A framework to evaluate the generalization capability of safety alignment for LLMs - View it on GitHub