Smarty Jones

A super lightweight debugging assistant that provides AI-powered error analysis with minimal setup.

Features

• AI-powered analysis: Uses Claude Sonnet via ChatOpenAI for intelligent error diagnosis
• Auto source code analysis: Automatically reads and analyzes source code from stack traces
• Rich context support: Pass files, directories, or complex nested data for better debugging
• Zero configuration: Just install and one line to activate
• Security first: Library-controlled prompts prevent prompt injection attacks

Installation

pip install smarty-jones

Quick Start

from smarty_jones import SmartyJonesHandler

# Install with your OpenAI-compatible endpoint
SmartyJonesHandler.install(
    endpoint_url="https://api.anthropic.com/v1/chat/completions",  # or your endpoint
    api_token="your-api-key",
    model="claude-4-6-sonnet"  # Optional, defaults to claude-4-6-sonnet
)

# Now any unhandled exception gets AI analysis
def test_function():
    return 1 / 0

test_function()  # This will trigger AI-powered error analysis

Advanced Usage

With Additional Context

# Pass additional context for better analysis
SmartyJonesHandler.install(
    endpoint_url="https://api.anthropic.com/v1/chat/completions",
    api_token="your-api-key",
    model="gpt-4",  # Use different model
    config_file="app.json",                    # Single file
    documentation="/path/to/docs/",            # Entire directory
    user_data={"session_id": "abc123"},        # Custom data
    business_rules="rules.yaml"                # Business context
)

Nested File Paths Support

# Complex nested structures with file paths
SmartyJonesHandler.install(
    endpoint_url="https://api.anthropic.com/v1/chat/completions", 
    api_token="your-api-key",
    model="claude-4-6-sonnet",  # Default model
    config={
        "business_rules_path": "rules.yaml", 
        "user_profiles": ["profile1.json", "profile2.json"]
    }
)

What You'll See

🤖 Smarty Jones Analysis:
========================================
📝 You're dividing by zero in line 15 of main.py
💡 Add a check: if denominator != 0 before division
🔍 Context: Variable 'denominator' was set to 0 in the loop above
📊 Confidence: 95%
========================================

Traceback (most recent call last):
  File "main.py", line 15, in <module>
    result = numerator / denominator
ZeroDivisionError: division by zero

How It Works

1. Global Exception Hooking: Captures all unhandled exceptions automatically
2. Source Code Analysis: Reads the actual source code from your stack trace
3. Context Collection: Processes files, directories, and nested data structures
4. AI Analysis: Sends structured data to Claude Sonnet for intelligent diagnosis
5. Formatted Output: Returns clear, actionable debugging advice

Security Considerations

⚠️ Important: While Smarty Jones includes security protections, you are responsible for the files and data you pass as additional context.

What Smarty Jones Protects Against:

• Blocks sensitive file types: Automatically excludes .env, .key, .pem, and other credential files
• Content filtering: Redacts common secret patterns (API keys, passwords, tokens) from file contents
• Prompt injection: Uses library-controlled prompts to prevent malicious input

Your Responsibility:

• Review file paths: Ensure directories don't contain unintended sensitive files
• Validate custom data: Check any custom context data before passing it
• Use appropriate endpoints: Only use trusted AI service endpoints for your data

Best Practices:

# ✅ Good - specific files you control
SmartyJonesHandler.install(
    endpoint_url="...",
    config_file="./app/config.yaml",
    docs_path="./docs/"
)

# ⚠️ Risky - entire home directory
SmartyJonesHandler.install(
    endpoint_url="...",
    everything="~/"  # May contain sensitive files
)

Uninstalling

SmartyJonesHandler.uninstall()  # Removes the exception handler

Development

# Clone and install in development mode
git clone https://github.com/FrankSiderio/smarty-jones.git
cd smarty-jones
pip install -e .

# Run the test
python test_minimal.py

Security

• No prompt injection: User data is passed as structured JSON, not interpolated into prompts
• Library-controlled prompts: All AI instructions are hardcoded in the library
• Local file reading: Only reads files you explicitly specify
• Safe by default: Won't read system files or execute arbitrary code

Requirements

• Python 3.8+
• langchain-openai
• langchain-core

Total dependencies: Just 2 lightweight packages!

License

MIT License - see LICENSE file for details.