阶段监督潜在推理实现单次JavaScript反混淆
原标题：Stage-Supervised Latent Reasoning for Single-Shot JavaScript Deobfuscation
AI 摘要
该论文提出一种阶段感知的潜在推理框架，用于JavaScript反混淆。方法将确定性反混淆工具产生的中间输出转化为监督信号，基于Coconut（连续思维链）范式训练模型，使其在训练时学习多阶段重写过程，但在推理时仅接收混淆代码并一次性生成清理后的程序。在JsDeObsBench上的初步结果显示，基于Coconut的模型将语法有效性提升至50%（直接微调为15%，零样本基线为25%），在有效输出中语义正确率达到80%，优于两个对比模型。
正文节选
Stage-Supervised Latent Reasoning for Single-Shot JavaScript Deobfuscation Abstract JavaScript obfuscation is widely used to protect code, but it also makes program analysis and security review substantially harder. Existing LLM-based deobfuscation methods usually treat the task as one-step translation, ignoring the staged structure of practical deobfuscation pipelines. This WIP paper proposes a stage-aware latent reasoning framework that converts intermediate outputs from a deterministic deobfu