EviRepair: Evidence-Driven Agents for Reliable Vulnerability Repair
EviRepair uses evidence review among LLM agents to improve repository-level vulnerability repair on PatchEval and SecBench.
EviRepair is a multi-agent framework for repository-level automated vulnerability repair that reviews evidence before patching. Agents collaboratively collect vulnerability evidence, verify consistency, and build a reviewed model used for patch synthesis and validation. Across agent frameworks and LLMs, average resolved-rate gains are 16.37% on PatchEval and 40.33% on SecBench. Code is released on GitHub.
- Agents collect evidence, check consistency, and build a reviewed vulnerability model.
- Resolved rates rise 16.37% on PatchEval and 40.33% on SecBench.
- Tests cover multiple agent frameworks and LLMs; code is public.
From the original article
1 min read at arxiv.orgRecent advances in large language models (LLMs) have enabled agent-based approaches for repository-level automated vulnerability repair (AVR), showing promising repair capabilities. However, an important issue has received limited attention: how can we ensure that the evidence generated during the repair process is reliable enough to support accurate vulnerability understanding and patch generation?
By Yilian Shi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.