Skip to main content

CLI Reference

Full documentation

npm

Install from npm

​
Overview

Phoenix CLI now supports prompt introspection with two new commands: px prompts and px prompt. These commands let you list, view, and export prompt templates directly from the terminal—and pipe them to AI coding assistants for optimization. 
# List all prompts
px prompts

# View a specific prompt
px prompt my-evaluator

# Get prompt text for piping
px prompt my-evaluator --format text --no-progress

​
Prompt Optimization with AI Assistants

The --format text option outputs prompt content in a clean format with XML-style role tags, designed for piping to AI assistants like Claude Code: 
# Get optimization suggestions
px prompt my-evaluator --format text --no-progress | claude -p "Review this prompt and suggest improvements"

# Analyze for edge cases
px prompt my-assistant --format text --no-progress | claude -p "What edge cases might this prompt fail to handle?"

# Generate test inputs
px prompt my-classifier --format text --no-progress | claude -p "Generate 5 diverse test inputs for this prompt"
The text output format: 
<system>You are an expert evaluator...

<rubric>
A correct answer:
- Matches the reference exactly
- Contains no hallucinations
</rubric></system>
<user>Evaluate this response: {{output}}</user>

​
Use Cases

​
Prompt Version Management

Access specific versions of your prompts: 
# Get the production-tagged version
px prompt my-assistant --tag production

# Get a specific version by ID
px prompt my-assistant --version UHJvbXB0VmVyc2lvbjox

​
Prompt Analysis

Export prompts for offline analysis or documentation: 
# Full JSON with model settings and tools
px prompt my-evaluator --format json

# List all prompts as JSON
px prompts --format raw --no-progress | jq '.[].name'

​
Interactive Development

Ask AI assistants to work with your prompts directly: 
Fetch my "correctness-evaluator" prompt from Phoenix and help me make the rubric more specific for code review tasks

​
Why This Matters

Prompts are code. They should be versioned, reviewed, and optimized with the same rigor as application code. By making prompts accessible through the CLI:
  1. Version Control Integration: Export prompts to files for git tracking
  2. AI-Assisted Optimization: Pipe prompts to Claude, GPT, or other models for improvement suggestions
  3. Automated Testing: Generate test cases programmatically from prompt templates
  4. Documentation: Extract prompt content for technical documentation
This continues the pattern of making Phoenix data accessible where developers already work—in the terminal and through AI coding assistants.

​
Commands Reference

​
px prompts

List all prompts with names and descriptions.
OptionDescription
--format <format>pretty, json, or raw
--limit <number>Maximum prompts to fetch

​
px prompt <identifier>

Show a specific prompt’s content and configuration.
OptionDescription
--tag <name>Get version by tag
--version <id>Get specific version
--format <format>pretty, json, raw, or text
The text format is optimized for piping—no progress messages, just the prompt content with role tags.

​
Installation

npm install -g @arizeai/phoenix-cli
Or update an existing installation: 
npm update -g @arizeai/phoenix-cli