Paper | Leaderboard
SusVibes is a benchmark and evaluation pipeline for measuring the functional correctness and security of AI-agent-generated code on real-world software engineering tasks.
This Hugging Face dataset hosts the source files used by the Harbor adapter for the SusVibes test split:
susvibes_dataset.jsonl: task records, metadata, Docker image names, prompts, oracle patches, and hidden security test patches.
components.json: per-task evaluation metadata… See the full description on the dataset page:
https://huggingface.co/datasets/dqwang122/susvibes.
