AI Detection

SnapBack monitors your coding session in real-time to detect AI tool activity and analyze code risks. This dual approach ensures protection is triggered when AI makes substantial changes or introduces common risks.

How It Works

AI Tool Detection

SnapBack detects when AI coding assistants are active and tracks their contribution to your code:

Guardian AI: Code Analysis

Separately, Guardian AI scans your code for risks that AI tools sometimes introduce:

Real-Time Analysis

Guardian AI performs real-time analysis on every save, providing immediate feedback:

Detection Speed

Both AI tool detection and Guardian AI operate in real-time:

• AI Tool Detection: Instant (passive monitoring)
• Guardian AI Scan: <50ms average
• Complex Pattern Analysis: <200ms

What Gets Detected

Secrets

Guardian AI identifies various types of secrets that should never be committed:

Copy// ❌ Detected as secret risk
const config = {
  apiKey: "sk-abcdefghijklmnopqrstuvwxyz1234567890", // API Key
  dbPassword: "supersecretpassword123", // Database Password
  jwtSecret: "my-jwt-secret-key", // JWT Secret
  awsAccessKey: "AKIAIOSFODNN7EXAMPLE", // AWS Access Key
};

Mocks

AI assistants sometimes generate test artifacts that shouldn’t be in production code:

Copy// ❌ Detected as mock risk
const userData = {
  id: 1,
  name: "John Doe",
  email: "john.doe@example.com", // Realistic but fake data
  role: "admin"
};

// This looks like mock data that shouldn't be in production
function processPayment() {
  console.log("Processing payment of $100.00"); // Mock payment
  return { success: true, transactionId: "txn_1234567890" }; // Fake transaction
}

Phantom Dependencies

Missing dependencies that are used in code but not declared in package.json:

Copy// ❌ Detected as phantom dependency risk
import { format } from 'date-fns'; // Used but not in package.json
import lodash from 'lodash'; // Used but not declared

export function formatDate(date) {
  return format(date, 'yyyy-MM-dd');
}

export function deepClone(obj) {
  return lodash.cloneDeep(obj);
}

False Positive Handling

We understand that not every detection is a real risk. Guardian AI is designed to minimize false positives while maintaining high accuracy:

Configuration

You can customize Guardian AI’s behavior to match your project’s needs:

Sensitivity Levels

Copy// .snapbackconfig
{
  "aiDetection": {
    "sensitivity": "high", // low, medium, high
    "ignorePatterns": [
      "**/*.test.js",
      "**/mocks/**"
    ],
    "customRules": [
      {
        "pattern": "API_KEY",
        "type": "secret",
        "severity": "high"
      }
    ]
  }
}

VS Code Integration

In the VS Code extension, you can:

1. View real-time detection results in the Problems panel
2. See inline decorations for detected risks
3. Configure detection settings through the UI
4. Provide feedback on detections

Accuracy Metrics

Our continuous testing shows Guardian AI maintains high accuracy across different codebases:

Best Practices

Related Documentation