Malva

Malva is a defect library for agentic GUI software, a class of systems in which large language model (LLM) agents autonomously operate graphical user interfaces to complete user-specified tasks without relying on programmatic APIs.This dataset supports an empirical study of defects in agentic GUI applications. It contains a benchmark of 40 open-source agentic GUI projects collected from GitHub, spanning web, desktop, and mobile environments. For each project, we record metadata including repository links, commit identifiers, deployment environments, and the LLM and perception mechanisms used.The dataset also includes a curated defect library that documents defects identified in these applications. For each defect instance, the dataset records the defect type, explanation, root cause, consequences, source-code location, and defect-triggering tests. Multiple cases of the same defect type are documented to capture different manifestations of similar issues.The artifact includes two main files: (1) application.csv, which describes the benchmark applications, and (2) defect.csv, which contains the organized defect library used in the study.