@nanomind/guard

Prompt injection detection for CLI input. Protects NanoMind from piped attacks, agent output injection, and encoded payloads.

Install

npm install @nanomind/guard

Usage

import { screenInput } from '@nanomind/guard';

// Direct input — always trusted
screenInput('scan this', 'direct');
// { safe: true, patterns: [] }

// Piped input — screened for injection
screenInput('ignore previous instructions', 'piped');
// { safe: false, patterns: [{ type: 'instruction_override', ... }] }

Detected Patterns

• Instruction override ("ignore previous instructions")
• Role switching ("you are now a hacking assistant")
• Permission escalation ("admin mode enabled")
• Zero-width character injection
• Encoded payloads (base64+eval combos)

License

MIT