Does it work with different types of Langfuse data?

Yes, it supports several retrieval modes including Input/Output, Flow for timing, Prompts for LLM quality, and Full metrics for comprehensive analysis including costs.

What specific symptoms can this skill investigate?

It is optimized to handle data gaps, execution crashes, high latency, poor LLM output quality, and unexpected API costs.

Can I analyze performance bottlenecks with this skill?

Yes, by using the 'flow' mode, the skill identifies the slowest steps in your workflow and suggests parallelization, caching strategies, or model changes to reduce latency.

Can it suggest actual code changes?

Absolutely. The skill generates structured reports that include recommended code diffs and explanations of why those changes address the identified root cause.

How does Langfuse Trace Analysis help with debugging?

It automatically retrieves execution data from Langfuse and matches it against your codebase to identify logic errors, missing data, or prompt instructions that caused a failure.

Langfuse Trace Analyzer

Name: Langfuse Trace Analyzer
Author: mberto10

bymberto10

Analytics & Monitoring

Diagnoses AI workflow issues by correlating Langfuse trace data with codebase logic to identify root causes and generate actionable fixes.

About

Langfuse Trace Analyzer is a specialized diagnostic tool for Claude Code designed to bridge the gap between LLM observability and source code implementation. It automates the process of retrieving execution traces, classifying symptoms like data gaps or high latency, and conducting deep-dive investigations into prompt quality or execution errors. By mapping trace observations directly to your project files, the skill helps developers move from identifying a failure to applying a verified code fix in a fraction of the time, producing structured reports complete with root cause analysis and suggested diffs.

Key Features

Comparative analysis patterns to identify differences between successful and failed traces
0 GitHub stars
Automated symptom classification for data gaps, latency, and quality issues
Automated generation of structured reports with specific code recommendations
Multi-mode trace retrieval including IO, Flow, Prompts, and Full metrics
Deep-dive investigation strategies for pinpointing LLM hallucinations or logic errors

Use Cases

Debugging intermittent 'wrong output' errors in complex LLM orchestration chains
Identifying and fixing performance bottlenecks in multi-step AI workflows
Investigating 'data gaps' where context or API results fail to propagate correctly through the system

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mberto10/mberto-compound trace-analysis

For use in Claude.ai and ChatGPT

Download Skill

GitHub