Prompt Optimizer Agent

English | 中文

An intelligent Prompt optimization system based on multi-Agent collaboration, implementing workflow orchestration with LangGraph, supporting multiple large language models, and providing both A2A (Agent-to-Agent) service interfaces and Web UI.

✨ Core Features

🤖 Multi-Agent Collaboration: Professional Agent workflow based on LangGraph
🔄 7-Step Optimization Process: Complete prompt engineering optimization pipeline
🌐 Multi-Model Support: Support for Google Gemini and OpenAI GPT models
🚀 A2A Service: Standardized Agent-to-Agent interface
🌟 Web Interface: Gradio-based streaming interactive interface
📊 Intelligent Evaluation: Automatically generate evaluation metrics and improvement suggestions
🎯 Role-Specific: Customized optimization strategies for different user roles
🔒 Proxy Support: Built-in proxy configuration for network-restricted environments
📝 Smart Parsing: Support for JSON and natural language input
🛡️ Robustness: Comprehensive error handling and state management
🔍 Example Validation: Field name consistency checking and format validation
📋 Session Management: Intelligent state management and history tracking

🔧 Supported Models

Provider	Model Name	Model Type ID	Configuration Required
Google	Gemini 2.0 Flash	`gemini`	`GOOGLE_API_KEY`
OpenAI	gpt-4o-2024-11-20	`openai`	`OPENAI_API_KEY`

🌐 Proxy Configuration

The system has built-in proxy support for convenient use in network-restricted environments:

Environment Variable Configuration

# Configure proxy in .env file
HTTPS_PROXY=http://127.0.0.1:7890
HTTP_PROXY=http://127.0.0.1:7890

# Or set when starting
export HTTPS_PROXY=http://127.0.0.1:7890
export HTTP_PROXY=http://127.0.0.1:7890
python main.py

Default Proxy Settings

If environment variables are not set, the system defaults to:

HTTPS Proxy: http://127.0.0.1:7890
HTTP Proxy: http://127.0.0.1:7890

Common Proxy Software Ports

Proxy Software	Default Port	Configuration Example
Clash	7890	`http://127.0.0.1:7890`
V2Ray	1081	`http://127.0.0.1:1081`
SSR	1080	`http://127.0.0.1:1080`

📋 7-Step Optimization Process

Generate Engineering Guide - Generate detailed prompt engineering guidelines for specific roles
Analyze Role and Requirements - Understand user roles and basic requirements
Generate Initial Prompt - Generate basic prompt based on role and requirements
Create Evaluation Framework - Create targeted prompt evaluation framework
Execute Prompt Evaluation - Comprehensive quality analysis of the prompt
Generate Improvement Plans - Provide 3 different optimization versions
Select Best Version - Automatically recommend or user-select final prompt

🚀 Quick Start

Environment Setup

Clone Project

git clone <repository-url>
cd prompt_agent

Install Dependencies

pip install -r requirements.txt

Configure API Keys and Server

# Copy configuration template
cp config_example.env .env

# Edit .env file, add your API keys and server settings
GOOGLE_API_KEY=your_google_api_key_here
OPENAI_API_KEY=your_openai_api_key_here  # Optional

# Server configuration
SERVER_HOST=0.0.0.0
SERVER_PORT=9999
WORKERS=4
RELOAD=false

# Web interface configuration
WEB_HOST=0.0.0.0
WEB_PORT=7860

# Logging configuration
LOG_LEVEL=info
VERBOSE_LOGGING=false
LOG_FILE_PATH=prompt_optimizer.log

# Proxy configuration
ENABLE_DEFAULT_PROXY=true
HTTPS_PROXY=http://127.0.0.1:7890  # Optional, used if ENABLE_DEFAULT_PROXY=true
HTTP_PROXY=http://127.0.0.1:7890    # Optional, used if ENABLE_DEFAULT_PROXY=true

# Model configuration
DEFAULT_MODEL_TYPE=openai
MODEL_TEMPERATURE=0.7
REQUEST_TIMEOUT=60
MAX_RETRIES=3

Starting Services

A2A Service (Recommended)

python main.py

Service will start at http://localhost:9999

Web Interface

python web.py

Web interface will start at http://localhost:7860

Development Mode

# Enable hot reload and debug logging
RELOAD=true
LOG_LEVEL=debug
VERBOSE_LOGGING=true
python main.py

🖥️ Web Interface Features

🚀 Prompt Optimization Page

Streaming Processing: Real-time optimization progress display
Multi-Format Support: JSON and text format example input
Real-time Validation: Automatic input parameter validation
Result Display: Formatted optimization result display

🔧 Manual Validation Page

Variable Management: Automatic extraction and validation of prompt variables
Real-time Preview: Preview of prompt after variable replacement
Error Messages: Detailed validation error information
Variable Hints: Intelligent variable definition suggestions

📖 Usage Guide Page

Complete Documentation: Detailed usage guidelines
Example Showcase: Multiple input format examples
Best Practices: Prompt engineering best practice guidance

📖 Usage Methods

Quick Start Example

The simplest way is to directly send keywords:

{
  "messages": [
    {
      "role": "user",
      "content": "software developer"
    }
  ]
}

The system will automatically generate basic configuration for the software developer role.

A2A Interface Call

Send POST request to http://localhost:9999/

Standard Request Format:

{
  "messages": [
    {
      "role": "user", 
      "content": {
        "role": "Target user role",
        "basic_requirements": "Basic task requirements",
        "examples": [...],  // Optional
        "additional_requirements": "Additional requirements",  // Optional
        "model_type": "Model type"  // Optional, defaults to openai
      }
    }
  ]
}

🤖 Gemini Model Example

{
  "messages": [
    {
      "role": "user",
      "content": {
        "role": "software developers",
        "basic_requirements": "Write high-quality, maintainable Python code, including functions, classes, and API design",
        "model_type": "gemini",
        "examples": [
          {
            "input": "{\"function_name\": \"calculate_fibonacci\", \"input_type\": \"int\", \"output_type\": \"int\", \"description\": \"Calculate the nth Fibonacci number\"}",
            "output": "def calculate_fibonacci(n: int) -> int:\n    \"\"\"Calculate the nth Fibonacci number\n    Args:\n        n: Position to calculate\n    Returns:\n        int: Fibonacci number\n    \"\"\"\n    if n <= 1:\n        return n\n    return calculate_fibonacci(n-1) + calculate_fibonacci(n-2)"
          }
        ],
        "additional_requirements": "Include type hints and documentation"
      }
    }
  ]
}

🧠 OpenAI Model Example

{
  "messages": [
    {
      "role": "user", 
      "content": {
        "role": "content creators",
        "basic_requirements": "Create engaging, well-structured blog articles and marketing copy",
        "model_type": "openai",
        "examples": [
          {
            "input": "{\"topic\": \"AI trends\", \"target_audience\": \"tech professionals\", \"tone\": \"professional\", \"word_count\": \"1000\"}",
            "output": "# The Future of AI: 5 Trends That Will Shape 2024\n\nArtificial Intelligence is revolutionizing how we work..."
          }
        ],
        "additional_requirements": "Engaging and accessible tone"
      }
    }
  ]
}

📝 Example Format Description

Field Name Consistency Requirements

Each example's input must be a JSON object, all examples' input field names must be consistent:

✅ Correct Example (consistent field names):

[
  {
    "input": "{\"function_name\": \"validate_email\", \"input_type\": \"str\", \"output_type\": \"bool\"}",
    "output": "def validate_email(email: str) -> bool: ..."
  },
  {
    "input": "{\"function_name\": \"calculate_sum\", \"input_type\": \"list\", \"output_type\": \"int\"}",
    "output": "def calculate_sum(numbers: list) -> int: ..."
  }
]

❌ Incorrect Example (inconsistent field names):

[
  {
    "input": "{\"function_name\": \"validate_email\", \"input_type\": \"str\"}",
    "output": "def validate_email(email: str) -> bool: ..."
  },
  {
    "input": "{\"method_name\": \"calculate_sum\", \"param_type\": \"list\"}",  // Different field names!
    "output": "def calculate_sum(numbers: list) -> int: ..."
  }
]

Supported Input Formats

JSON Format (recommended):

[
  {
    "input": "{\"topic\": \"sustainability\", \"audience\": \"general public\", \"tone\": \"casual\"}",
    "output": "🌱 Small changes, BIG impact! ..."
  }
]

Simple Text Format:

Input:
topic=sustainability
audience=general public
tone=casual
Output:
🌱 Small changes, BIG impact! Here are 3 easy tips...

Input:
topic=AI trends
audience=tech professionals
tone=professional
Output:
# The Future of AI: 5 Trends That Will Shape 2024...

Supported User Roles

software developers - Software developers
content creators - Content creators
customer support representatives - Customer service representatives
data scientists - Data scientists
marketing professionals - Marketing professionals
teachers - Teachers
Or any custom role

🛠️ Local Development and Testing

Running Demo Scripts

# Complete feature demo
python demo.py

# Test Web interface components
python test_web.py

# Test proxy configuration and optimization
python test_proxy_optimization.py

# A2A client test
python test_client.py

Development Environment Configuration

# Development mode environment variables
export RELOAD=true
export LOG_LEVEL=debug
export VERBOSE_LOGGING=true
export SERVER_PORT=8888
export WEB_PORT=7861

# Start development server
python main.py

📊 Output Format

{
  "role": "Target user role",
  "basic_requirements": "Basic task requirements",
  "model_type": "Model used",
  "original_examples": [
    {
      "input": "Example input",
      "output": "Example output"
    }
  ],
  "generated_prompt": "Generated initial prompt",
  "evaluations": [
    "Evaluation result 1",
    "Evaluation result 2"
  ],
  "alternative_prompts": [
    "Improvement plan 1",
    "Improvement plan 2",
    "Improvement plan 3"
  ],
  "final_recommendation": "Final recommended prompt",
  "step": "completed"
}

🏗️ System Architecture

Core Components

Multi-Agent System: Specialized generator, evaluator, improver
LangGraph Workflow: State management and process orchestration
Model Factory: Unified model creation and management interface with instance caching
A2A Integration: Standardized Agent service interface
Web Interface: Gradio-based interactive interface with streaming support
Session State: SessionState class for user session and history management

New Features

Error Handling: Classified error handling (input errors, connection errors, runtime errors)
Proxy Support: Built-in proxy configuration with automatic network request handling
Logging System: Complete logging with file output and different levels
Input Validation: Comprehensive input validation including field name consistency checking
State Management: Safe session state management avoiding global variables
Configuration Management: Rich environment variable configuration options

🔧 Configuration Options

Environment Variables

Variable Name	Required	Default	Description
`GOOGLE_API_KEY`	Conditional*	-	Google Gemini API key
`OPENAI_API_KEY`	Conditional*	-	OpenAI API key
`SERVER_HOST`	No	`0.0.0.0`	Server host address
`SERVER_PORT`	No	`9999`	Server port
`WEB_HOST`	No	`0.0.0.0`	Web interface host address
`WEB_PORT`	No	`7860`	Web interface port
`LOG_LEVEL`	No	`info`	Log level (debug/info/warning/error)
`WORKERS`	No	`1`	Number of worker processes
`RELOAD`	No	`false`	Enable hot reload
`HTTPS_PROXY`	No	`http://127.0.0.1:7890`	HTTPS proxy address
`HTTP_PROXY`	No	`http://127.0.0.1:7890`	HTTP proxy address
`MODEL_TEMPERATURE`	No	`0.7`	Model temperature parameter
`REQUEST_TIMEOUT`	No	`60`	Request timeout (seconds)
`MAX_RETRIES`	No	`3`	Maximum retry attempts

*At least one API key must be configured

Performance Optimization

Model Instance Caching: Avoid repeated model instance creation
Workflow Caching: Reuse workflow instances for different model types
Error Recovery: Graceful degradation when individual steps fail
Input Validation: Early validation to avoid invalid requests
Logging: Detailed performance and error logs
Session Management: Intelligent state management and history tracking

🚨 Troubleshooting

Common Issues

ImportError: cannot import name 'StateGraph'
- Solution: Ensure correct langgraph version is installed: pip install langgraph>=0.4.1
API Key Error
- Check API key configuration in .env file
- Ensure at least one valid API key is configured
- Verify key format is correct
Port Occupied
- Modify port configuration in environment variables: SERVER_PORT=8888
- Or stop processes occupying the port: lsof -ti:9999 | xargs kill
Example Field Name Inconsistency Error
- Ensure all examples' input field names are exactly the same
- Check JSON format is correct
- Refer to example formats in documentation
Proxy Connection Issues
- Check if proxy software is running normally
- Verify proxy address and port configuration
- Try different proxy ports
Web Interface Inaccessible
- Check firewall settings
- Confirm web server started normally
- Check if port is occupied

Logging and Debugging

# Enable verbose logging
export LOG_LEVEL=debug
export VERBOSE_LOGGING=true

# View log file
tail -f prompt_optimizer.log

# Check server status
curl http://localhost:9999/.well-known/agent.json

Environment Check

The system automatically checks at startup:

API key configuration status
Proxy configuration status
Dependency package installation status
Port availability

📚 More Documentation

System Architecture - Detailed system architecture description
API Documentation - A2A interface specifications
Development Guide - Development and contribution guide
Changelog - Version update records

🤝 Contributing

Fork the project
Create feature branch (git checkout -b feature/AmazingFeature)
Commit changes (git commit -m 'Add some AmazingFeature')
Push to branch (git push origin feature/AmazingFeature)
Create Pull Request

📄 License

MIT License - See license file for details

🔗 Related Links

Developer: Multi-Agent prompt optimization system based on LangGraph
Version: 1.2.0 - Enhanced Robustness and Web Interface Version
Last Updated: December 2024

🎯 Version Features

v1.2.0 New Features

✨ Complete Web interface support
🛡️ Enhanced error handling and state management
🔍 Example field name consistency validation
📋 Session state and history management
🔧 Rich environment variable configuration
📊 Detailed logging and monitoring
🚀 Streaming processing and real-time feedback

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
ARCHITECTURE.md		ARCHITECTURE.md
LICENSE		LICENSE
README.md		README.md
README_CN.md		README_CN.md
config_example.env		config_example.env
demo.py		demo.py
main.py		main.py
prompt_optimizer.py		prompt_optimizer.py
prompt_optimizer_executor.py		prompt_optimizer_executor.py
requirements.txt		requirements.txt
test_client.py		test_client.py
test_proxy_optimization.py		test_proxy_optimization.py
test_web.py		test_web.py
web.py		web.py

License

longmans/prompt_agent

Folders and files

Latest commit

History

Repository files navigation