LLM Security

Audits LLM/AI application security — prompt injection and data exfiltration via tools, unsafe model/artifact deserialization (pickle, torch.load, LangChain unsafe_deserialization), hardcoded LLM API keys (OpenAI, Anthropic, Google, Groq, Hugging Face), over-powered agent tools (shell/REPL), prompt/PII logging, and telemetry capturing prompts (LangSmith, W&B). Use when auditing LangChain/LlamaIndex/AutoGen apps, OpenAI/Anthropic integrations, RAG pipelines, chatbots, or repos containing model files.

jiayaoqijia Updated 81 repo stars

File contents

jiayaoqijia/cryptoskill commit 4b7b2d3179

Frequently asked questions

npx get-skill add jiayaoqijia/llm-security