●builderYou need to implement sanitization mechanisms in the skill-learning pipeline to prevent long-term behavioral corruption.
●policyYou must consider new security risks where agents' own learning processes can be hijacked to create permanent backdoors.