langfuse-traces - SKILL.md Agent Skill

name: langfuse-traces description: Query Langfuse traces for debugging LLM calls, analyzing token usage, and investigating workflow executions. Use when debugging AI/LLM behavior, checking trace data, or analyzing observability metrics. allowed-tools: Bash, Read

Langfuse Traces Skill

Query and analyze Langfuse trace data directly from Claude Code.

Usage

Invoke the query script using the base directory shown above:

bash <base_directory>/scripts/query.sh <command> [options]

Commands

Command	Args	Description
`traces`	[limit] [session_id] [name]	List recent traces
`trace`	trace_id	Get full trace with observations
`observations`	[limit] [trace_id]	List spans/generations
`sessions`	[limit]	List sessions
`summary`	[limit]	Compact one-line-per-trace view

Examples

# List last 20 traces
bash <base_directory>/scripts/query.sh traces 20

# Get specific trace detail
bash <base_directory>/scripts/query.sh trace tr-abc123

# List observations for a trace
bash <base_directory>/scripts/query.sh observations 50 tr-abc123

# Quick summary of recent activity
bash <base_directory>/scripts/query.sh summary 10

Credentials

The script reads credentials from .env.local, .env, or environment variables:

LANGFUSE_PUBLIC_KEY=pk-lf-...
LANGFUSE_SECRET_KEY=sk-lf-...
LANGFUSE_BASE_URL=https://cloud.langfuse.com  # optional, default

Requirements

curl (standard on macOS/Linux)
jq for JSON parsing

Output

All commands return JSON (piped through jq). Use jq filters for specific fields:

bash <base_directory>/scripts/query.sh traces 5 | jq '.data[].name'