prompt-guard

Detect jailbreak and prompt-injection attempts in LLM applications.

Updated Apr 3, 2026

Search Tags:#pytorch #multilingual #jailbreak #transformers #prompt-injection #llm-security #content-filtering

Ready to run

npx skills add https://github.com/handsomelong922/my-codex-skills --skill prompt-guard-handsomelong922

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: prompt-guard
Source: https://github.com/handsomelong922/my-codex-skills/tree/main/skills/prompt-guard
Command: npx skills add https://github.com/handsomelong922/my-codex-skills --skill prompt-guard-handsomelong922

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires transformers, torch.

What problem does it solve?

Filters malicious prompts and third-party data for LLM apps by detecting jailbreak attempts and prompt injections in real time, helping teams reduce leakage, manipulation, and unsafe outputs.

Core Features & Use Cases

Lightweight 86M-parameter classifier with fast inference for chat interfaces and RAG pipelines.
Multilingual support across 8 languages to cover global user bases.
Deployable via HuggingFace and batch processing workflows for scalable security checks.

Quick Start

Run the prompt-guard detector on an input to evaluate jailbreak risk.

Frequently Asked Questions about prompt-guard

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

FAQPage Schema

How do I detect prompt injection attempts in my LLM application?▼

You can detect prompt injection attempts by running user inputs or retrieved data through a lightweight 86M-parameter classifier built with transformers and PyTorch to evaluate jailbreak risk in real time.

What is the best way to filter jailbreak prompts in a RAG pipeline?▼

Filtering jailbreak prompts in a RAG pipeline is best handled by applying a real-time content classifier to scan retrieved documents and user queries before they reach the LLM, preventing manipulation and unsafe outputs.

Does this prompt injection detector support multilingual inputs?▼

Yes, the prompt injection detector supports multilingual inputs across 8 languages, allowing you to secure global user bases interacting with your chat interfaces or API endpoints.

Can I use transformers and PyTorch to secure chat interfaces against jailbreaks?▼

Yes, you can deploy an 86M-parameter PyTorch and transformers model via HuggingFace to secure chat interfaces, achieving low latency inference to block malicious prompts before processing.

When do I need real-time LLM security and content filtering?▼

You need real-time LLM security and content filtering when your application ingests third-party data or user chats, requiring immediate detection of jailbreak attempts to reduce data leakage and manipulation.