Developed a custom configuration field for the main Config model in the nvidia-cosmos/cosmos-rl repository, enabling users to define script-specific parameters independently from global settings. This feature was implemented as a dictionary in Python, defaulting to empty, to support flexible and user-defined parameterization for experiments. The approach focused on configuration management and software design, reducing coupling between global and per-script configurations while improving reproducibility and traceability. By introducing this new field, the work streamlined parameter passing for experiments and minimized global configuration drift, providing a more maintainable and adaptable configuration structure for ongoing development and experimentation workflows.