Worked on the FoundationAgents/OpenManus repository to deliver a feature optimizing token counting logic, directly improving efficiency in token usage and downstream performance. Focused on simplifying calculations based on image dimensions and detail level, the approach emphasized code refactoring and cleanup to enhance maintainability and readability. By removing redundant comments and clarifying logic paths, the work ensured that default token calculations are handled efficiently, supporting better cost control and reduced processing time. The project relied on Python and demonstrated skills in performance optimization and code hygiene, resulting in a more maintainable and efficient token handling module for the OpenManus workflow.