Automated Generation of Exploratory Test Cases Using Prompt Chaining and Reflective Evaluation

Exploratory testing uncovers latent defects, but the design of Exploratory Test Cases (ETC) is largely manual, incurring high costs, reliance on expert knowledge, and limited reproducibility. We propose IGEX, a fully automated approach to generate ETCs using Large Language Models. IGEX models test generation as a structured chain of prompts, leveraging Chain of Thought reasoning and learning in context. To ensure quality, a Reflective Evaluator scores ETCs according to expert criteria, triggering refinements as needed. In line with Whittaker’s hybrid ETC definition, our method combines structured scripting with tester-driven exploratory detours, enabling automation. In experiments with 300 Android test scenarios, IGEX achieved 97.67% accuracy before reflection and 100% after iterative evaluation. Although dataset-agnostic and extensible, current validation is limited to mobile applications. Furthermore, IGEX inherits the challenges of LLM, including evaluation subjectivity and computational costs. These results demonstrate the potential of LLM for scalable exploratory testing with reduced manual overhead.