nemo-relay-debug-runti…
Use this skill when NeMo Relay is installed or imported but application-side runtime behavior is missing or incorrect, including…
Use this skill when choosing or running NeMo Relay installation for the CLI, Python, Node.js, Rust, OpenClaw, or maintained framework integrations, or when explaining Hermes Agent's built-in Relay integration.
$ npx -y skills add NVIDIA/NeMo-Relay --skill nemo-relay-install --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/nemo-relay-installContext preview
The summary Claude sees to decide when to auto-load this skill.
Use this skill when choosing or running NeMo Relay installation for the CLI, Python, Node.js, Rust, OpenClaw, or maintained framework integrations, or when explaining Hermes Agent's built-in Relay integration.
name: nemo-relay-install description: Use this skill when choosing or running NeMo Relay installation for the CLI, Python, Node.js, Rust, OpenClaw, or maintained framework integrations, or when explaining Hermes Agent's built-in Relay integration. license: Apache-2.0 metadata: author: NVIDIA Corporation and Affiliates
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.
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 built-in Hermes Agent support?
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](references/cli-install.md). 2. **Maintained integration** when the user already uses OpenClaw, Hermes, LangChain, LangGraph, or Deep Agents. Read [Maintained Integration Installation](references/maintained-integrations.md). 3. **Language package** when a Python, Node.js, or Rust application directly owns its tool or model call sites. Read [Language Package Installation](references/language-packages.md). 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.
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](references/cli-install.md). 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.
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.
Know about `nemo-relay doctor`, but use it in the right scope:
gateway, agent-readiness, plugin, exporter, or model-pricing issues.
inspect checks programmatically.
codex`, or `nemo-relay doctor --plugin all` only for persistent host-plugin installations.
not require persistent host-plugin state.
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.
Choose the next workflow from the user's immediate outcome:
or trial plugin setup.
host-plugin workflow, not application runtime setup.
plugin files, observability, or adaptive behavior.
Avoid these installation-scope mistakes:
package. -
NVIDIA NeMo Relay provides visibility into and control over agent runs without requiring changes to the existing agent stack.
Use this skill when NeMo Relay is installed or imported but application-side runtime behavior is missing or incorrect, including…
Use this skill when first-time NeMo Relay users want to try Relay, choose the least-complex supported quick start, or verify…
Use this skill when an application owns tool or LLM/provider call sites and needs to wrap them with NeMo Relay scopes and managed…
Use this skill when concurrent requests, async tasks, threads, workers, goroutines, or agents need independent NeMo Relay scope…
Use this skill when adding NeMo Relay typed wrappers, domain types, or provider codecs while preserving JSON middleware semantics…
Use this skill when assessing, extending, or implementing NeMo Relay support in an agent harness or agent framework, including…