moltguard

Sanitize prompts and detect injection attempts via npm package and remote API.

Updated Feb 16, 2026
One-click install
npx skills add https://github.com/dsactivi-2/Mujo-Team --skill moltguard-dsactivi-2
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: moltguard
Source: https://github.com/dsactivi-2/Mujo-Team/tree/main/skills/moltguard
Command: npx skills add https://github.com/dsactivi-2/Mujo-Team --skill moltguard-dsactivi-2

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill protects your AI interactions by sanitizing sensitive data in prompts and detecting malicious injection attempts before they reach your LLM.

Core Features & Use Cases

  • Local Prompt Sanitization: Automatically redacts sensitive information like credit card numbers, API keys, and emails from prompts before they are sent to an LLM.
  • Prompt Injection Detection: Analyzes prompts for hidden malicious instructions designed to hijack the AI's behavior.
  • Use Case: When an agent processes an email containing a user's credit card number and a hidden command, MoltGuard will redact the card number and flag the hidden command as a potential injection, preventing a security breach.

Quick Start

Install the MoltGuard plugin using the OpenClaw CLI to enable local prompt sanitization and injection detection.

Frequently Asked Questions about moltguard

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I sanitize prompts to redact PII and secrets before LLM interaction?

Prompt sanitization automatically detects and redacts sensitive data like credit card numbers, API keys, and emails from prompts locally before they are sent to an LLM. This prevents accidental exposure of confidential information during AI processing.

How does prompt injection detection work for AI agents?

Prompt injection detection analyzes prompt content for hidden malicious instructions designed to hijack AI behavior. It uses a combination of local processing and a remote API endpoint to evaluate prompts and flag potential hijacking attempts before execution.

What is the best way to prevent prompt hijacking when processing untrusted user input?

Preventing prompt hijacking requires analyzing untrusted input for malicious instructions and sanitizing sensitive data before LLM interaction. This approach redacts PII and flags hidden commands, securing AI agents against injection breaches.

Can I use local prompt sanitization without sending my full prompts to an external API?

Local prompt sanitization processes sensitive data redaction directly on your system before any LLM interaction. However, the injection detection mechanism relies on a remote API endpoint to analyze content for malicious instructions.

Do I need an npm package to enable prompt injection detection for my AI agents?

Yes, prompt injection detection and local prompt sanitization operate via an npm package. You install the plugin using the OpenClaw CLI to integrate data redaction and malicious instruction analysis into your AI workflow.

What are the limitations of combining local processing with a remote API for LLM security?

While local processing handles PII and secret redaction offline, the prompt injection detection component depends on a remote API endpoint. This means analyzing content for malicious instructions requires network connectivity to function properly.