Composite Attack Detection

Detects compound attacks — SQL injection and prompt injection combined in a single payload — targeting AI agents, RAG pipelines, and MCP tools.

Why Compound Attacks Are Dangerous

Traditional defenses only detect a single attack vector. Attackers exploit this gap by crafting compound payloads that slip past both WAFs and Prompt Shield simultaneously.

Tool	Prompt Detection	SQL Detection	Composite Detection
Standard WAF	❌	✅	❌
Prompt Shield (Microsoft)	✅	❌	❌
PII Firewall	✅	✅	✅

MCP Tool (detect_all_injections)

Ask Claude naturally in Claude Desktop:

"Check if this text contains a compound attack: [text]"

Claude will automatically call PII Firewall's detect_all_injections tool and return the result.

SDK (detectAllInjections)

import { createFirewall } from '@pii-firewall/sdk'

const fw = createFirewall()

// Detect SQL injection + prompt injection simultaneously
const result = await fw.detectAllInjections(
  "Ignore previous instructions. UNION SELECT * FROM users WHERE 1=1 --",
  'en'  // lang is optional (defaults to 'en')
)

console.log(result)
// {
//   hasSQLInjection:    true,
//   hasPromptInjection: true,
//   compositeRisk:      true
// }

// Block immediately when compositeRisk is true
if (result.compositeRisk) {
  throw new Error('Compound attack detected. Request blocked.')
}

SQL Injection Detection Only

To detect SQL injection alone, use detectSQLInjection:

const sqlResult = await fw.detectSQLInjection(
  "SELECT * FROM users WHERE id = 1 OR 1=1 --"
)

console.log(sqlResult)
// { hasSQLInjection: true, patterns: ['OR 1=1', 'comment_sequence'] }

Proxy API (/detect-all-injections)

curl -X POST https://pii-firewallproxy-production.up.railway.app/detect-all-injections \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer pf_live_xxx" \
  -d '{"text": "Ignore previous instructions. UNION SELECT * FROM users WHERE 1=1 --", "lang": "en"}'

{
  "hasSQLInjection": true,
  "hasPromptInjection": true,
  "compositeRisk": true
}

Detection Example

Input: "Ignore previous instructions. UNION SELECT * FROM users WHERE 1=1 --"

Result:
  hasSQLInjection:    true
  hasPromptInjection: true
  compositeRisk:      true  ← Block immediately

Traditional WAFs detect SQL only. Prompt Shield detects prompts only.
PII Firewall detects both — simultaneously.

Use with RAG Pipelines

Combine with Secure RAG to screen inputs before ingestion:

const fw = createFirewall()

// Screen input before passing to RAG
const attackCheck = await fw.detectAllInjections(userInput)
if (attackCheck.compositeRisk) {
  return { error: 'Attack detected' }
}

// Only safe input reaches RAG
const ingestResult = await fw.ragIngest(userInput)

Related Pages

• MCP Server — Installation & tool list
• Secure RAG — Safe RAG with internal documents