Feel free to correct me, but the ML classifier appears to be rather bare. Less than 20 hardcoded payloads with randomized URL encoding as the only augmentation.

Feel free to correct me, but the ML classifier appears to be rather bare. Less than 20 hardcoded payloads with randomized URL encoding as the only augmentation. How does this generalize to novel evasion techniques? Genuinely curious what your eval numbers look like against real traffic. https://github.com/theghostshinobi/Shibuya-waf-light-version/blob/main/waf-killer/ml/training/generate_attack_dataset.py https://github.com/theghostshinobi/Shibuya-waf-light-version...