
How I built vexproof: using an LLM for security triage without trusting its answers
A case study of vexproof: combining static analysis, Muse Spark, and a deterministic verifier to triage Python dependency alerts, then evaluating the results against the running project.