Extended the list of dangerous keywords in the _partially_shield_content method to cover more attack vectors, including system commands, privilege escalation, information leakage, and social engineering. This enhances the robustness of the message shielding mechanism against prompt injection and related attacks.