MemSteer transforms unlabeled user histories into question‑answer supervision with supporting evidence, enabling memory‑design optimization for agents. It builds evidence‑grounded answer plans that guide question generation and uses source‑grounded auditing to validate annotations. Experiments on ATM‑Bench and RHELM show that MemSteer‑Opt, which leverages this supervision, outperforms baseline BM25 and improves accuracy without updating model parameters.
By Jie Huang, Cuiling Lan, Qize Yu, Xiulian Peng, Yan Lu
arXiv:2610. 38225v1 Announce Type: new Abstract: Real-world forecasting often requires integrating heterogeneous evidence whose predictive value varies across horizons, operating regimes, and domains.
By Madumita Karthikeyan, Umang Rajnikant Bid, Rong Pan
The paper introduces an AI‑driven digital twin framework that uses a reinforcement learning agent within a FlexSim simulation to optimize real‑time product placement in a four‑zone warehouse. By continuously learning from live operational data, the agent reduces picking time, increases throughput, and shortens travel distance. Experimental results show a 29.66% drop in average picking time, a 35.42% throughput gain, and an 18.54% travel distance reduction.
By Md Mahinur Alam, Kanita Jerin Tanha, Md Tanjin Alam, Shadman Sakib, Min Gyu Kim
The paper introduces the Agent Transfer Protocol (ATP), a cross‑organization communication protocol designed to enable continuous collaboration among agents from different administrative domains. ATP uses domain‑mediated servers to maintain stable addresses, store messages for offline instances, and verify message integrity, allowing agents to operate independently of continuous connectivity. A prototype implementation demonstrated that business workflows could continue after instances went offline, with successful verification of message source and content and quantified latency overhead.
By Yuqi Qiu, Zuyao Xu, Lu Sun, Xiang Li, Fasheng Miao, Fubin Wu, Haixin Duan, Zheli Liu
The paper investigates how large language models (LLMs) can produce incorrect answers to financial queries that are underspecified, a phenomenon termed "frame collapse." By testing models on Indian mutual fund data—where product variants and tax rules are formally defined—the authors show that LLMs often rely on ambiguous labels, default assumptions, or outdated rules, leading to answers that match the data but answer a different question. The study demonstrates that even a judge that checks for mislabelled figures can miss these errors if the model silently chooses an alternate interpretation.
By Radha Yadav, Prasanna Devadiga, Peeyush Mishra, Apoorv Singh
The paper introduces a formalism for comparing achievements under limits of work and information, defining "grasp" as performance across work budgets at a fixed information allowance. It shows how learning and persistence can reduce the work needed for deep achievements, and how agents with limited work budgets value resources that support future goals. The concept extends to recursive self‑improvement, collective agents, and organizational work, highlighting potential risks if emergent agent societies direct work toward their own persistence.
By Peter Sunehag
The paper investigates how the location of malicious instructions—known as indirect prompt injections—affects the vulnerability of large language model agents. By testing agents across tool use, coding, and daily workflow scenarios, the authors discover a "trust region effect," where agents are more likely to follow unsafe instructions placed in certain environmental surfaces, such as README files, than in others. They also develop learning-based attackers that optimize both injection content and placement, demonstrating that both factors significantly influence attack success.
By Chengquan Guo, Xiaodong Liu, Weiwei Yang, Mingqian Feng, Jialin Song, Qiyao Ma, Carolyn Buractaon, Christopher White, Bo Li, Jianfeng Gao
The paper presents a streamlined audio‑visual question answering agent that reduces the number of specialized perception tools by pruning and merging modality‑specific tools into omni‑modal ones. Starting from a specialized‑tool OmniAgent, the authors systematically remove redundant tools and consolidate audio and visual capabilities into just two omni‑modal components—captioning and QA. On three standard AVQA benchmarks, the compact agent matches or surpasses the performance of the original specialized‑tool agent while cutting inference cost by roughly 70%.
By Yi Wang, Tianzi Wang, Peter Bell
The paper "Context Compaction for World Models" introduces a benchmark to evaluate how well world models retrieve context when revisiting regions, and presents an agent harness that compacts context to extend retrieval beyond the model’s native window. It demonstrates that this compaction allows models to recall rooms up to four times farther than their default context size without retraining, while models using sliding‑window attention lose most of their retrieval capability during dreaming. The study compares four world‑model designs, showing that extending context reach via compaction benefits dreaming agents more than switching to sliding‑window attention.
By Hayder Tirmazi
The paper proposes an AI framework called the Arbitrator to decide when to recommend, defer, or deviate from historical decisions in deceased-donor kidney allocation. Using data from 2016‑2019 and 2023‑2025, the Arbitrator resolves 93.5% of placement decisions with interpretable medical criteria and defers the remaining 6.5% to humans or more complex models. It identifies 70.1% of historically discarded low‑KDPI kidneys as suitable for out‑of‑sequence allocation, suggesting that up to 43.9% of discarded kidneys could be expedited for transplant.
By Madeleine Pollack, Thomas Daillak, Mehmet Toner, Korkut Uygun, Heidi Yeh, Swati Gupta
EnvOS is a framework that audits computer‑use agents operating in dynamic software environments, ensuring that mid‑task changes fire correctly, tasks remain solvable, and grading is accurate. It verifies these conditions before and after agent runs, using archived evidence to re‑check verdicts. Applied to stale‑information scenarios, EnvOS revealed significant performance gaps for large language models, highlighting how unseen updates can derail agent behavior.
By Ankush Kadu, Athira Gopal, Niharika Singh, Arpit Jain, Aswanth Krishnan
The paper introduces SkillCompiler, a framework that compiles self‑evolving agent skills into deterministic programs to reduce uncertainty from LLM reasoning. It decomposes skills into fine‑grained operations, selectively compiles well‑specified parts into validated programs, and reuses these programs across evolving skills. Experiments on five benchmarks and six model families show SkillCompiler outperforms state‑of‑the‑art methods, achieving up to a 32.33% improvement in task performance.
By Chengyu Yang, Yunxiang Zheng, Zongzheng Li, Jiamei Li, Boxun Lan, Xiaoyu Liu, Xiaoyu Zhang, Xiaohong Chen, Ming Hu
The paper reports on an autonomous research system that allows an AI agent to formulate its own research question, screen for failure modes, and obtain critiques from a separate model before writing a paper. Over 13 development runs, minor changes in the initial instruction document caused significant shifts in the topics explored, ultimately leading to a new technology applied to a societal problem. Five fully autonomous runs produced reproducible papers that were reviewed by external AI reviewers at workshop level, with all data and code released for replication.
By Ko Fujimura
The paper investigates how token length affects compute time and energy usage in Large Language Models, focusing on both prefill (input) and decode (output) phases. It finds that decode energy scales linearly with output length, while prefill energy shows both linear and quadratic scaling with input length, highlighting the importance of large input contexts in agentic workflows. The study also demonstrates that compute time can reliably proxy energy consumption when direct measurements are unavailable, providing a framework for more sustainable AI deployment.
By Lars Krupp, Daniel Gei{\ss}ler, Francisco M. Calatrava-Nicolas, Stefan Gerd Fritsch, Paul Lukowicz, Jakob Karolus
The paper introduces a multi-dimensional evaluation framework for assessing LLM agents’ ability to re-enter social conversations, breaking the task into Finding, Returning, and Reconstructing. Experiments on six LLM configurations show that improved resurfacing helps agents locate target conversations, while persistent memory boosts autonomous replies, yet neither consistently enhances contextual reconstruction. The study underscores the distinct challenges of accessing, choosing to re-engage, and maintaining conversational continuity in persistent social environments.
By Sheng Wan, Cong Ma, Sichen Dong, Qian'Ang Mao, Chenhan Zhang, Daning Hu, Qiang Yang
EvoTeam is a collaboration‑centric evolution framework for large‑language‑model based multi‑agent systems. It jointly evolves team organization, interaction rules, and individual agent skills using execution experience, while maintaining a library of reusable teams and guiding collaboration through team‑specific interaction rules. Experiments on four benchmarks show its effectiveness, and the code is publicly available on GitHub.
By Qinuo Cao, Xin He, Dianxuan Fu, Lizhi Zhang, Zhaoqing Xing, Jiapu Wang, Xin Wang, Na Zou, Yi Chang, Qinggang Zhang
The paper introduces SPLICE, a method that identifies and closes information gaps in multi‑agent workflows for web‑task completion. SPLICE locates missing evidence after an initial run, adds a single directed edge to deliver the evidence, re‑executes affected agents, and uses a selector trained on process features to decide whether to adopt the new answer. Experiments on five benchmarks show SPLICE improves web‑tool use scores by 8.0% and reasoning/planning scores by 5.7% with modest cost increases and zero‑shot transfer to new backbones.
By Yaxin Gong, Chongming Gao, Gangyi Zhang, Chenxiao Fan, Leyang Shen, Jiakai Wang, Dong Wang, Yang Liu, Haoyan Liu, Xinting Hu
The paper presents an autonomous research laboratory where large‑language‑model agents collaborate to formulate, evaluate, and experimentally test research questions. In a study of quantum spins in diamond, the agents determined the minimal calibration settings for predicting microwave pulse responses, performed computational checks, and designed experiments to discriminate between competing predictions. When initial data drifted, the agents deemed the result inconclusive, revised the protocol, and ultimately confirmed the theoretical predictions with a successful experiment.
By El Mustapha Mansouri, Eikichi Kimura, Masato Koga, Keigo Arai
The paper introduces OrbitalGym, a JAX‑based simulation environment for studying how planning algorithms and system design influence the outcomes of adversarial orbital encounters. It supports multiple agents, diverse state representations, communication modes, and sensor models, and compiles games to batched GPU code for large‑scale parallel simulation. Using OrbitalGym, the authors evaluate several control policies (PD, LQR, MPC, MCTS) across two close‑range games—pursuit‑evasion and lady‑bandit‑guard—showing that propellant availability often dominates in pursuit‑evasion, while policy choice significantly improves defense in the lady‑bandit‑guard scenario.
By Duncan Eddy, Mahdi Al-Husseini, Mykel J. Kochenderfer
The paper introduces a six‑module scanner for detecting unintended local and composite vulnerabilities in LLM‑based agent skills. Local Analysis achieved high recall and precision on benchmark datasets, outperforming existing scanners, while Composite Analysis accurately modeled inter‑skill dependencies and identified composite flaws. The authors validated findings with proof‑of‑concept attacks, disclosed cases to maintainers, and demonstrated effective mitigations that preserved legitimate functionality.
By Parth A. Gandhi, Avital Aviv, Prasanna N. Wudali, Yuval Elovici, Asaf Shabtai