Reproduce “Knowdit: Agentic Smart Contract Vulnerability Detection with Auditing Knowledge Summarization” 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

Ziqiao Kong, Wanxu Xia, Chong Wang, Yue Xue, Yi Lu, Pan Li, Shaohua Li, Zong Cao and 1 more

KNOWDIT introduces a knowledge-driven, multi-agent system for smart contract vulnerability detection, which builds an auditing knowledge graph by connecting fine-grained DeFi semantics to recurring vulnerability patterns. The system achieved 100% detection of high-severity and 90.5% of medium-severity vulnerabilities on the AuditEval benchmark without false positives, and identified 45 previously unknown high/medium severity vulnerabilities in real-world projects.

Key insights

  • Recurring smart contract vulnerabilities often stem from shared 'DeFi semantics' (underlying economic mechanisms) rather than direct code reuse, suggesting a need for abstraction beyond implementation details.
  • A structured auditing knowledge graph, systematically linking abstract DeFi semantics to concrete vulnerability patterns, can effectively guide large language models (LLMs) in generating precise auditing specifications and exploits.
  • An iterative, multi-agent framework with executable Proof-of-Concept (PoC) validation, incorporating iterative drafting and guardrails, significantly improves the precision and recall of automated smart contract vulnerability detection.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.