What problem does it solve? Debugging an LLM application in production is hard when you cannot see what prompts were sent, which spans failed, or why latency spiked. This Skill lets you pull existing trace and span data out of Arize with the ax CLI so you can inspect errors, sessions, and model behavior directly. ## Core Features & Use Cases - Trace and Span Export: Download spans by trace ID, span ID, or session ID, or export complete traces with all child spans using ax spans export and ax traces export. - SQL-like Filtering: Query spans with filters on status codes, latency, model names, error types, and OpenInference span kinds to isolate exactly the data you need. - Root-Cause Investigation: Follow guided workflows to debug failing traces, download multi-turn conversation sessions, and analyze error messages and tracebacks. - Use Case: Your LLM agent started returning wrong answers yesterday. Export the last day's traces with --start-time, filter for status_code = 'ERROR', and inspect attributes.error.message and the LLM input messages to find the broken prompt. ## Quick Start Ask the assistant to export the recent traces from your Arize project and show which spans have errors.
