fir3storm/ANI — reverse-engineered prompt

Build me a tool called ANI for authorized security testing of AI chat websites. I want it to help me check how resistant a chat interface is to prompt injection, jailbreaks, hidden prompt leaks, data exfiltration, RAG style document poisoning, tool abuse, and multi turn social engineering tricks.

It should work in two simple ways. One is a Firefox sidebar that runs inside my existing logged in browser session so I can test real chat apps without rebuilding login flows. The other is a Python command line app for repeatable scans, saved sessions, baseline versus diff checks, and reports I can use in automation. I want adaptive testing where an LLM can look at each response and decide the next attack, plus an offline rules only mode. Keep payloads editable in separate JSON files, detect common chat UI elements even in tricky page structures, and save auth profiles and sessions encrypted locally. The output should give clear evidence, a verdict, a risk score, and exportable HTML, JSON, and SARIF reports. Look up current docs online if you need to.