Published June 7, 2025 | Version d49afdd
Software Open

Source Code for "LLMxCPG: Context-Aware Vulnerability Detection Through Code Property Graph-Guided Large Language Models"

  • 1. Qatar Computing Research Institute

Contributors

  • 1. Qatar Computing Research Institute

Description

This repository contains the source code for "LLMxCPG: Context-Aware Vulnerability Detection Through Code Property Graph-Guided Large Language Models" accepted at Usenix Security 2025, a framework for vulnerability detection using Code Property Graphs (CPG) and Large Language Models (LLM).

The core methodology involves a two-phase process:

  1. Slice Construction: An LLM generates specific queries for a Code Property Graph to extract a minimal, relevant "slice" of code that may contain a vulnerability.

  2. Vulnerability Detection: A second LLM analyzes the extracted code slice to classify it as either vulnerable or safe.

Files

llmxcpg-main.zip

Files (38.4 MB)

Name Size Download all
md5:5ca4fa759343777613e3751e232339a9
38.4 MB Preview Download

Additional details

Dates

Accepted
2025-06-06

Software

Repository URL
https://github.com/qcri/llmxcpg/
Programming language
Python
Development Status
Active