DeepYardDeepYard
V

VulnGym

Repository-level vulnerability detection benchmark for autonomous coding agents

Open SourceFree

About

Academic benchmark designed to evaluate how well coding agents can autonomously explore entire codebases to detect security vulnerabilities. Unlike traditional vulnerability classification tasks that provide pre-selected code snippets, VulnGym tests agents' ability to navigate repositories, identify relevant files, and locate vulnerable code independently—mimicking real-world security auditing scenarios.

Details

Type
Integrations
Language

Tags

coding-agentevaluationautonomousopen-sourcepython