Worked on the mlcommons/inference repository to enhance the reliability of MLPerf benchmarking scripts using Shell scripting. Addressed a bug in the reference scripts by refining flag handling, ensuring the --accuracy flag is applied exclusively to the accuracy script and removed from the performance script. This adjustment prevents misinterpretation of command-line arguments and helps maintain the integrity of benchmarking results. Additionally, corrected minor typographical errors in both reference_mlperf_perf.sh and reference_mlperf_accuracy.sh, further improving script stability. The work focused on Shell scripting and command-line interface management, aligning the scripts with MLPerf standards and reducing troubleshooting time for evaluators.