Diagnoses and resolves LLM workflow issues by performing structured root-cause analysis on Langfuse traces.
Langfuse Trace Analysis is a specialized skill for Claude Code designed to streamline the debugging and optimization of LLM-powered applications. By integrating Langfuse observability data directly with codebase investigation, it allows developers to quickly classify symptoms like latency, data gaps, or quality issues and map them to specific code segments or prompts. The skill automates the retrieval of execution flows and I/O data, providing a structured framework to identify why a workflow failed and generating concise reports with actionable fixes to bridge the gap between production logs and source code.
Key Features
01Multi-mode trace retrieval for I/O, execution flow, and prompt inspection
02Automated report generation with evidence-based fix recommendations
03Automated symptom classification for data gaps, latency, and quality issues
04Pattern-based investigation heuristics for common LLM failure modes
050 GitHub stars
06Structured root-cause analysis linking trace evidence to specific code files
Use Cases
01Identifying latency bottlenecks in complex multi-step chains or RAG pipelines
02Analyzing prompt quality and input/output mismatches at failing steps
03Debugging why an LLM agent failed to call a specific tool or API
