analyzing-mlflow-trace

Analyze MLflow trace JSON files to identify errors and performance bottlenecks.

Updated Mar 25, 2026
One-click install
npx skills add https://github.com/JingyiWng/databricks_ai_dev_kit_price_watcher --skill analyzing-mlflow-trace-jingyiwng
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: analyzing-mlflow-trace
Source: https://github.com/JingyiWng/databricks_ai_dev_kit_price_watcher/tree/main/.claude/skills/analyze-mlflow-trace
Command: npx skills add https://github.com/JingyiWng/databricks_ai_dev_kit_price_watcher --skill analyzing-mlflow-trace-jingyiwng

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It helps users debug and investigate single MLflow traces by extracting key information, identifying errors, performance bottlenecks, and interpreting assessment feedback.

Core Features & Use Cases

  • Full trace retrieval: Guides users to fetch the complete JSON trace via the MLflow CLI.
  • Structured health check: Provides ready-made jq commands to summarise state, span count, error spans, and assessment issues.
  • Insight extraction: Interprets assessments, span inputs/outputs, and timing to pinpoint root causes.
  • Code correlation: Maps spans to source‑code functions for deeper root‑cause analysis.
  • Typical scenarios: Debugging failed agent runs, investigating unexpected LLM outputs, and monitoring token usage for cost optimisation.

Quick Start

Analyze the MLflow trace with ID 12345 to get a detailed error and performance report.

Frequently Asked Questions about analyzing-mlflow-trace

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug a failed MLflow trace to find errors and performance bottlenecks?

To debug a failed MLflow trace, fetch the complete JSON via the MLflow CLI, parse span inputs and outputs, and interpret assessment fields to pinpoint root causes. This identifies errors, performance bottlenecks, and assessment issues within a single trace.

What is the best way to extract insights from an MLflow trace JSON file?

Extracting insights from an MLflow trace JSON involves using ready-made jq commands to summarize state, span count, and error spans, then interpreting timing and assessment feedback to investigate failed agent runs or unexpected LLM outputs.

Do I need the MLflow CLI to analyze trace data for token usage and cost optimization?

Yes, you need the MLflow CLI to fetch the complete trace JSON file and the ability to parse JSON for span and assessment fields. This access enables monitoring token usage for cost optimization and correlating trace data with code behavior.

How does mapping spans to source-code functions help with root-cause analysis in MLflow?

Mapping spans to source-code functions provides deeper root-cause analysis by correlating MLflow trace data with code behavior. It connects the trace JSON span inputs and outputs directly to application functions for debugging AI/ML applications.

Can I use jq commands to summarize state and error spans in an MLflow trace?

Yes, you can use provided jq commands to perform a structured health check on an MLflow trace. These commands summarize the trace state, span count, error spans, and assessment issues directly from the fetched JSON file.