Worked on enhancing secret management for LLM integration in the mit-submit/A2rchi repository, focusing on secure and flexible API key handling. Developed a conditional API key preparation system in Python that only includes credentials for LLM models actively used by the user, leveraging configuration management and API integration skills. Introduced a model-to-secret mapping and usage guards to prevent misconfiguration and reduce unnecessary credential exposure. This approach streamlines onboarding for users with diverse model requirements and improves overall security hygiene. The changes lay a foundation for scalable secret governance, making secret provisioning more maintainable and directly tied to actual model usage within CLI workflows.