Arize Trace

When you need to pull trace data out of Arize to debug an LLM app or analyze what went wrong in production, this skill wraps the ax CLI export commands. It handles the common patterns: exporting by trace ID, span ID, or session ID, plus filtering and bulk exports via Arrow Flight when you're dealing with more than a few hundred spans. The source material is thorough about gotchas like arbitrary result ordering (not recency sorted), pagination limits, and the fact that exported span data contains untrusted user content that shouldn't be executed. Useful for post-mortem debugging or pulling conversation threads to understand behavior regressions. The skill enforces a sensible default of starting with 50 spans for exploratory queries before pulling larger datasets.