Conceptual

LLM-Tailored Prompt Optimization for Software Test Case Generation

A method for automatically discovering a distinct, per-model prompt that maximizes the quality of unit tests a large language model generates for given source code. Students learn how iterative prompt search is made effective for this task by three cooperating ideas: forcing diverse modification directions to escape local optima, mining clusters of failed-test errors into validated rules that prevent recurring mistakes, and injecting cross-file code context so the model instantiates classes and passes arguments correctly.