What is the difference between Phoenix and Arize AX?

Phoenix is a self-hosted, local-first observability tool for rapid development, while Arize AX is a cloud-based or on-prem enterprise platform for long-term LLM monitoring and evaluation.

Do I need to install additional dependencies?

For Arize AX (cloud), you must install 'opentelemetry-proto' and 'grpcio' via pip to facilitate gRPC data transmission.

Where are my tracing credentials stored?

Settings are saved in the standard Claude configuration directory, specifically in ~/.claude/settings.json for global use or .claude/settings.local.json for project-specific needs.

Does this skill work with the Python Agent SDK?

Yes, it provides specific code snippets and configuration steps to integrate tracing into both Python and TypeScript versions of the Claude Agent SDK.

How can I verify the tracing setup is working?

You can set ARIZE_DRY_RUN=true in your settings to verify that hooks are firing without sending data, and check the logs located at /tmp/arize-claude-code.log.

Arize Claude Code Tracing

Name: Arize Claude Code Tracing
Author: Arize-ai

byArize-ai

•

分析与监控

Configures comprehensive observability and OpenInference tracing for Claude Code sessions and Agent SDK applications.

This skill simplifies the process of integrating Arize AX and Phoenix tracing into Claude Code and the Anthropic Agent SDK. It provides a guided decision-tree workflow for setting up self-hosted Phoenix instances or Arize AX cloud accounts, managing OTLP endpoints, and configuring essential environment variables like Space IDs and API keys. Whether you are debugging local sessions or deploying production-grade AI agents, this skill ensures your LLM interactions are fully observable, providing detailed spans for prompts, tool executions, and subagent workflows.

主要功能

01Built-in troubleshooting tools including dry-run modes and verbose debug logging.

02Guided setup for both Arize AX (cloud/on-prem) and Phoenix (self-hosted) backends.

03Automated dependency validation for OpenTelemetry and gRPC requirements.

04Full support for OpenInference tracing in both CLI and Agent SDK (Python/TypeScript).

05Automated configuration of global and project-local Claude settings files.

068 GitHub stars

使用场景

01Monitoring LLM performance and tool calls in real-time during local development.

02Centralizing observability data from multiple Claude Code projects into a unified Arize dashboard.

03Debugging complex multi-turn agentic workflows within the Claude Agent SDK.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add arize-ai/arize-claude-code-plugin setup-claude-code-tracing

For use in Claude.ai and ChatGPT

Download Skill