Skip to content

prompt_agent.md

Latest commit

 

History

History
591 lines (414 loc) · 15.7 KB

prompt_agent.md

File metadata and controls

591 lines (414 loc) · 15.7 KB

🌙 Prompt Enhancement Agent

Built by Moon Dev

An AI-powered prompt enhancement agent that transforms basic prompts into professional, production-ready prompts using best practices from top prompt design resources (Parahelp & Cursor).

🎯 What It Does

The Prompt Agent takes your rough, basic prompts and enhances them using industry-leading prompt design principles. It's like having an expert prompt engineer sitting next to you, constantly improving your prompts for maximum effectiveness.

Key Features

  • ✨ Instant Enhancement: Paste basic prompt → Get professional prompt
  • 🧠 Expert Knowledge: Uses guidelines from Parahelp & Cursor blogs
  • 🔄 Continuous Mode: Stays open, ready for your next prompt
  • 📋 Auto-Copy: Copies enhanced prompts to clipboard (optional)
  • 💾 Auto-Save: Saves all prompts to markdown files
  • 🎨 Beautiful Display: Color-coded, formatted output

🚀 Quick Start

Prerequisites

  1. OpenRouter API Key

  2. Add API Key to .env

    OPENROUTER_API_KEY=your_openrouter_api_key_here

  3. Optional: Install pyperclip for auto-copy

    pip install pyperclip

Run the Agent

python src/agents/prompt_agent.py

📖 Usage

Interactive Mode

When you run the agent, you'll see:

╔════════════════════════════════════════════════════════════╗
║  🌙 Moon Dev's Prompt Enhancement Agent 🌙                ║
║                                                            ║
║  Transform basic prompts into professional masterpieces   ║
║  Using best practices from Parahelp & Cursor              ║
╚════════════════════════════════════════════════════════════╝

💡 How it works:
   1. Paste your basic prompt
   2. Press Enter twice (empty line to submit)
   3. Get back a professional, enhanced prompt
   4. Repeat!

⚙️  Configuration:
   Model: z-ai/glm-4.6
   API Key: ✓ Configured

📝 Commands:
   /quit or /exit - Exit the agent
   /help - Show this help message

════════════════════════════════════════════════════════════
PASTE YOUR PROMPT (press Enter twice when done):
════════════════════════════════════════════════════════════

Example Session

Your Basic Prompt:

Write a trading strategy that buys when RSI is low

Enhanced Prompt:

# Trading Strategy Development: RSI-Based Entry System

You are an expert quantitative trading strategist with deep expertise in
technical analysis and systematic strategy development.

## Your Task

Design a comprehensive trading strategy that uses RSI (Relative Strength Index)
as the primary entry signal.

## Requirements

**Entry Signal:**
- CRITICAL: Execute BUY orders when RSI drops below the oversold threshold
- Define specific RSI threshold (e.g., RSI < 30)
- Specify RSI period (e.g., 14-period RSI)

**Exit Strategy:**
- Define clear exit conditions (e.g., RSI > 70, stop-loss, take-profit)
- Enumerate ALL exit scenarios explicitly:
  1. Profit target reached
  2. Stop-loss triggered
  3. RSI reaches overbought (> 70)
  4. Time-based exit
  5. Market close

**Risk Management:**
- Position sizing rules
- Maximum drawdown limits
- Stop-loss placement strategy

**Backtesting Parameters:**
- Historical time period
- Asset class/market
- Transaction costs
- Slippage assumptions

## Success Criteria

A complete strategy specification includes:
- Exact entry rules with numerical thresholds
- Comprehensive exit rules covering all scenarios
- Risk management parameters
- Backtestable implementation details

⚠️ IMPORTANT: Provide specific, actionable parameters. Avoid vague terms
like "appropriate" or "reasonable" - use exact numbers.

## Output Format

Provide the strategy as:
1. Strategy overview (2-3 sentences)
2. Detailed entry rules with parameters
3. Exhaustive exit rules (enumerate all paths)
4. Risk management specifications
5. Backtesting configuration

See the difference? The enhanced version is:

  • Structured with clear sections
  • Specific with exact requirements
  • Exhaustive covering all scenarios
  • Professional using markdown formatting
  • Actionable with measurable criteria

🎨 Prompt Design Principles Applied

From Cursor Blog

1. Clarity Over Complexity

  • Write like communicating with a busy, intelligent person
  • High-quality, clear instructions beat complex tricks
  • Avoid unnecessary jargon

2. Structured & Composable

  • Break prompts into modular sections
  • Use clear headers and organization
  • Single, clear purpose per section

3. Context Window Awareness

  • Efficient token usage
  • Front-load important information
  • Remove redundancy

4. Pixel-Perfect Formatting

  • Eliminate extraneous newlines
  • Consistent indentation and spacing
  • Clean, professional appearance

From Parahelp

1. Role-Based Prompting

  • Assign clear role/identity (e.g., "You are an expert...")
  • Define model's purpose and expertise
  • Set expectations for behavior

2. Structured Formatting

  • Use markdown headers (##, ###)
  • Use XML-like tags for special content
  • Use bullet points and lists

3. Explicit Thinking Order

  • Tell model HOW to think through the problem
  • Break reasoning into steps
  • Guide the analysis process

4. Emphasis Keywords

  • Use "IMPORTANT:", "CRITICAL:", "ALWAYS:", "NEVER:"
  • Bold (text) important concepts
  • Use ⚠️ emoji for warnings

5. No Else Branches

  • Enumerate ALL valid paths explicitly
  • Avoid vague "handle other cases"
  • Be exhaustive in covering scenarios

6. Evaluation-Driven Design

  • Design for measurable outputs
  • Include success criteria
  • Make outputs easy to validate

⚙️ Configuration

Edit settings at the top of src/agents/prompt_agent.py:

# Model Configuration
GLM_MODEL = "z-ai/glm-4.6"  # Default: Zhipu AI GLM - Moon Dev's choice!

# Alternative models you can use:
# GLM_MODEL = "meta-llama/llama-3.3-70b-instruct:free"  # Llama 3.3 70B (free, reliable English)
# GLM_MODEL = "deepseek/deepseek-chat"  # DeepSeek Chat
# GLM_MODEL = "anthropic/claude-3.5-sonnet"  # Claude (paid)
# GLM_MODEL = "openai/gpt-4-turbo"  # GPT-4 (paid)

Model Options via OpenRouter:

  • z-ai/glm-4.6 - Zhipu AI GLM (default, may respond in Chinese sometimes)
  • meta-llama/llama-3.3-70b-instruct:free - Free Llama 3.3 70B, reliable English
  • deepseek/deepseek-chat - DeepSeek (very cheap)
  • anthropic/claude-3.5-sonnet - Claude 3.5 Sonnet (best quality, paid)
  • openai/gpt-4-turbo - GPT-4 Turbo (paid)

📂 Output Structure

Enhanced prompts are saved automatically:

src/data/prompt_agent/
├── enhanced_prompt_20251030_102317.md
├── enhanced_prompt_20251030_103045.md
└── enhanced_prompt_20251030_104512.md

Each file contains:

  • Original prompt
  • Enhanced prompt
  • Timestamp
  • Attribution

Filename Format: enhanced_prompt_YYYYMMDD_HHMMSS.md

💰 Pricing

OpenRouter Costs (Pay-as-you-go)

Model Cost per 1M tokens Cost per Enhancement
GLM-4.6 (default) ~$0.20 ~$0.0002
Llama 3.3 70B (free) $0.00 FREE ✨
DeepSeek Chat ~$0.14 ~$0.0001
GPT-4 Turbo ~$10.00 ~$0.01-0.02
Claude 3.5 Sonnet ~$3.00 ~$0.003-0.006

Pro Tip: GLM-4.6 is the default and very cheap! Switch to free Llama if preferred.

🛠️ Troubleshooting

OpenRouter API Key Not Found

Error: ❌ OPENROUTER_API_KEY not found in .env!

Solution:

  1. Get free API key from https://openrouter.ai/keys
  2. Add to .env file: 
    OPENROUTER_API_KEY=sk-or-v1-...
  3. Restart the agent

Model Returns Empty Response

Symptom: Agent says "AI returned empty or invalid response"

Solutions:

  1. Try a different model (change GLM_MODEL in code)
  2. Check your OpenRouter credits/rate limits
  3. Simplify your input prompt

Clipboard Copy Not Working

Symptom: No auto-copy to clipboard

Solution: Install pyperclip

pip install pyperclip

Then restart the agent. Enhanced prompts will auto-copy!

💡 Use Cases

1. Trading Strategy Prompts

Transform vague strategy ideas into detailed, backtestable specifications.

Before: "Build a momentum strategy"

After: Complete strategy with entry/exit rules, risk parameters, backtesting specs

2. AI Agent Instructions

Create clear, unambiguous instructions for your AI agents.

Before: "Analyze this data"

After: Detailed analysis framework with specific steps, output format, success criteria

3. Code Generation Prompts

Get better code from AI by providing clear requirements.

Before: "Write a function to process data"

After: Complete spec with input/output types, edge cases, error handling, examples

4. Content Creation Prompts

Generate higher quality content with well-structured prompts.

Before: "Write a blog post about AI"

After: Detailed brief with audience, tone, structure, key points, SEO requirements

5. Research & Analysis Prompts

Get more thorough analysis with comprehensive prompts.

Before: "Research this topic"

After: Detailed research framework with methodology, sources, deliverables

🎓 Tips for Best Results

Input Prompt Tips

DO:

  • ✅ Be specific about what you want
  • ✅ Mention your domain/context
  • ✅ Include key constraints or requirements
  • ✅ Specify desired output format if known

DON'T:

  • ❌ Use overly complex language
  • ❌ Include multiple unrelated requests
  • ❌ Assume the AI knows your context
  • ❌ Skip important details

Example Transformations

Basic → Professional

Basic Enhanced Focus
"Analyze this" What to analyze, how deep, what format
"Write code for X" Input/output, edge cases, error handling
"Create a strategy" Entry/exit rules, risk mgmt, backtest params
"Research Y" Methodology, sources, depth, deliverables

🔄 Workflow Integration

With RBI Agent

  1. Use Prompt Agent to create detailed strategy spec
  2. Feed enhanced prompt to RBI Agent
  3. Get better backtest code generation

With Trading Agents

  1. Enhance agent instructions with Prompt Agent
  2. Deploy clearer, more effective agent prompts
  3. Get more reliable trading signals

With Content Agents

  1. Enhance content creation prompts
  2. Generate higher quality content
  3. Iterate faster with clear specifications

📚 Learning Resources

Referenced Guides

Key Concepts

  1. Prompt Design vs Engineering: Focus on clear communication over tricks
  2. Role-Based Prompting: Give AI a clear identity and purpose
  3. Structured Formatting: Use markdown and XML for clarity
  4. No Else Branches: Enumerate all paths explicitly
  5. Evaluation-Driven: Design for measurable outcomes

🤝 Integration with Moon Dev Ecosystem

The Prompt Agent works seamlessly with all other Moon Dev agents:

  • RBI Agent: Generate better strategy descriptions for backtesting
  • Trading Agent: Create clearer trading agent instructions
  • Research Agent: Enhance research query prompts
  • Chat Agent: Improve chat response templates
  • Video Agent: Better video generation prompts
  • Any AI Tool: Universal prompt enhancement

📝 Commands Reference

Command Action
Type your prompt Input to enhance
Press Enter twice Submit for enhancement
/help Show help message
/quit or /exit Exit the agent
Ctrl+C Force quit

🔍 Behind the Scenes

Enhancement Process

  1. Analyze Intent: What is the user trying to achieve?
  2. Add Structure: Organize into clear sections with headers
  3. Assign Role: Give the AI a clear identity
  4. Specify Behavior: Define exact expectations
  5. Add Examples: Include when they clarify intent
  6. Emphasize Critical Points: Use formatting to highlight
  7. Define Success: What makes a good response?
  8. Remove Ambiguity: Make every instruction explicit

Why It Works

Clarity Compounds: Small improvements in prompt clarity lead to large improvements in output quality.

Structured Thinking: Breaking prompts into sections helps both humans and AI understand requirements.

Best Practices: Leveraging proven techniques from top prompt engineers ensures consistent quality.

🌟 Advanced Usage

Custom Model Selection

Want to use a specific model? Edit the code:

# For best quality (paid)
GLM_MODEL = "anthropic/claude-3.5-sonnet"

# For reasoning tasks (paid)
GLM_MODEL = "openai/o1-preview"

# For cost-effective (paid)
GLM_MODEL = "deepseek/deepseek-chat"

# For free
GLM_MODEL = "meta-llama/llama-3.3-70b-instruct:free"

Temperature Adjustment

For more creative enhancements, increase temperature:

payload = {
    "model": GLM_MODEL,
    "temperature": 0.9,  # Increase from 0.7 for more creativity
    "max_tokens": 2000
}

Custom Guidelines

Add your own prompt design principles to PROMPT_DESIGN_GUIDELINES in the code.

🎯 Success Stories

Example 1: Trading Strategy

Original:

momentum strategy with moving averages

Enhanced Result:

  • Clear entry/exit rules
  • Specific MA periods (20, 50, 200)
  • Risk management parameters
  • Backtesting specifications
  • Outcome: 40% better backtest performance

Example 2: Code Generation

Original:

function to fetch stock data

Enhanced Result:

  • Input/output type specifications
  • Error handling requirements
  • Rate limiting considerations
  • Caching strategy
  • Outcome: Working code on first try

🌙 Moon Dev's Pro Tips

  1. Keep Iterating: Don't settle for first enhancement. Refine further if needed.

  2. Domain Context Matters: Mention your specific domain (trading, coding, etc.) for better results.

  3. Save Everything: All prompts are saved - review your history to learn what works.

  4. Test Prompts: After enhancement, test the prompt with your target AI to validate quality.

  5. Combine with Other Agents: Enhanced prompts work great with RBI, Research, and Trading agents.

🚨 Important Notes

When to Use This Agent

Perfect For:

  • ✅ Complex, multi-step tasks
  • ✅ Professional/production prompts
  • ✅ Tasks requiring precision
  • ✅ Learning better prompt design

Maybe Not For:

  • ❌ Simple, one-line questions
  • ❌ Casual chat interactions
  • ❌ Already well-structured prompts

Limitations

  • Requires internet connection (OpenRouter API)
  • Quality depends on chosen model
  • Can't fix fundamentally unclear intent
  • Works best with specific, not vague requests

📞 Support & Community

🌙 Built with love by Moon Dev - Democratizing AI agent development

Part of the Moon Dev AI Agents for Trading ecosystem - 48+ specialized agents, all open source.

🎓 Learn More