Worked on deepmodeling/abacus-develop to enhance the stability and reliability of matrix input and output operations in C++. Focused on improving error handling and memory management, the work addressed potential crashes caused by file I/O failures and invalid memory allocations. By validating matrix dimensions before allocation and ensuring consistent error paths for both reading and writing operations, the changes reduced the risk of runtime failures during production simulations. The updates required careful use of C++ features for file I/O and robust error handling, resulting in a more resilient codebase that better safeguards against edge cases in scientific computing workflows.