Why use Langfuse instead of LangSmith with Claude Code?

Langfuse is open-source and self-hostable, offering superior data privacy and cost control while providing built-in prompt management and tracing capabilities that are production-ready.

What is the @observe decorator used for?

The @observe decorator is a Python utility that automatically instruments your functions, sending execution data and metadata to Langfuse without requiring manual span creation.

Is this skill compatible with multi-agent systems like LangGraph?

Yes, it specifically supports distributed tracing with nested spans, making it ideal for visualizing and debugging the execution flow of complex multi-agent architectures.

How does this skill help with AI cost management?

It provides automatic token counting and cost calculation per trace, allowing you to monitor spending trends by user, session, or specific AI agent through SQL-ready dashboards.

Langfuse LLM Observability

Name: Langfuse LLM Observability
Author: yonatangross

byyonatangross

•

分析与监控

Implements professional-grade LLM observability using Langfuse for distributed tracing, cost monitoring, prompt management, and quality evaluation.

The Langfuse Observability skill integrates the open-source Langfuse platform into Claude Code workflows to provide deep insights into LLM application performance. It enables developers to implement distributed tracing with parent-child spans, track token usage and associated costs in real-time, and manage production prompts through a centralized registry. By providing tools for automated G-Eval quality assessments and multi-judge evaluations, this skill helps developers move from prototype to production with confidence while maintaining full control over their observability data through self-hosting options.

主要功能

0169 GitHub stars

02Real-time token usage and granular LLM cost tracking

03Distributed tracing with automatic @observe instrumentation

04Session and user tracking for multi-turn conversation analytics

05Automated LLM evaluation using G-Eval and multi-judge scoring

06Centralized prompt management with versioning and A/B testing

使用场景

01Monitoring and optimizing production LLM costs across different models and users

02Debugging complex multi-agent LLM workflows to identify latency or logic errors

03Managing prompt iterations and regression testing through a centralized registry

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit langfuse-observability

For use in Claude.ai and ChatGPT

Download Skill