LLM-Guided Issue Generation from Uncovered Code Segments

Researchers from multiple institutions have unveiled IssueSpecter, a novel tool that leverages LLMs to automatically detect and report bugs from uncovered code segments in Python projects. By combining coverage analysis with LLM-based defect identification, IssueSpecter generates structured issue reports complete with severity ratings, reproduction steps, and suggested fixes—addressing the growing problem of AI-generated reports that lack actionability and reproducibility. In evaluations across 13 actively maintained Python projects, the tool produced 10,467 issue reports, with manual annotation of the top-130 ranked issues confirming that 84.6% were valid or warranted further investigation, against only 15.4% false positives.

IssueSpecter's LLM-based ranking outperformed rule-based ranking by 50% at P@3 and 41% in MRR, covering a wide variety of bug types from logic and boundary errors to security vulnerabilities and state consistency bugs. Compared against CoverUp, a state-of-the-art coverage-driven test generation tool, IssueSpecter achieved a higher bug validity rate (81.0% vs. 76.2%) under identical evaluation conditions while additionally providing structured reports with reproduction steps and candidate fixes that are immediately actionable. Case studies reproducing real bugs from its generated reports further validated its practical value for automatic bug discovery in open-source Python projects.