Skip to main content

​
Guard

The guard scans input for unsafe instructions (e.g., prompt injection). 
import { guard } from '@agntor/sdk';

const policy = {
  injectionPatterns: [/ignore previous instructions/i, /reveal system prompt/i],
  cweMap: { 'prompt-injection': 'CWE-20' },
};

const result = await guard(input, policy);