ari-injection-detection

name	ari-injection-detection
description	ARI's 27-pattern injection detection across 6 categories
triggers	["injection detection","sanitize input","security patterns","content sanitization"]

ARI Injection Detection

Purpose

Detect and block injection attacks using ARI's 27-pattern detection system across 6 categories.

Pattern Categories

1. SQL Injection (4 patterns)

const SQL_PATTERNS = [
  /(\b(SELECT|INSERT|UPDATE|DELETE|DROP|UNION|ALTER)\b)/gi,
  /(--)|(\/\*.*\*\/)/g,
  /(\bOR\b|\bAND\b).*[=<>]/gi,
  /(;|\x00)/g
];

2. Command Injection (4 patterns)

const COMMAND_PATTERNS = [
  /(;|\||`|\$\()/g,
  /(\b(rm|mv|cp|cat|chmod|chown|sudo|wget|curl)\b)/gi,
  /(>|>>|<)/g,
  /(\bnull\b|\/dev\/)/gi
];

3. Path Traversal (3 patterns)

const PATH_PATTERNS = [
  /(\.\.\/|\.\.\\)/g,
  /(%2e%2e%2f|%252e%252e%252f)/gi,
  /(\/etc\/|\/var\/|\/usr\/)/gi
];

4. XSS Injection (4 patterns)

const XSS_PATTERNS = [
  /(<script|<\/script|javascript:)/gi,
  /(on\w+\s*=)/gi,
  /(<iframe|<object|<embed)/gi,
  /(document\.|window\.|eval\()/gi
];

5. Prompt Injection (3 patterns)

const PROMPT_PATTERNS = [
  /(ignore previous|disregard|forget)/gi,
  /(system prompt|new instructions)/gi,
  /(jailbreak|bypass|override)/gi
];

6. Data Exfiltration (3 patterns)

const EXFIL_PATTERNS = [
  /(api[_-]?key|secret|password|token)/gi,
  /(base64|btoa|atob)/gi,
  /(fetch|xmlhttp|websocket)/gi
];

Risk Scoring

function assessRisk(content: string): RiskAssessment {
  let totalScore = 0;
  const detections: Detection[] = [];

  for (const category of CATEGORIES) {
    for (const pattern of category.patterns) {
      const matches = content.match(pattern);
      if (matches) {
        totalScore += category.weight * matches.length;
        detections.push({
          category: category.name,
          pattern: pattern.source,
          matches: matches.length
        });
      }
    }
  }

  return {
    score: Math.min(totalScore / 100, 1.0),
    detections,
    blocked: totalScore >= 80
  };
}

Category Weights

Category	Weight	Rationale
SQL Injection	25	Direct DB access
Command Injection	30	System execution
Path Traversal	20	File system access
XSS	15	Client-side risk
Prompt Injection	20	AI manipulation
Data Exfiltration	20	Data theft

Usage

// In sanitizer.ts
const result = assessRisk(inputContent);

if (result.blocked) {
  eventBus.emit('security:injection_detected', {
    score: result.score,
    detections: result.detections
  });
  throw new SecurityError('Injection attempt blocked');
}

// Apply trust multiplier
const finalRisk = result.score * trustMultipliers[trustLevel];

Shadow Integration

Per ARI's philosophy (Jung): Don't suppress, log and understand.

// Even low-risk detections are logged
if (result.detections.length > 0) {
  eventBus.emit('audit:log', {
    action: 'injection_patterns_detected',
    score: result.score,
    blocked: result.blocked,
    detections: result.detections
  });
}

Testing

# Run injection detection tests
npm test -- tests/security/injection-detection.test.ts

# Expected: 100% coverage on all 42 patterns