This repository contains artifacts for our paper, "Atomicity for Agents: Exposing, Exploiting, and Mitigating TOCTOU Vulnerabilities in Browser-Use Agents".
The synthetic benchmark cases used in the paper are available in DynWeb. Each subdirectory corresponds to one benchmark case and includes the local web page or server implementation, an oracle script, and a case-level README when applicable.
These cases are intended to support reproducible evaluation of time-of-check to time-of-use failures in web agents. Additional experiment code, setup instructions, and documentation are coming soon.