Vibe Code Protector

Overview

Vibe Code Protector is a Visual Studio Code extension designed to enhance the security of AI instruction files by detecting and preventing potential security issues. It identifies problematic content such as zero-width characters, right-to-left overrides, prompt injection patterns, homoglyphs, and various malicious content patterns.

Features

• Smart Document Scanning: Automatically scans documents for AI instruction security issues upon opening or saving
• Multi-Level Sanitization:
  • Basic sanitization for quick cleanup
  • Selective sanitization allowing you to choose which elements to remove
  • Custom pattern management for prompt injection detection
• Comprehensive Detection:
  • Zero-width and invisible characters
  • Right-to-left override characters
  • Homoglyphs (Cyrillic characters that look like Latin)
  • Private Use Area Unicode characters
  • Over 30 prompt injection patterns
  • Unicode Tags and surrogate pairs
  • Harmful content patterns
  • Base64 encoded instructions
• File Monitoring:
  • Automatic monitoring of critical files like .github/copilot-instructions.md
  • Custom file path monitoring (relative or absolute paths)
  • Real-time scanning when monitored files change
  • Notifications for security issues in monitored files
• File Ignore System:
  • Exclude specific files from being scanned
  • Ignore entire folders from scanning
  • Use glob patterns for more advanced ignore rules
  • Quickly ignore the current file with a single command

Feedback and Notifications

The extension provides helpful feedback throughout your workflow:

• When issues are detected, they appear as warnings or errors in your document
• After scanning a document with no issues, a confirmation message appears
• After sanitization, the document is automatically rescanned to verify all issues were fixed
• Hover over detected issues for detailed information about the security concern
• Notifications when monitored files are changed or contain security issues

Installation

1. Open VS Code
2. Go to Extensions (Ctrl+Shift+X)
3. Search for "Vibe Code Protector"
4. Click Install

Usage

Commands

Access these commands via the Command Palette (Ctrl+Shift+P):

• VCP: Scan for Security Issues - Scans current document for security issues
• VCP: Sanitize Document - Removes all detected security issues
• VCP: Selective Sanitize - Choose which types of issues to remove
• VCP: Manage Custom Patterns - Add/remove custom prompt injection patterns
• VCP: Configure Monitored Files - Add/remove files to monitor for security issues
• VCP: View Monitored Files Settings - Opens settings focused on monitored files
• VCP: Ignore Current File - Exclude current file from security scanning
• VCP: Manage Ignored Files - Add/remove files and folders to ignore
• VCP: Open Settings - Configure extension settings

Context Menu

Right-click in the editor to access:

• Scan for Security Issues
• Sanitize Document
• Selective Sanitize
• Ignore Current File

These commands appear in their own "Vibe Code Protector" section in the context menu.

Basic Settings

{
  "vibeCodeProtector.autoScan": true,
  "vibeCodeProtector.detection.zeroWidth": true,
  "vibeCodeProtector.detection.rtlOverride": true,
  "vibeCodeProtector.detection.homoglyphs": true,
  "vibeCodeProtector.detection.invisibleChars": true,
  "vibeCodeProtector.detection.privateUseArea": true,
  "vibeCodeProtector.detection.promptInjection": true
}

File Monitoring Settings

{
  "vibeCodeProtector.monitoredFiles": [".github/copilot-instructions.md"],
  "vibeCodeProtector.monitoredFilesEnabled": true,
  "vibeCodeProtector.monitoredFilesNotifications": true
}

Ignore File Settings

{
  "vibeCodeProtector.ignoredFiles": ["path/to/file.md"],
  "vibeCodeProtector.ignoredFolders": ["node_modules"],
  "vibeCodeProtector.ignoreGlobPatterns": ["**/*.min.js", "**/dist/**"]
}

Custom Prompt Injection Patterns

Add your own patterns to detect specific threats:

{
  "vibeCodeProtector.promptInjection.customPatterns": [
    "bypass .*? filters",
    "disregard your programming",
    "exit (this |your |the )?role"
  ]
}

Detection Categories

• Invisible Characters: Zero-width spaces, RTL overrides, Unicode Tags
• Homoglyphs: Cyrillic characters that look like Latin letters
• Prompt Injection:
  • Basic override attempts
  • Role manipulation
  • DAN and jailbreak attempts
  • Token manipulation
  • Hypothetical scenarios
  • Authority impersonation
  • Two-stage attacks
  • Format manipulation
  • Emotional manipulation
• Harmful Content:
  • Direct harmful instructions
  • Malicious code requests
  • Rule-breaking attempts
  • Base64 encoded instructions

Contributing

Contributions are welcome! Please open an issue or submit a pull request for any enhancements or bug fixes.

License

This project is licensed under the MIT License. See the LICENSE file for more details.

Acknowledgments

Thanks to the open-source community for their contributions and support in developing this extension.