nemo-relay-get-started

por nvidia

Usa esta skill cuando los usuarios de NeMo Relay por primera vez quieran probar Relay, elegir el inicio rápido de menor complejidad compatible, o verificar el valor inicial a través de la CLI, a…

npx skills add https://github.com/nvidia/skills --skill nemo-relay-get-started

Get Started With NeMo Relay

Guide a new user to visible Relay value with the least complicated applicable trial. Do not begin with production deployment or Relay's full architecture. Keep the first run focused on one observable success path.

Choose A Try-Now Path

Evaluate these paths in order. Use the first one that fits the user's stated goal and existing environment.

  1. CLI try-now (default): choose this for a generic "try Relay" request or when the user wants value without modifying application code. Run Codex or Claude Code through the local CLI wrapper. Read CLI Try-Now.
  2. Hermes Agent native path: choose this when Hermes Agent owns the execution boundary. Explain that NeMo Relay is built in and requires no separate Relay installation, observability plugin, or Relay CLI setup. State explicitly: "Hermes Agent understands NeMo Relay plugin configurations." Stop after confirming the native integration. Do not provide installation or observability-configuration instructions, and do not continue into the generic plugin progression.
  3. Built-in integrations try-now: choose this when an existing LangChain, LangGraph, Deep Agents, or OpenClaw application owns the execution boundary. Prefer the maintained supported integration over manual wrapping. Read Built-In Integrations Try-Now.
  4. Language-specific manual try-now: choose this when the user's Python, Node.js, or Rust application directly owns its tool or LLM call sites and no maintained integration is the better boundary. Read Manual Language Try-Now.

Do not ask the user to choose among all four when their request, manifest, or framework already identifies the boundary. For an unspecified request, use the CLI path. When more than one CLI agent is available, ask one concise question to select the agent.

Resolve Installation Without Looping

Select the try-now path before choosing an install package.

  • CLI path -> verify nemo-relay --version; if missing, use nemo-relay-install for the CLI outcome.
  • Hermes Agent native path -> do not install or configure Relay separately.
  • Built-in integration path -> use nemo-relay-install for the named framework or harness package.
  • Manual language path -> use nemo-relay-install for the detected language package.

If installation already succeeded, preserve the chosen path and continue from its next step. Do not ask the install-path question again or bounce between the install and get-started skills.

Apply The Common First-Value Contract

Do not apply this section to the Hermes Agent native path.

Follow the selected reference, then:

  1. Inspect the target environment and existing Relay configuration before proposing changes. Treat repository-local .nemo-relay/config.toml and .nemo-relay/plugins.toml as unsupported project configuration. Do not create, edit, merge, or trust those files as active Relay configuration.
  2. Explain the attachment boundary and show the exact minimal change.
  3. Obtain confirmation before writing configuration, modifying application code, or launching a model-consuming run.
  4. Use a non-sensitive, read-only trial and exercise one representative tool and LLM path when the selected surface exposes both.
  5. Verify observable evidence from the selected path rather than treating a successful application result as proof that Relay is active.
  6. Summarize the captured root, tool, and model relationships without dumping prompts, credentials, or complete event payloads.

Explain only the concepts visible in the result: the chosen attachment boundary, scopes and parentage, captured lifecycle events, and how subscribers or the Observability plugin make those events inspectable. Keep instrumentation and export distinct.

Continue With One Plugin

Do not apply this section to the Hermes Agent native path.

Stop the initial try-now workflow when the selected path's success checks pass. Then make one additional built-in plugin the primary suggested next step.

Explain Relay's core progression: instrument an execution boundary once, then change or extend behavior through plugin configuration without repeatedly rewriting those call sites. Easy plugin configuration and reconfiguration is the main value to demonstrate after the first observable proof.

If the selected path did not use plugin-managed Observability, add it first to establish the reusable plugin path. If Observability already produced the proof, ask what outcome matters next and recommend exactly one plugin:

  • Adaptive -> adaptive runtime behavior and optimization
  • NeMo Guardrails -> policy checks around managed execution
  • PII Redaction -> sanitization of sensitive observability payloads
  • Model Pricing -> cost estimates for managed LLM responses

Use the plugin overview to select the next component. Preview its smallest configuration, obtain confirmation, and verify its behavior before layering in another plugin.

Use another handoff only after the user accepts or declines plugin progression, or when the demonstrated boundary does not yet cover the intended workflow:

  • Direct application expansion -> nemo-relay-instrument-calls
  • Unsupported framework or harness integration -> nemo-relay-integrate-upstream
  • Additional exporters or durable observability configuration -> nemo-relay-plugin-observability
  • A different package or supported integration -> nemo-relay-install
  • Persistent Claude Code or Codex loading -> nemo-relay-install; for Codex Desktop, complete its recovery-note safety gate before changing global config
  • Missing hooks, gateway traffic, or events -> nemo-relay doctor, nemo-relay doctor --json, or nemo-relay-debug-runtime-integration

Do not configure production OTLP backends, model pricing, guardrails, adaptive tuning, custom plugins, Go or FFI examples during the quick start. Mention optional plugins only after the initial proof and add only the one the user selects.

Public Entry Points

Use these public entry points for current product documentation:

Más skills de nvidia

compileiq-debug
nvidia
Úsalo cuando algo esté mal: Search() se cuelga, todas las evaluaciones devuelven INVALID_SCORE, las puntuaciones no mejoran, cada configuración devuelve el mismo número, errores de ptxas…
create-github-pr
nvidia
Crear solicitudes de extracción de GitHub usando la CLI gh. Usar cuando el usuario quiera crear un nuevo PR, enviar código para revisión o abrir una solicitud de extracción. Palabras clave de activación -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
Escanea otros issues abiertos para encontrar aquellos que un PR dado también podría corregir o romper accidentalmente. Genera oportunidades de corrección adyacente y riesgos de contradicción con archivo:línea…
fhir-basics
nvidia
Enseña a los agentes cómo funcionan las APIs de FHIR R4, qué recursos están disponibles, cómo consultarlos con parámetros de búsqueda y cómo analizar correctamente todos los formatos de respuesta…
compileiq-validate-result
nvidia
Usar DESPUÉS de que una Búsqueda haya finalizado y ANTES de reclamar cualquier aceleración o enviar un ACF. Carga el CSV de dump_results, extrae los mejores K candidatos (de un solo objetivo)…
changelog-audit
nvidia
Auditar el CHANGELOG.md de Warp antes de un lanzamiento: recuperar entradas perdidas, ordenar por impacto en el usuario, refinar el lenguaje de las entradas, ajustar saltos de línea y (en modo rama de lanzamiento) incrementar comparación…
maintain-dynamic-plugins
nvidia
Mantener los cargadores de plugins dinámicos de NeMo Relay, manifiestos, SDKs nativos de Rust, protocolo de trabajador gRPC, SDK de trabajador Python, documentación, pruebas y cobertura del flujo de trabajo de lanzamiento
dgx-diagnose
nvidia
Diagnostica problemas comunes de la DGX Station GB300: fallos de CUDA, direccionamiento incorrecto de GPU, errores de contenedores vLLM/SGLang, problemas de estado MIG, errores de NVLink/Fabric Manager,…