1.0.0

🛡️ AI Code Guard Pro

Industry-grade security scanner for AI-generated code with AST analysis, taint tracking, and LLM-specific vulnerability detection.

AI coding assistants (GitHub Copilot, Claude, ChatGPT, Cursor) are revolutionizing development—but they can introduce security vulnerabilities that slip past code review. AI Code Guard Pro is a next-generation security scanner specifically designed to catch these issues.

🚀 Key Improvements Over Basic Scanners

🎯 What It Detects

🔐 Secrets & Credentials

• API Keys: OpenAI, Anthropic, AWS, GCP, GitHub, Stripe, and 15+ providers
• Private Keys: RSA, SSH, PGP, EC
• Database Credentials: Connection strings, passwords
• High-Entropy Strings: AI placeholder secrets

💉 Injection Vulnerabilities

• SQL Injection: f-strings, .format(), concatenation in queries
• Command Injection: os.system, subprocess with shell=True
• Code Execution: eval(), exec() with user input
• SSRF: User-controlled URLs in requests

🤖 AI/LLM-Specific Issues

• Direct Prompt Injection: User input in system prompts
• Indirect Injection: RAG/retrieval injection risks
• Unsafe Deserialization: pickle, yaml.load without SafeLoader

📦 Supply Chain Attacks

• Typosquatting: Similar names to popular packages
• Dependency Confusion: Internal package name patterns
• Known Malicious Packages: Database of suspicious packages

📦 Installation

pip install ai-code-guard

Or with development dependencies:

pip install ai-code-guard[dev]

🔧 Quick Start

# Scan a directory
ai-code-guard scan ./src

# Scan with specific output format
ai-code-guard scan ./src --format sarif -o results.sarif

# Quick CI check
ai-code-guard check ./src

# List all rules
ai-code-guard rules

# Create config file
ai-code-guard init

📊 Example Output

🛡️  AI Code Guard Pro v1.0.0
   Scanning ./my-project...

┌─────────────────────────────────────────────────────────────────────┐
│ 🔴 CRITICAL: SQL Injection Vulnerability                            │
├─────────────────────────────────────────────────────────────────────┤
│ 📍 src/db/queries.py:42                                             │
│                                                                     │
│ SQL query constructed using f-string interpolation. User-controlled │
│ data may be interpolated directly into the query, enabling SQL      │
│ injection attacks.                                                  │
│                                                                     │
│ Code: query = f"SELECT * FROM users WHERE id = {user_id}"          │
│                                                                     │
│ ✅ Fix: Use parameterized queries:                                  │
│    cursor.execute('SELECT * FROM users WHERE id = ?', (user_id,))  │
│                                                                     │
│ CWE: CWE-89                                                         │
└─────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────┐
│ 🟠 HIGH: Prompt Injection Vulnerability                             │
├─────────────────────────────────────────────────────────────────────┤
│ 📍 src/api/chat.py:23                                               │
│                                                                     │
│ User input directly embedded in LLM prompt via f-string. Attackers  │
│ can inject malicious instructions to manipulate the AI's behavior.  │
│                                                                     │
│ Code: prompt = f"You are a helper. User says: {user_input}"        │
│                                                                     │
│ ✅ Fix:                                                              │
│ 1. Separate system prompts from user content using message roles    │
│ 2. Sanitize user input (remove control characters, limit length)    │
│ 3. Use structured output formats to detect injection attempts       │
│                                                                     │
│ CWE: CWE-74 | OWASP: LLM01                                          │
└─────────────────────────────────────────────────────────────────────┘

────────────────────────────────────────────────────────────────────────
📊 SUMMARY
────────────────────────────────────────────────────────────────────────
Files scanned    47
Issues found     3
Scan time        127ms

🔴 CRITICAL: 1  🟠 HIGH: 2  🟡 MEDIUM: 0  🔵 LOW: 0
────────────────────────────────────────────────────────────────────────

⚙️ Configuration

Create .ai-code-guard.yaml in your project root:

# Minimum severity to report
min_severity: low  # critical, high, medium, low, info

# Patterns to ignore
ignore:
  - "tests/**"
  - "**/test_*.py"
  - "examples/**"
  - "docs/**"

# Rules to disable
disable_rules: []
  # - "SEC001"  # If using example API keys
  # - "PRI001"  # If false positives on prompt construction

# Secret detection tuning
entropy_threshold: 4.5  # Shannon entropy threshold
min_secret_length: 16

# AI-specific detection
detect_placeholder_secrets: true
detect_prompt_injection: true

# Performance
max_file_size_kb: 1024
parallel_workers: 4

🔌 CI/CD Integration

GitHub Actions

name: Security Scan

on: [push, pull_request]

jobs:
  security:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      
      - uses: actions/setup-python@v5
        with:
          python-version: '3.11'
      
      - run: pip install ai-code-guard
      
      - name: Run security scan
        run: ai-code-guard scan . --format sarif -o results.sarif --fail-on high
      
      - name: Upload SARIF to GitHub Security
        uses: github/codeql-action/upload-sarif@v2
        if: always()
        with:
          sarif_file: results.sarif

Pre-commit Hook

# .pre-commit-config.yaml
repos:
  - repo: local
    hooks:
      - id: ai-code-guard
        name: AI Code Guard Security Scan
        entry: ai-code-guard check
        language: python
        types: [python]
        pass_filenames: false

GitLab CI

security-scan:
  image: python:3.11
  script:
    - pip install ai-code-guard
    - ai-code-guard scan . --format json -o gl-sast-report.json
  artifacts:
    reports:
      sast: gl-sast-report.json

📋 Rule Reference

🔬 Technical Details

AST-Based Analysis

Unlike regex-based scanners, AI Code Guard Pro parses Python code into an Abstract Syntax Tree, enabling:

• Taint tracking: Follow user input through variable assignments
• Context awareness: Understand function calls and their arguments
• Reduced false positives: Skip patterns in comments and strings

Entropy-Based Secret Detection

Uses Shannon entropy to distinguish real secrets from placeholders:

# High entropy (likely real secret) - DETECTED
api_key = "sk-proj-aB3xK9mL2pQrStUvWxYz..."

# Low entropy (placeholder) - IGNORED
api_key = "your-api-key-here"

LLM Security Focus

Specifically targets vulnerabilities in AI/LLM applications:

• Detects prompt injection in OpenAI, Anthropic, and LangChain code
• Identifies indirect injection risks in RAG pipelines
• Flags unsafe patterns in agent/tool implementations

🤝 Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

Adding Detection Patterns

# ai_code_guard_pro/analyzers/my_analyzer.py
from ai_code_guard_pro.models import Finding, Severity, Category

class MyAnalyzer:
    def analyze(self) -> list[Finding]:
        findings = []
        # Your detection logic
        return findings

📚 References

• OWASP Top 10 for LLM Applications
• CWE - Common Weakness Enumeration
• SARIF Specification

📄 License

MIT License - see LICENSE for details.

---

Built for the AI era by security engineers who use AI coding assistants daily 🛡️