Datasets:

bbkdevops
/

cass-vibe-security-bench

license: apache-2.0
pretty_name: CASS-Vibe Security Bench
language:
  - en
task_categories:
  - text-classification
tags:
  - security
  - code-repair
  - calibration
  - benchmark
  - datasets
configs:
  - config_name: triples
    data_files: data/triples.jsonl
  - config_name: truth20
    data_files: data/truth20.jsonl
  - config_name: diversevul100
    data_files: data/diversevul100.json

CASS-Vibe Security Bench

19 teacher triples + 20 human-labeled rows + 100 real-CVE rows with reports.

from datasets import load_dataset
ds = load_dataset("bbkdevops/cass-vibe-security-bench", "truth20")

Reproduce: python eval.py (needs the repo's cass/ package for scanners; numbers must match EVIDENCE.md scoreboard).