time NeMo Relay users toward the least-complex applicable quick start. It supports a local CLI path, maintained integrations for selected frameworks, and manual instrumentation for Python, Node.js, or Rust. The workflow stays focused on one observable success path rather than production deployment or the full Relay architecture. After the initial proof, it recommends configuring one additional plugin.
Selects a CLI, built-in integration, or language-specific manual trial based on the user’s goal and environment; checks the nemo-relay version and chooses the corresponding installation outcome; reads the relevant trial guide; inspects the target environment and existing Relay configuration; explains the attachment boundary and minimal change; requests confirmation before changing configuration or application code or launching a model-consuming run; performs a non-sensitive trial and verifies observable Relay evidence; summarizes root, tool, and model relationships; and previews one plugin configuration after the initial proof.
- A first-time Relay user who wants to try Relay without modifying application code can use the default CLI path.
- A team with an existing LangChain, LangGraph, Deep Agents, or OpenClaw application can use the maintained integration at its execution boundary.
- A developer whose Python, Node.js, or Rust application directly owns tool or LLM call sites can use manual instrumentation when no maintained integration is a better fit.
- A user whose installation has already succeeded can continue from the selected trial path without repeating installation decisions.
- A user who has completed the first proof can evaluate one next plugin such as Observability, Adaptive, NeMo Guardrails, PII Redaction, or Model Pricing.
What are this skill's strengths and limitations?
- Uses an ordered decision process across CLI, maintained integrations, and manual instrumentation.
- Requires confirmation before modifying configuration, application code, or launching a model-consuming run.
- Verifies Relay lifecycle evidence instead of treating a successful application result as proof of instrumentation.
- Demonstrates a reusable extension path through one plugin after the initial proof.
- The skill references three trial guides whose contents are not included, so their exact commands cannot be verified here.
- No complete installation-package version matrix, platform matrix, or test results are provided.
- The quick-start scope excludes production OTLP backends, full production setup, Go examples, and FFI examples.
- The supplied source does not specify cost, resource consumption, or a detailed permission model.
How do you install this skill?
Install the skill with the skills CLI:
npx skills add nvidia/skills --skill nemo-relay-get-started --yes
You can target a specific agent, for example:
npx skills add nvidia/skills --skill nemo-relay-get-started --agent codex
The README does not document another dedicated installation method for this individual skill.
How do you use this skill?
Use a prompt such as: “I’m a first-time NeMo Relay user. Choose the simplest trial for my environment and verify Relay with one non-sensitive example.” For an unspecified request, the skill defaults to the CLI path; when multiple CLI agents are available, it asks which agent to use. After the initial proof, accept or decline one plugin recommendation before taking another handoff.
How does this skill compare with similar options?
The skill explicitly chooses among three real trial approaches: the default CLI path for a minimal no-code trial, maintained integrations for LangChain, LangGraph, Deep Agents, or OpenClaw, and manual Python, Node.js, or Rust instrumentation when the application directly owns the call sites.