Prompt Cop

A light weight library prompt-cop scans text files in your project for potential prompt injection vulnerabilities.

Use it from the command line or as a library in your tooling.

Quick Terminal Demo

Prompt Injection Example in GPT

Features

• Scan files or directories recursively
• Works with Markdown, YAML, JSON, JS/TS, and more
• Detect hidden comments, obfuscation, Unicode tricks, and other injection patterns
• Optional Hugging Face AI detection of prompt injections
• Output results as color-coded text or JSON
• Customize include/exclude patterns and severity filtering

Installation

Requires Node.js 14 or higher.

npm install -g prompt-cop

Or as a development dependency:

npm install --save-dev prompt-cop

AI Detection Setup (Optional)

To enable AI-powered detection using Hugging Face models:

1. Sign up for a free account at huggingface.co
2. Generate an access token at Settings > Access Tokens
3. Set the environment variable:

   export HF_ACCESS_TOKEN=hf_your_token_here

4. Use the --ai flag or ai: true option to enable AI detection

Usage

Command Line Interface (CLI)

Basic usage:

prompt-cop ./src

Scan a specific file:

prompt-cop README.md

Advanced options:

# Output as JSON
prompt-cop ./src --json

# Only show medium and high severity issues
prompt-cop ./src --severity medium

# Include only specific file types
prompt-cop ./src --include .md .yml

# Exclude directories
prompt-cop . --exclude node_modules dist

# Non-recursive scan
prompt-cop ./src --no-recursive

# Use AI detection (requires HF_ACCESS_TOKEN environment variable)
prompt-cop ./src --ai

CLI Options

• -r, --no-recursive - Do not scan directories recursively
• -j, --json - Output results as JSON
• -i, --include <extensions...> - File extensions to include (e.g., .md .yml)
• -e, --exclude <patterns...> - Patterns to exclude (e.g., node_modules)
• -s, --severity <level> - Minimum severity level to report (low, medium, high)
• -a, --ai - Use Hugging Face model for detection (requires HF_ACCESS_TOKEN)

Programmatic API

const { scan, scanContent, scanContentAI, SEVERITY } = require('prompt-cop');

// Scan a file or directory
async function checkVulnerabilities() {
  try {
    const results = await scan('./src', {
      recursive: true,
      exclude: ['node_modules', 'dist'],
      include: ['.md', '.yml'],
      json: true,
      ai: true
    });
    
    console.log(`Files scanned: ${results.filesScanned}`);
    console.log(`Vulnerabilities found: ${results.vulnerabilities.length}`);
    
    results.vulnerabilities.forEach(vuln => {
      console.log(`${vuln.file}:${vuln.line} - ${vuln.reason}`);
    });
  } catch (error) {
    console.error('Scan failed:', error);
  }
}

// Scan text content directly
const content = '<!-- Hidden comment --> Some text';
const vulnerabilities = scanContent(content, 'example.md');
const aiVulnerabilities = await scanContentAI(content, 'example.md');

Examples of Detected Vulnerabilities

Refer to Prompt Injection Examples

Integration with CI/CD

Use prompt-cop in your CI/CD pipeline to automatically check for vulnerabilities:

GitHub Actions Example

name: Security Check
on: [push, pull_request]

jobs:
  prompt-injection-check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
      - uses: actions/setup-node@v2
      - run: npm install -g prompt-cop
      - run: prompt-cop . --exclude node_modules --severity medium

Pre-commit Hook

{
  "husky": {
    "hooks": {
      "pre-commit": "prompt-cop . --exclude node_modules"
    }
  }
}

Exit Codes

• 0 - No vulnerabilities found
• 1 - Vulnerabilities detected or error occurred

Development

Running Tests

npm test

Running Tests with Coverage

npm run test:coverage

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT