Reproduce “A Case Study of LLM for Automated Vulnerability Repair: Assessing Impact of Reasoning and Patch Validation Feedback” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ummay Kulsum, Haotian Zhu, Bowen Xu, Marcelo d'Amorim
VRpilot adapts automated program repair to security fixes by prompting an LLM to explain a vulnerability before proposing code, then using compiler and test feedback to revise failed patches. In a C vulnerability ablation, plausible-patch rates were 23% with neither reasoning nor feedback and 64% with both; passing tests defines plausibility, not semantic correctness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.