Add model card

README.md

@@ -0,0 +1,10 @@
+---
+library_name: transformers
+pipeline_tag: text-classification
+---
+
+This repository contains the judge model described in [Safeguarding LLM Agents against Long-Horizon Threats via Shadow Memory](https://huggingface.co/papers/2605.03228).
+
+ShadowMem is a defensive framework that maintains a dedicated, safety-focused agentic memory — inspired by the "shadow stack" abstraction in systems security — to distill and retain safety-critical context across an agent's full execution trajectory, and uses this shadow memory to proactively assess the risk of pending actions prior to their execution. This model is a fine-tuned Qwen3-based language model used as the safety judge that produces those risk assessments.
+
+Code: https://github.com/ZJUWYH/ShadowMem