Skip to content
Agent Orchestration
Skill

/nemo-relay-debug-runtime-integration

Use this skill when NeMo Relay is installed or imported but application-side runtime behavior is missing or incorrect, including load failures, inactive scopes, missing events, and plugin or adaptive wiring problems.

BOOST
From plugin
nemo-relay
20111 skills1 MCP
Install
$ npx -y skills add NVIDIA/NeMo-Relay --skill nemo-relay-debug-runtime-integration --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/nemo-relay-debug-runtime-integration

Context preview

The summary Claude sees to decide when to auto-load this skill.

Use this skill when NeMo Relay is installed or imported but application-side runtime behavior is missing or incorrect, including load failures, inactive scopes, missing events, and plugin or adaptive wiring problems.

SKILL.md

nemo-relay-debug-runtime-integration.SKILL.md
name: nemo-relay-debug-runtime-integration
description: Use this skill when NeMo Relay is installed or imported but application-side runtime behavior is missing or incorrect, including load failures, inactive scopes, missing events, and plugin or adaptive wiring problems.
license: Apache-2.0
metadata:
  author: NVIDIA Corporation and Affiliates

Debug Runtime Integration

Use this skill when NeMo Relay is present in the application but something is not working. Start by proving which runtime layer is failing before changing configuration.

First Checks

  • Can the binding or native artifact load?
  • Is there an active scope when the failing call runs?
  • Is the work happening on the expected scope stack?
  • Is the subscriber/exporter/plugin configuration actually active?
  • Did the app choose the right public API layer: managed execute vs manual

lifecycle vs typed wrappers vs adaptive/plugins?

Common Failure Classes

  • Python native extension missing
  • Go dynamic library not on the loader path
  • Node native addon not built or not loading
  • Execute call outside a scope
  • Missing events because registration never happened
  • Concurrency causing the wrong scope stack to be active
  • Adaptive component never initialized or config validation ignored

Embedded Troubleshooting Matrix

  • **Rust build failure**: run the narrowest core build first, then expand to the

affected binding or workspace command.

  • **Python import failure**: rebuild the virtual environment and native

extension with `uv sync`, then run a small Python test or import check from the same environment as the application.

  • **Node.js addon failure**: reinstall and rebuild the native addon from the

Node binding package before debugging application code.

  • **Go loader failure**: build the release FFI shared library and point both the

linker and runtime loader at the release output directory; use the macOS dynamic-library path variable (`DYLD_LIBRARY_PATH`) on macOS.

  • **Scope stack empty**: the work is outside an active scope or crossed a thread,

task, goroutine, or worker boundary without the intended stack.

  • **Work leaks across requests**: separate requests are sharing one scope stack;

create a fresh stack per independent request or agent.

  • **Middleware missing or ordered incorrectly**: check global vs scope-local

registration, active scope ancestry, names, and priority values.

  • **Subscriber missing events**: register before the events are emitted; for

scope-local subscribers, ensure the current scope is the owner or descendant.

  • **Event fields missing**: managed helpers populate semantic fields; manual

lifecycle calls require explicit params for input, output, model names, and tool call IDs.

  • **ATIF empty or mixed**: register before work starts, use one exporter per run

or clear between runs, and separate concurrent agents by root scope.

  • **Provider payload conversion failure**: convert non-JSON provider objects,

SDK handles, callbacks, streams, or class instances with explicit codecs.

  • **Plugin validation failure**: validate config independently from runtime

registration and check required fields, value types, defaults, and config source.

  • **Adaptive behavior unchanged**: confirm instrumentation emits events, the

adaptive component is enabled, policy allows the behavior, and the call path reaches the configured component.

  • **OpenTelemetry or OpenInference export failure**: first identify the Relay

version. For 0.6, check the separate exporter, `http_binary` versus `grpc`, endpoint, headers, target support, and an active Tokio runtime for native gRPC. For 0.7, check the typed projection, endpoint, headers, TLS, and target support; the subscriber owns the native gRPC runtime.

  • **Callback succeeded but no lifecycle events appear**: confirm the integration

uses managed execute helpers or balanced manual start/end APIs, not only the underlying business callback.

Related Skills

  • `nemo-relay-get-started`
  • `nemo-relay-instrument-context-isolation`
  • `nemo-relay-plugin-adaptive-tuning`
  • `nemo-relay-plugin-build`
Read more
Ships withnemo-relay

NVIDIA NeMo Relay provides visibility into and control over agent runs without requiring changes to the existing agent stack.

Get the whole plugin

Other skills on nemo-relay.