VulnSeeker explores how static program analysis and LLM reasoning can be combined for automated security assessment.

This work was conducted at the Fudan Software Engineering (CodeWisdom) Lab under the guidance of Lecturer Yi Li, in collaboration with Qicai Chen, a Ph.D. student.

Key components:

  • CodeQL-based vulnerability detection;
  • automated expansion of relevant external function context;
  • prompt workflows for CodeQL and LLM interaction;
  • structured security judgement states;
  • combined assessment from static rules and model classification.

The source code is publicly available for exploring the analysis workflow.