Skip to content
InfoResearchPreprintLLM-specific

SCARVA: Interpretable Agentic Reasoning in Software Supply-Chain Analysis

Published
Record updated
View JSON

Summary

SCARVA is a framework for interpretable LLM-agent analysis of software supply-chain vulnerabilities. An ANALYST agent assesses dependencies while every tool call, file access and web request is recorded, and independent JUDGE agents check the result and produce an evidence-backed report. On 54 curated cases, the best ANALYST model, Qwen 3.8, reached 82% balanced accuracy, and human annotation found Claude Opus 5.5 agreed with annotators on 94% of JUDGE claim checks.

Topics