What environment variables are required for setup?

You need to set LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY. You can also optionally set LANGFUSE_HOST if you are using a self-hosted instance.

How does this skill help with LLM context limits?

It uses surgical output modes like 'io' or 'prompts' to filter out metrics bloat, ensuring only the most relevant trace data is added to your conversation without overwhelming the context window.

Can I filter traces by specific metadata?

Yes, you can filter by any custom metadata field (such as project_id or user_id) or tags to isolate specific workflows or deployment environments.

Does this support score-based filtering?

Absolutely. You can retrieve traces based on minimum or maximum quality scores to find failing examples or identify 'golden' traces for your evaluation datasets.

Langfuse Data Retrieval

Name: Langfuse Data Retrieval
Author: mberto10

bymberto10

Analytics & Monitoring

Extracts and filters Langfuse observability traces to provide surgical debugging insights directly within the Claude Code environment.

About

Langfuse Data Retrieval is a specialized Claude Code skill designed for the surgical extraction of LLM observability data. It allows developers to pull specific traces, recent logs, or filtered activity directly into their terminal, formatted perfectly for LLM analysis. By offering specialized output modes like 'io' for core logic and 'prompts' for LLM interaction analysis, it prevents context window bloat while providing the precise data needed to debug complex AI workflows, investigate tool calls, and optimize prompt performance.

Key Features

Five specialized output modes (io, minimal, prompts, flow, full) to minimize context usage
Surgical trace extraction by ID, tags, or metadata filters
0 GitHub stars
Score-based filtering to identify high-quality or failing traces for dataset curation
Markdown-formatted output optimized for Claude and LLM consumption
Flexible time-range retrieval to investigate specific windows of application activity

Use Cases

Comparing prompt performance across different versions using the specialized prompts mode
Debugging failing LLM chains by inspecting input/output flows and tool call results
Identifying regression test candidates by filtering for low-scoring production traces

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mberto10/mberto-compound data-retrieval

For use in Claude.ai and ChatGPT

Download Skill

GitHub