nemo-fabric-build-adapter

Build and validate third-party NVIDIA NeMo Fabric adapters against the public adapter contract.

3.2k|370|Updated Feb 25, 2026
One-click install
npx skills add https://github.com/NVIDIA/skills --skill nemo-fabric-build-adapter
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: nemo-fabric-build-adapter
Source: https://github.com/NVIDIA/skills/tree/main/skills/nemo-fabric-build-adapter
Command: npx skills add https://github.com/NVIDIA/skills --skill nemo-fabric-build-adapter

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Building a third-party adapter for NVIDIA NeMo Fabric requires correctly implementing the southbound adapter contract: descriptors, configuration mapping, lifecycle methods, discovery packaging, and conformance checks. This Skill guides an agent through that contract so adapters are structured correctly and avoid unsupported fields or hidden behavior.

Core Features & Use Cases

  • Descriptor Authoring: Create self-contained *.fabric-adapter.json and per-target *.fabric-target.json descriptors with aligned contract versions, closed schemas, and declared capabilities.
  • Lifecycle Implementation: Implement start, ordered invoke, and stop methods with isolated runtime state, plus optional native OpenAI streaming via invoke_openai_stream.
  • Validation & Handoff: Run discovery, doctor, failure, cleanup, and two-runtime isolation checks before shipping an adapter.
  • Use Case: A developer building a Python adapter for a custom agent harness uses this Skill to define the adapter descriptor, map a validated AgentConfig into target-native constructs, package discovery metadata under share/nemo-fabric, and verify conformance before release.

Quick Start

Use the nemo-fabric-build-adapter skill to build a typed NVIDIA NeMo Fabric adapter for my agent harness.

Frequently Asked Questions about nemo-fabric-build-adapter

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build a third-party NVIDIA NeMo Fabric adapter?

Start by reading the published adapter contract, then define a self-contained *.fabric-adapter.json descriptor before writing code. Map a validated AgentConfig at the adapter boundary, implement start, invoke, and stop lifecycle methods, and install discovery metadata under share/nemo-fabric.

How do I add native OpenAI streaming to a NeMo Fabric adapter?

Set capabilities.streaming in the descriptor and implement async invoke_openai_stream(request, context, emit). Invoke the target exactly once, await emit only for valid openai.chat_completions.chunk/v1 chunks, and return one separate terminal AgentRunResult.

When should I use a shared framework adapter versus a dedicated custom-agent adapter?

Use a shared framework adapter with registered targets when the framework provides stable loading and invocation semantics across multiple agents. Use a dedicated adapter when the agent itself is the only unambiguous execution boundary.

Does this skill cover consumer applications that call the NeMo Fabric SDK?

No. Consumer-side SDK integration, such as selecting a harness and invoking it from an application, is out of scope and should use the nemo-fabric-integrate skill instead. This skill is only for authoring third-party adapters.

What validation is required before handing off a NeMo Fabric adapter?

Install the built wheel in an isolated environment, confirm descriptor discovery and resolution in Fabric().plan(...), exercise accepted and rejected configs, run doctor checks, and test lifecycle failures, cleanup, streaming edge cases, and two-runtime isolation.