Patterns & rules¶

Built-in categories: - Prompt injection patterns - PII leakage patterns - Toxic content patterns - Malicious URL patterns

Files live in src/security/patterns/.

Add a new pattern¶

import { PromptInjectionDetector } from "resk-llm-ts";

const detector = new PromptInjectionDetector({ enabled: true, level: 'advanced' });
detector.addCustomPattern('high', 'leakage', /\b(?:\d{1,3}\.){3}\d{1,3}\b/);

Vector similarity corpus¶

Combine regex patterns with semantic similarity. A minimal corpus is exported from src/security/patterns/llm_injection_patterns.ts as canonicalInjectionCorpus.

import { ReskLLMClient } from 'resk-llm-ts';
import { canonicalInjectionCorpus } from '../src/security/patterns/llm_injection_patterns';

const client = new ReskLLMClient({ provider: 'openai', providerConfig: { apiKey: '...' } });
await client.seedInjectionCorpus(canonicalInjectionCorpus, { source: 'builtin' });