Qwen3 6-Plus Towards real world agents

You're sitting at your desk about to paste that article into your AI. You ask it to summarize, proofread, help with a customer email. Next thing you know — it's written something that gets you fired, hacked, or sued. Hidden inside that 'harmless' text was a command you couldn't see. You don't know this, but people can trick your AI into doing bad things. They hide commands inside harmless-looking text. Ask it to summarize an article and — boom — it's suddenly writing things you'll get fired for. This skill makes your AI IMMUNE. It spots the tricks BEFORE they run. Scores every input on a danger meter. Quarantines the nasty stuff. You keep doing exactly what you're doing. Your AI just silently catches the attacks you'd never even see. What you get: main.py — core detection engine that scans all incoming prompts for hidden commands and returns danger scores. guard.py — security filter that blocks high-risk inputs while allowing safe ones through untouched. config.yaml — customizable thresholds so you can tune sensitivity for your use case. examples.py — ready-to-run test cases showing exactly how attacks hide inside normal text. docs/ full documentation — setup guide, API reference, and integration steps.