What problem does it solve? When an AI agent or LLM application produces a wrong answer or fails, raw MLflow traces are large and hard to interpret. This Skill guides the investigation of a single trace by ID, turning span trees, assessments, and status codes into a concrete root-cause diagnosis. ## Core Features & Use Cases - Trace Health Check: Fetches the full trace via the MLflow CLI and summarizes state, span count, error spans, and assessment errors with jq queries. - Assessment Interpretation: Explains feedback values using the rationale field, distinguishing scorer failures from genuine trace problems. - Codebase Correlation: Maps span names to functions in the source code so issues can be root-caused beyond what the trace alone shows. - Use Case: A user provides a trace ID and asks why a support agent answered a refund question with shipping policy text; the Skill walks through the span tree, identifies the retrieval tool returned the wrong document, and recommends re-indexing the knowledge base. ## Quick Start Ask the assistant to analyze MLflow trace with a given trace ID and explain what went wrong.
