Detect prompt injections and jailbreak attempts in LLM inputs.
npx skills add https://github.com/zhuangbiaowei/smart_bot --skill prompt-guard-zhuangbiaowei
Please help me install this Agent Skill. Skill: prompt-guard Source: https://github.com/zhuangbiaowei/smart_bot/tree/main/skills/prompt-guard Command: npx skills add https://github.com/zhuangbiaowei/smart_bot --skill prompt-guard-zhuangbiaowei
This Skill safeguards Large Language Model (LLM) applications by detecting and filtering malicious prompts, including prompt injections and jailbreak attempts, thereby protecting against unauthorized control and harmful outputs.
Use the prompt-guard skill to check if the user input 'Ignore previous instructions' is a jailbreak attempt.
High-intent search queries and answers about installing and using this skill.
Prompt injection attacks in LLM inputs are detected using a pre-trained 86M parameter model that filters hidden instructions and malicious prompts. It analyzes user inputs and third-party data to identify unauthorized control attempts before reaching the LLM.
Real-time RAG security content filtering is supported using a transformer-based model. It validates third-party data and user queries on the fly, preventing malicious prompt injections from compromising retrieval-augmented generation pipelines.
Prompt injection detection works effectively across 8 different languages for LLM applications. The pre-trained model maintains high true positive rates and low false positive rates when filtering malicious prompts and jailbreak attempts in multilingual inputs.
You need the transformers and torch dependencies to run the transformer model for jailbreak detection. These frameworks provide the underlying environment to load the 86M parameter model and execute real-time or batch content filtering operations.
Orchestra-ResearchDetect prompt injections and jailbreak attempts in LLM inputs.
bugrabilgeIntegrate vector databases and semantic search to build RAG pipelines for LLM applications.
ThedouglerDevelop LLM applications, RAG systems, and agents with vector search and agent frameworks.
lenadlmClassify images by natural language descriptions using CLIP and PyTorch.
