nemo-relay-install

作成者: nvidia

NeMo RelayのインストールをCLI、Python、Node.js、Rust、OpenClaw、Hermes、またはメンテナンスされたフレームワーク統合のために選択または実行する際に、このスキルを使用してください…

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

Install NeMo Relay

Choose the package or executable from the user's desired outcome. Stop after installation and basic availability checks. Do not configure runtime behavior, write plugins.toml, create scopes, register middleware, or build a first app example from this skill. Keep installation separate from first-use instrumentation.

Choose The Install Path

If the user asks to install NeMo Relay but does not identify a target or desired outcome, ask one short clarifying question before giving commands:

Which install path do you want: CLI for coding-agent/local gateway use, language package for a Python/Node.js/Rust app, or framework integration for LangChain, LangGraph, Deep Agents, OpenClaw, or Hermes?

Do not ask when the user already names a CLI, language, framework, harness, source checkout, or target project file such as pyproject.toml, package.json, or Cargo.toml.

Use this order from least to most application-specific:

  1. CLI for a generic "try Relay" request, a temporary coding-agent run, the local gateway, or explicit persistent host-plugin setup. Read CLI Installation.
  2. Maintained integration when the user already uses OpenClaw, Hermes, LangChain, LangGraph, or Deep Agents. Read Maintained Integration Installation.
  3. Language package when a Python, Node.js, or Rust application directly owns its tool or model call sites. Read Language Package Installation.
  4. Source checkout only for contributors or unpublished changes. Follow the repository development guide instead of published package commands.

For "try Relay" requests, default to the CLI and temporary transparent run. Do not make persistent host-plugin installation the default.

Protect Codex Desktop Continuity

Before persistent Codex installation, determine whether the user is operating from Codex Desktop. Persistent nemo-relay install codex changes the active Codex provider and can make the current or older Desktop threads appear missing after restart because of an upstream provider-filtering bug. The threads are not deleted.

For Codex Desktop users:

  1. Recommend temporary transparent run first.
  2. If the user still wants persistent installation, read the Codex Desktop section in CLI Installation.
  3. Preview the install and proposed recovery-note location.
  4. Obtain confirmation before writing either the recovery note or global Codex configuration.
  5. Render assets/codex-desktop-recovery.md as NEMO_RELAY_CODEX_DESKTOP_RECOVERY.md in the user's workspace root before running the persistent installer.
  6. Do not restart Codex Desktop until the user has the recovery-file path.

Do not directly inspect, copy, delete, edit, or rewrite Codex session files, private application configuration, or SQLite state to work around the visibility bug. Supported nemo-relay install, uninstall, and doctor commands may manage the Relay-generated provider and hook configuration.

Install And Verify

  1. Inspect the target manifest, environment, operating system, architecture, and existing installation before changing anything.
  2. Load only the reference for the selected path.
  3. Preserve the project's existing package manager and virtual environment.
  4. Use the latest compatible release unless the user or project requires an exact version.
  5. Show the exact install command before running a remote installer or changing a project manifest.
  6. Run only the selected path's basic availability check.
  7. Report what was installed, where it was installed, and the verification result. Then stop.

The primary documented language paths are Rust, Python, and Node.js. Treat Go and raw FFI as source-first advanced surfaces, not normal package installs. Do not treat a first scope, subscriber, gateway, plugin config, or LLM call as installation verification.

Use Doctor For CLI Readiness And Configuration Issues

Know about nemo-relay doctor, but use it in the right scope:

  • Run nemo-relay doctor when the user installed the CLI and reports config, gateway, agent-readiness, plugin, exporter, or model-pricing issues.
  • Use nemo-relay doctor --json when structured output will help an agent inspect checks programmatically.
  • Use nemo-relay doctor --plugin claude-code, nemo-relay doctor --plugin codex, or nemo-relay doctor --plugin all only for persistent host-plugin installations.
  • Do not require plugin doctor for transparent runs. Transparent-run setup does not require persistent host-plugin state.
  • Do not use doctor as proof that a Python, Node.js, or Rust package dependency was installed. Verify those with the language package manager/import checks.

When doctor reports failures, summarize the failed checks and the specific remediation it suggests. Do not loop back to reinstalling every package unless the failed check points to a broken or missing install.

Hand Off After Install

Choose the next workflow from the user's immediate outcome:

  • Use nemo-relay-get-started for a first working scope, tool call, LLM call, or trial plugin setup.
  • Use the NeMo Relay CLI documentation or host-specific setup for a local CLI host-plugin workflow, not application runtime setup.
  • Use the matching plugin or instrumentation skill for runtime configuration, plugin files, observability, or adaptive behavior.

Common Mistakes

Avoid these installation-scope mistakes:

  • Using repository development setup when the user only needs a published package.
  • Installing the CLI when the user needs an application binding, or installing a binding when the user only needs the local nemo-relay executable.
  • Treating persistent host-plugin installation as the default way to try Relay, instead of starting with temporary transparent run.
  • Pinning old versions unless the user or project explicitly requires that version.
  • Continuing into plugins.toml, middleware registration, scopes, or quick-start examples before the install step has been verified.

Public Docs To Reference

Use these public entry points to confirm current installation guidance:

nvidiaのその他のスキル

compileiq-debug
nvidia
何かがおかしいときに使用:Search()がハングする、すべての評価がINVALID_SCOREを返す、スコアが改善しない、すべての設定が同じ数値を返す、ptxasエラー…
create-github-pr
nvidia
gh CLIを使用してGitHubのプルリクエストを作成します。ユーザーが新しいPRを作成したい、コードをレビューに提出したい、またはプルリクエストを開きたい場合に使用します。トリガーキーワード -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
他のオープンなIssueをスキャンし、特定のPRが修正する可能性があるものや、誤って壊す可能性があるものを見つけます。隣接修正の機会や矛盾リスクをfile:line…と共に出力します。
fhir-basics
nvidia
エージェントにFHIR R4 APIの動作方法、利用可能なリソース、検索パラメータを使ったクエリ方法、およびすべてのレスポンス形式を正しく解析する方法を教えます…
compileiq-validate-result
nvidia
検索が完了した後、かつスピードアップの申請やACFの発送の前に使用します。dump_results CSVを読み込み、トップK候補(単一目的)を抽出します…
changelog-audit
nvidia
リリース前にWarp CHANGELOG.mdを監査:失われたエントリを復元、ユーザー影響で並べ替え、エントリの文言を洗練、行折り返し、および(リリースブランチモードで)比較をバンプ…
maintain-dynamic-plugins
nvidia
NeMo Relayの動的プラグインローダー、マニフェスト、RustネイティブSDK、gRPCワーカープロトコル、PythonワーカーSDK、ドキュメント、テスト、およびリリースワークフローのカバレッジを維持する
dgx-diagnose
nvidia
一般的なDGX Station GB300の問題(CUDAクラッシュ、誤ったGPUターゲット、vLLM/SGLangコンテナのバグ、MIG状態の問題、NVLink/Fabric Managerエラーなど)を診断します。