CodeSense is a benchmark designed for fine-grained code reasoning across real-world Python, C, and Java software projects. It enables deeper semantic understanding beyond coarse input/output tasks, offering execution traces and tools to evaluate LLMs in practical software engineering contexts.