ps-fuzz

Why we included this project

Your system prompt is the part of a GenAI app that most teams never think to test, and ps-fuzz treats it as a real attack surface. It runs the prompt through dynamic, LLM-driven attacks: jailbreaks, prompt injection, RAG and vector-database poisoning, and system prompt extraction, then reports how well it held up. The tests adapt to your application's domain and configuration instead of firing generic payloads, so the results say something about your actual setup. You can drive it interactively or in batch mode from the CLI, and it supports a wide range of LLM providers, including self-hosted endpoints through Ollama. A built-in playground lets you revise the prompt and re-run the attacks until the weak spots close, so you can see the improvement before you ship.