instrument-existing-agent-with-prefactor-sdk

LLM tracing is added by instrumenting boundaries with Prefactor, wrapping run, LLM, and tool calls in spans while preserving the existing business logic and context propagation.

Yes, agents built with common adapters like LangChain, AI SDK, or OpenClaw are supported. The instrumentation enforces adapter-specific imports to add tracing around boundaries safely.

Error capture in agent tracing works by recording error metadata within spans and rethrowing the original errors. Proper finish paths ensure both success and error cases are traced safely.

You add a top-level run span with child spans around LLM and tool boundaries. This approach targets boundaries specifically, avoiding modifications to the underlying business logic.

No, adding Prefactor tracing spans preserves context propagation. The instrumentation enforces proper finish paths for success and error cases, ensuring safe production usage without altering behavior.

You need to bootstrap Prefactor resources and install dependencies before wrapping agent run, LLM, and tool boundaries with the appropriate adapter imports.