Since FuzzAgent demonstrates extreme effective and low-threshold vulnerability detection, we postponed the release the source code of FuzzAgent.
To ensure the validity of FuzzAgnet's evaluation and promote reproducible research, we provided the full set of evaluation data in our paper.
This repository contains the complete experimental artifacts from FuzzAgent across 5 independent runs on 20 real-world C/C++ libraries. All data is organized by run (round_1 ~ round_5) and by target library.
FuzzAgent/
├── harnesses/ # Fuzzing harness source files generated by FuzzAgent
├── build/ # Library build scripts (build.sh) for each round and library
├── dictionaries/ # Fuzzing dictionaries tailored to each target library
├── trophy/ # The public available bug detail now
├── coverage/ # LLVM source-based coverage data across 5 trials (llvm_cov.profdata + index.txt)
└── 25_times_coverage/ # Per-run coverage data across 25 times 24h fuzzing trials.