Worked on the CherryHQ/cherry-studio repository to enhance vision capabilities for custom models within Agent sessions. Addressed a critical issue by adding the missing 'capabilities' field to the AgentSessionInputbar, ensuring that vision settings for custom models are now properly recognized and utilized. The solution involved a targeted refactor of the agent session assistant object creation, reducing complexity and improving maintainability. Leveraged React and TypeScript to implement and validate these changes, focusing on front end development and robust session initialization. This work improved reliability and user experience for vision-enabled workflows, while also laying the groundwork for smoother future enhancements and CI validation.