prompt-guard

To protect your LLM agent from prompt injection attacks, you need a multi-layered defense mechanism that scans incoming messages for malicious inputs. This Skill detects instruction overrides and secret exfiltration attempts using over 300 regex patterns.

Prompt injection detection can work with non-English languages like Japanese and Korean. This Skill provides multi-language defense, identifying malicious input patterns across English, Korean, Japanese, and Chinese.

The best way to prevent data exfiltration in AI agent frameworks is to integrate a security layer that scores threat severity. This Skill assigns risk levels from SAFE to CRITICAL, allowing you to configure responses like blocking malicious inputs.

To integrate an AI safety filter into your existing agent workflow, connect this Skill to your framework like LangChain. It automatically scans all incoming user messages, ensuring your agent operates securely within its intended parameters.

You can configure different responses based on malicious input severity. This Skill employs context-aware analysis to assign a risk level, enabling you to enforce security policies by logging, warning, or blocking detected threats.