Wen Liu

The paper introduces STRIATUM-CTF, a modular agentic framework that uses a standardized context protocol to enable LLMs to perform multi-step, stateful reasoning for general-purpose CTF solving, achieving first place in a live competition.

The paper introduces CanaryRAG, a novel dual-path runtime defense mechanism that detects RAG Knowledge Base Leakage attacks by embedding canary tokens into retrieved knowledge chunks.

The paper analyzes Ethereum builder transactions to show that builder centralization is an emergent property of the Proposer-Builder Separation (PBS) architecture, driven by specific order flow and MEV patterns.

The paper introduces a diagnostic-driven iterative refinement process for improving LLM-generated reward functions in sparse, structured reinforcement learning tasks, significantly boosting agent performance.

The paper distinguishes between a model's ability to generate useful updates for external agent components (harness-updating) and its ability to benefit from those updates (harness-benefit), finding that updating capabilities are surprisingly uniform while benefit is maximized in mid-tier models.

DynaTree introduces a two-stage framework that pre-constructs a reusable retrieval tree offline using coordinated agents, allowing for efficient, structure-aware, and highly effective time-sensitive news retrieval online.

SkillPager is a novel two-stage framework that efficiently selects minimal, execution-sufficient context from large procedural skill documents by leveraging typed semantic nodes, significantly reducing token usage while maintaining high context sufficiency for LLM agents.

Adaptive Auto-Harness introduces a framework that enables LLM agents to sustain self-improvement and maintain high performance over open-ended, shifting task streams, outperforming existing fixed-benchmark auto-harness systems.

The paper introduces and analyzes cross-session stored prompt injection, demonstrating that persistent system state transforms prompt injection from a temporary model-level threat into a long-lived, system-level vulnerability in agentic systems.

This paper proposes a four-layer technical architecture for large model inference optimization, including Multi-model Fusion, Model Optimization, Compute-Model Fusion, and Compute-Network-Model Fusion.

xDSM is a full-space, elastic DSM system built over CXL that transparently scales unmodified multithreaded applications by employing an OS-runtime co-design, dynamic data placement policy, and spatial locality-aware elasticity.