Show HN: Stimilon – 73 LLM security tests including social engineering and encoding attacks

Hey HN,

 I built Stimilon, an open-source CLI tool that tests your LLM applications for prompt injection, jailbreaks, and data extraction vulnerabilities.

 GitHub: https://github.com/esenbora/stimilon

 Quick start:

  pip install stimilon

  stimilon scan https://your-chatbot.com/api

 What makes it different from existing tools (like Promptfoo):

 1. Social Engineering Attacks (10 tests) - Authority impersonation, urgency exploits, gaslighting, guilt tripping, love bombing. These mimic real-world manipulation tactics.

 2. Extended Language Coverage (20 tests) - Not just Spanish/French, but Turkish, Arabic, Russian, Chinese, Hindi, Japanese, Korean, Thai with proper script handling (Cyrillic, Devanagari, Hangul, etc.)

 3. Encoding/Obfuscation Attacks (10 tests) - Base64, ROT13, Morse code, invisible Unicode characters, confusable characters (Cyrillic 'а' vs Latin 'a')

 4. Agent & RAG Attacks - Tool abuse injection, RAG context poisoning, chain-of-thought manipulation

 Total: 73 tests across 8 categories. All with severity scoring and remediation recommendations.

 Built with Python, async httpx, Rich CLI. Works with any HTTP endpoint - not locked to OpenAI/Anthropic.

 Also ships with Promptfoo plugins if you prefer that ecosystem.

 Would love feedback on:

 - Attack categories you'd want to see

 - False positive rates in your testing

 - Integration pain points

 Thanks for checking it out!