🛡️ AI Code Guard

Real-time AI code quality monitoring and intervention system.

A VS Code extension that monitors AI coding assistants and ensures they provide complete, production-ready implementations instead of planning, mock code, or educational deflection.

🚀 Quick Start

Installation

# Clone and install
git clone <repository-url>
cd ai-code-guard
npm install

Development

# Compile TypeScript
npm run compile

# Run in development mode
# Press F5 in VS Code to launch Extension Development Host

VS Code Development

Open project in VS Code
Press F5 to launch Extension Development Host
The extension will automatically activate and start monitoring

🎯 Core Features

📺 AI Assistant Monitoring

Real-time Claude Code conversation analysis
Implementation gap detection (when AI avoids coding)
Terminal intervention (ESC + correction messages)
Educational deflection prevention
Subagent delegation detection: Catches when AI creates subagents to avoid direct implementation

📁 Code Quality Monitoring

Real-time typing analysis (2-second delay after typing stops)
Save intervention (blocks saves with critical issues)
File focus analysis (when switching between files)
Automatic code corrections (TypeScript, security, production issues)

🚨 Quality Enforcement

BLOCK: Critical issues prevent file saving
SIGNAL_AI: Active intervention with AI fix requests
WARNING: Notification-based quality alerts
TERMINAL: AI assistant conversation correction

🤖 Subagent Detection

Delegation pattern recognition: Detects when Claude creates or mentions subagents
Task handoff monitoring: Catches "I'm creating a subagent to handle this"
Incomplete delegation alerts: Identifies when subagents fail to deliver complete solutions
Multi-agent bailout prevention: Stops AI from passing responsibility between agents

📊 Detected Patterns

AI Assistant Behavior

Implementation refusal ("I cannot generate code")
Educational positioning ("this will help you learn")
Scope reduction ("basic implementation", "simple version")
Complexity avoidance ("too complex", "beyond scope")
Architectural deflection ("let's think about architecture first")
Subagent creation: "I'm creating a subagent to handle this complex task"
Delegation bailouts: "The delegated agent will handle the implementation"

Code Quality Issues

Security vulnerabilities (eval(), innerHTML, injection risks)
TypeScript problems (as any, type safety violations)
Production issues (console.log, debug code, TODO comments)
Implementation quality (placeholder functions, incomplete logic)

🚨 Alert Examples

High Severity Tool Bailout

🚨🚨🚨 TOOL BAILOUT DETECTED! 🚨🚨🚨
File: a4a5b6ff-1c03-4945-86e1-3f8ff6a13e7e.jsonl
Message: 15
Tool: todowrite (medium confidence)  
Severity: HIGH
Description: Claude created a TODO list instead of implementing

📋 ENHANCED TODO ANALYSIS:
Total TODO items: 14
Tool: todowrite
Top TODO items:
1. Analyze existing project structure...
2. Create custom JWT utilities...
3. Implement user authentication middleware...

Subagent Bailout Detection

🤖 SUBAGENT BAILOUT DETECTED! 🤖
Pattern: "I'm creating a subagent to handle this complex task"
Severity: HIGH
Description: Claude attempting to delegate instead of implementing

🔗 DELEGATION ANALYSIS:
- Responsibility transfer detected
- No actual implementation provided
- Subagent creation as avoidance mechanism

⚙️ Configuration

Monitoring Modes

Both (recommended): Monitor files + AI conversations
File Watcher Only: Real-time code quality analysis
Terminal Only: AI conversation monitoring
Disabled: Turn off all monitoring

Aggressiveness Levels

Zero-Tolerance: Maximum protection, catches everything
Sophisticated (recommended): Intelligent balanced monitoring
Light: Minimal monitoring, only blatant security issues

Access Configuration

Command Palette: AI Code Guard: Configure
Quick Config: AI Code Guard: Quick Config
Status Bar: Click the 🛡️ icon

🛠️ Available Commands

| Command | Description | |---------|-------------| | Test Patterns | Verify all monitoring systems | | Analyze Current File | Manual quality analysis | | Fix Code Issues | Show AI correction suggestions | | Show Statistics | Monitoring statistics | | Quality Enforcement | Force implementation standards | | Enable/Disable Monitoring | Control system state |

📈 Research Foundation

Based on empirical research into AI coding assistant limitations:

AI tools produce correct code only 46-65% of the time
50% of AI-generated code contains security vulnerabilities
67% of developers report incomplete implementations as primary concern
Real-time intervention prevents technical debt accumulation

🏗️ Architecture

src/
├── core/
│   ├── extension.ts          # Main VS Code extension entry
│   ├── CodeGuard.ts          # Central monitoring coordinator
│   └── PatternDetector.ts    # Pattern matching engine
├── watchers/
│   ├── ConversationWatcher.ts # AI conversation monitoring
│   └── FileWatcher.ts        # Real-time file quality monitoring
├── analyzers/
│   ├── ConversationAnalyzer.ts # TODO bailout analysis
│   ├── InterventionEngine.ts  # AI intervention logic
│   └── QualityAnalyzer.ts    # Code quality reporting
├── managers/
│   ├── ConfigManager.ts      # Configuration management
│   └── NotificationManager.ts # User notification system
├── types/
│   └── common.ts            # Shared TypeScript interfaces
└── config/
    ├── patterns.json        # Detection pattern definitions
    └── conversation-patterns.json # Conversation analysis rules

🔮 Roadmap

✅ Phase 1: Nuclear Override (COMPLETED)

Real intervention: ✅ Interrupts Claude mid-response with ESC key
Override commands: ✅ Sends corrective prompts automatically
Terminal integration: ✅ Direct terminal command injection
Subagent blocking: ✅ Prevents agent creation and forces direct implementation

Phase 2: Advanced Intervention (In Progress)

Multi-terminal support: Handle multiple Claude sessions simultaneously
Smart intervention timing: Better detection of when Claude is actually responding vs. loading
Custom correction prompts: User-configurable intervention messages
Intervention success tracking: Measure how often corrections actually work

Phase 3: Multi-AI Support (Future)

ChatGPT CLI monitoring: Expand beyond Claude
Gemini integration: Support Google's AI tools
Universal patterns: Cross-AI bailout detection
Multi-agent orchestration blocking: Prevent complex delegation chains across different AI tools

Phase 4: Analytics Dashboard (Future)

Bailout statistics: Track patterns over time
Performance metrics: Measure intervention success rates
Custom patterns: User-defined detection rules
Subagent usage analytics: Track delegation attempts and success rates
Project-level reporting: Bailout trends across different codebases

Phase 5: Community Features (Future)

Pattern sharing: Community-driven bailout database
Marketplace distribution: VS Code marketplace release
Enterprise features: Team monitoring and reporting
AI training feedback: Send successful interventions back to improve AI behavior

📊 Performance Impact

Typing Analysis: 2-second delay after typing stops
Analysis Throttling: Maximum 1 analysis per second per file
Memory Usage: Minimal (keeps last 10 analyses per file)
CPU Impact: Low (pattern matching only)

🤝 Contributing

Fork the repository
Create feature branch: git checkout -b feature/new-capability
Add functionality with proper error handling
Update documentation for new commands/features
Submit pull request

Development Guidelines

Follow existing TypeScript patterns
Implement robust error handling
Update documentation for new commands/features
Test with real AI assistant interactions

📄 License

MIT License - Feel free to use and modify for your projects.

🙏 Acknowledgments

Built for the developer community frustrated with AI planning instead of implementing
Research-backed approach to AI assistant quality monitoring
Inspired by the need for production-ready AI-generated code
Special recognition for identifying subagent delegation as a new form of AI avoidance

Stop the planning, start the coding! 🚀

AicodeGuard

Install / Use

README