JailbrokenAI/wallbreaker — reverse-engineered prompt

Build me a Python terminal app called wallbreaker for authorized LLM security testing only. I want it to feel like a Claude Code style agent where I can chat in a TUI or run a single automatic objective from the command line, and it should keep reasoning, using tools, and retrying attacks until it succeeds or asks me for input. Make it configurable so I can choose an attacker model, a target model, and a judge model, with support for OpenAI style and Anthropic style APIs, OpenRouter, local Claude Code, and custom base URLs with bearer auth if needed.

It should come with a strong red team toolkit, including prompt mutation and transform tools, benchmark driven testing with HarmBench style behaviors, validation that repeats runs to measure real success rate, persona and system prompt generation tools, and optional multimodal image edit attack flows. Please include session autosave and resume, a config file plus env support, and basic check commands so setup is easy. Look up current docs online if you need to.