Ask HN: Are there good security benchmarks for LLMs?

A discussion is being sought on security benchmarks for large language models. The person initiating the discussion is new to benchmarking LLMs and is looking for a benchmark that involves an agent scanning a repository. They found one potential option, eyeballvull, but it lacks wide support. The goal is to identify other existing benchmarks that may not be well-known. The discussion also touches on security-aware software engineering with and without agents.