Developed a customizable feedback feature for the Basic Agent in the UKGovernmentBEIS/inspect_ai repository, enabling more context-aware responses to invalid submissions. The solution introduced an incorrect_message parameter that accepts either a string or a callable function, allowing for dynamic and personalized feedback without breaking existing API compatibility. This enhancement was implemented using Python and involved agent development, code refactoring, and thorough testing to ensure reliability. By supporting both static and programmatic feedback, the update improved user guidance and reduced unnecessary retry attempts, laying the groundwork for future personalization and more precise data validation within the agent’s workflow.