What problem does it solve?
Adding strong domain types to NeMo Relay tool and LLM integrations risks breaking middleware, intercepts, and guardrails that operate on raw JSON. This Skill guides you through typed wrappers and codecs that convert values at the boundary so middleware keeps seeing predictable JSON.
Core Features & Use Cases
- Typed Value Codecs: Use JsonPassthrough, DataclassCodec, PydanticCodec, BestEffortAnyCodec in Python, or custom Codec<T> implementations in Node.js to translate domain objects to and from JSON.
- Provider Codecs: Normalize LLM provider payloads with OpenAIChatCodec, OpenAIResponsesCodec, and AnthropicMessagesCodec so middleware can inspect messages, tools, usage, and finish reasons.
- Validation Checklist: Verify codec output is JSON-compatible, required fields survive round-trips, and response codec failures never break the underlying LLM call.
- Use Case: A team with stable Pydantic models wants typed tool calls in NeMo Relay without breaking existing guardrails; the Skill explains how PydanticCodec encodes to JSON before middleware runs and decodes back afterward.
Quick Start
Ask your agent to add Pydantic typed wrappers around NeMo Relay tool calls while keeping middleware operating on JSON.
