Deobfuscating npm Packages Before the LLM Sees Them Raises Malicious-Detection Coverage From 69% to 99%

arXiv 2609.18862·medium signalXBlueskyLinkedInCopy link

arXiv 2609.18862 (16 Sep 2026) addresses how threat actors evade LLM-based npm scanners by exploiting limited context windows, using obfuscation that yields high token density and bundling malicious code into benign packages so detectors skip large files. CASHEWS is a JavaScript preprocessor that iteratively deobfuscates, extracts bundled modules and dynamically executed code, identifies malicious sinks and computes backward slices reaching them, and abbreviates long literals and identifiers. Across 512 large package files, two scanner types and three LLMs it raises analysis coverage from 69.1-85.7% to 98.8-100% and cuts the false-negative rate by up to 18.6 percentage points, with the recent Shai-Hulud supply-chain attack cited as the motivating case.