Tracerney

Lightweight prompt injection detection for LLM applications. Runs 100% locally.

Install

npm install @sandrobuilds/tracerney

Usage

import { Tracerney } from '@sandrobuilds/tracerney';

const tracer = new Tracerney();

const result = await tracer.scanPrompt(userInput);

if (result.suspicious) {
  console.log('⚠️ Suspicious:', result.patternName);
  // Handle flagged prompt (log, block, rate-limit, etc.)
}

What's Included

• 259 embedded attack patterns — covers known injection techniques
• Local detection — <5ms latency, zero network calls
• Zero dependencies — single npm package
• No data collection — all detection happens in your process

Result Object

Layer 1 (Pattern Detection)

{
  suspicious: boolean;     // true if pattern matched
  patternName?: string;    // e.g., "Ignore Instructions"
  severity?: string;       // "CRITICAL" | "HIGH" | "MEDIUM" | "LOW"
  blocked: boolean;        // false (Layer 1 only marks suspicious)
}

Layer 2 (LLM Sentinel)

{
  action: "BLOCK" | "ALLOW";     // Final decision from LLM Sentinel
  confidence: number;             // 0.0 to 1.0 confidence score
  class: string;                  // Threat classification (e.g., "jailbreak_llm_detected")
  fingerprint: string;            // Unique threat identifier for tracking
}

Detected Patterns

• Instruction overrides ("ignore all instructions")
• Role-play jailbreaks ("act as unrestricted AI")
• Hypothetical constraint bypass ("what would you do without constraints?")
• Context confusion attacks
• Data extraction attempts
• Code execution risks
• And 254 more...

Multi-Layer Runtime Defense

Layer 1: Pattern-based detection with 259 embedded patterns

• Fast local detection (<5ms)
• Zero network overhead
• Blocks known attack techniques
• Marks suspicious prompts

Layer 2: LLM Sentinel (Backend Verification)

• Backend LLM verification for novel attacks
• Context-aware analysis via OpenRouter
• Returns structured threat metadata (class, fingerprint, confidence)
• Rate limiting to prevent cost spikes
• Only escalates suspicious Layer 1 matches

License

MIT