每篇包含 Motivation, Method, Data, Evaluation, Contribution, Quality note 与 Lucian 下一步。
Top Pick 1 · 2605.17101
SEMA-RAG: A Self-Evolving Multi-Agent Retrieval-Augmented Generation Framework for Medical Reasoning
Total 22/25cs.AI, cs.CLarXiv official recent/list pagesAbs date 2026/05/16
一句话结论:把静态 RAG 改成“多智能体、可自我演化”的医疗推理流程,适合作为高风险领域 agentic RAG 的工程蓝图。
作者/机构:Huang, Yongfeng, Chen, Ruiying, Cheng, James
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
医疗问答暴露了 RAG 最典型的断点:一次性检索、缺少临床语义转换、没有检索充分性反馈。SEMA-RAG 将这些步骤显式拆成协作 agent,和 Lucian 关注的 long-horizon reasoning / retrieval feedback loop 高度贴合。
Method
以多 agent 分工完成 query 语义解释、检索、答案生成和充分性反馈;框架会根据推理阶段动态改写检索策略,并让反馈信号推动后续轮次。
Data
论文面向 medical QA / medical reasoning 场景;可把其任务流程迁移到研究助理、代码库问答和个性化知识库。
Evaluation
摘要显示作者针对 hallucination、knowledge obsolescence 和 multi-stage clinical reasoning 的结构性缺陷设计系统;需要进一步查看完整表格确认对比 baseline 和消融。
Contribution
把静态 RAG 改成“多智能体、可自我演化”的医疗推理流程,适合作为高风险领域 agentic RAG 的工程蓝图。
Quality note
高相关、高工程启发;证据强度取决于完整 benchmark 与医生/专家评估细节。
局限/风险:医疗域系统可能依赖高质量临床语义解析;若检索器或 feedback agent 出错,错误会在多轮中被放大。
Lucian 下一步:复现一个小型“self-evolving RAG for paper reading”:让 retriever / verifier / answerer 三个 agent 在同一论文库上迭代检索充分性。
摘要摘录:Retrieval-Augmented Generation (RAG) is widely employed to mitigate risks such as hallucinations and knowledge obsolescence in medical question answering, yet its predominantly single-round, static retrieval paradigm misaligns with the multi-stage process of clinical reasoning. This compressed workflow induces two structural deficiencies: question-to-query translation often lacks clinically grounded semantic interpre…
Top Pick 2 · 2605.18032
PROTEA: Offline Evaluation and Iterative Refinement for Multi-Agent LLM Workflows
Total 23/25cs.AI, cs.CL, cs.HC, cs.SEarXiv official recent/list pagesAbs date 2026/05/18
一句话结论:PROTEA 把 multi-agent workflow 的离线调试变成 test-driven loop,是构建可维护 agent 产品的关键缺口。
作者/机构:Kawamura, Kazuki, Waki, Satoshi, Tateno, Kei
Relevance5/5Novelty4/5Substance5/5Evidence4/5Actionability5/5Total23/25
Motivation
多 agent workflow 失败常来自中间节点的细小偏差,单看最终答案很难定位。PROTEA 的 offline evaluation + iterative refinement 正好对应 Lucian 想做的 agent runtime 可观测、可回放、可修复能力。
Method
执行完整 workflow,给中间输出和最终输出打分,定位哪个 agent/节点最需要修改,再进行迭代 refinement;强调 trace inspection 与 test-driven improvement。
Data
面向多角色 LLM calls 构成的 workflow;可应用在研究报告、代码 agent、工具链 agent 的离线回归集。
Evaluation
摘要称系统统一了执行、评分和细粒度诊断;需阅读全文确认用户研究/任务集规模。
Contribution
PROTEA 把 multi-agent workflow 的离线调试变成 test-driven loop,是构建可维护 agent 产品的关键缺口。
Quality note
方法实质强、对工程落地极具 actionability;优先读。
局限/风险:如果评分器本身偏置,refinement loop 可能优化到错误目标;离线测试也可能覆盖不到真实线上分布。
Lucian 下一步:为 fjai.pw 的论文日报/研究 agent 建一个 PROTEA-like trace:每个候选选择、摘要、评分都留下可回放中间产物。
摘要摘录:Multi-agent LLM workflows -- systems composed of multiple role-specific LLM calls -- often outperform single-prompt baselines, but they remain difficult to debug and refine. Failures can originate from subtle errors in intermediate outputs that propagate to downstream nodes, requiring developers to inspect long traces and infer which agent to modify. We present PROTEA, a unified interface for offline, test-driven imp…
Top Pick 3 · 2605.18652
MementoGUI: Learning Agentic Multimodal Memory Control for Long-Horizon GUI Agents
Total 22/25cs.CVHugging Face Daily Papers + arXiv official recent/list pagesAbs date 2026/05/18
一句话结论:MementoGUI 针对 GUI agent 的长程视觉状态记忆,是 computer-use agent 从“看当前屏幕”走向“记任务过程”的代表工作。
作者/机构:Zeng, Ziyun, Hua, Hang, Zou, Bocheng, Cai, Mu, Feris, Rogerio, Luo, Jiebo
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
GUI agent 在真实任务中需要跨很多界面保留局部视觉证据;纯文本 memory 或 raw history replay 都不够。该文对长程 GUI / browser agent 记忆模块很有参考价值。
Method
提出 plug-in agentic memory control:压缩/选择多模态历史,保存对未来决策有用的局部视觉证据,并在后续 GUI action prediction 时调用。
Data
面向 long-horizon GUI tasks;Hugging Face Daily Papers 也收录,说明社区热度较高。
Evaluation
摘要强调相对 raw history replay/text-only memory 的问题;完整论文应包含 GUI benchmark 上的 action accuracy 或 success rate 对比。
Contribution
MementoGUI 针对 GUI agent 的长程视觉状态记忆,是 computer-use agent 从“看当前屏幕”走向“记任务过程”的代表工作。
Quality note
长程 agent memory 的优质工程型论文,值得进入 Top 3。
局限/风险:视觉记忆控制可能对具体 UI 分布敏感;跨网站、跨 App 泛化仍需验证。
Lucian 下一步:在网页自动化 agent 中加入“视觉证据卡片”:每轮只保留按钮/表单/结果区域截图摘要,而不是完整历史。
摘要摘录:Recent GUI agents have made substantial progress in visual grounding and action prediction, yet they remain brittle in long-horizon tasks that require maintaining task state across many interface transitions. Existing agents typically rely on raw history replay or text-only memory, which either overwhelms the model with redundant screenshots or discards localized visual evidence needed for future decisions. To addres…
Top Pick 4 · 2605.17641
Causal Intervention-Based Memory Selection for Long-Horizon LLM Agents
Total 21/25cs.AI, cs.CLarXiv official recent/list pagesAbs date 2026/05/17
一句话结论:用因果干预而不是相似度来选择 agent memory,直击长期记忆中“相关但误导”的问题。
作者/机构:Srivastava, Saksham Sahai
Relevance5/5Novelty4/5Substance4/5Evidence3/5Actionability5/5Total21/25
Motivation
Lucian 长期关注 personalized/user-feedback agents;记忆检索不能只靠 embedding 相似度。CMI 把“这条记忆是否改变模型行为”作为选择依据,方向很重要。
Method
Causal Memory Intervention 估计候选记忆对模型预测/行动的影响,过滤 stale、irrelevant 或 misleading memory。
Data
面向 long-horizon LLM agents 与跨 session interactions。
Evaluation
摘要提出明确失效模式和算法思想;实验强度需看是否有长期交互 benchmark 与反事实消融。
Contribution
用因果干预而不是相似度来选择 agent memory,直击长期记忆中“相关但误导”的问题。
Quality note
研究问题非常准,方法值得深挖。
局限/风险:因果影响估计可能昂贵;对闭源 LLM 只能近似干预,在线部署成本需控制。
Lucian 下一步:在个人化小说/研究 agent 的 memory retrieve API 增加“影响估计”字段:相似度 + 反事实回答差异。
摘要摘录:Long-horizon LLM agents rely on persistent memory to support interactions across sessions, yet existing memory systems often retrieve context using semantic similarity or broad history inclusion, treating retrieved memories as uniformly useful. This assumption is fragile because memories may be topically related while remaining irrelevant, stale, or misleading. We propose Causal Memory Intervention (CMI), a causal me…
Top Pick 5 · 2605.18854
Evaluating Memory Condensation Strategies for Coding Agents in Data-Driven Scientific Discovery
Total 22/25cs.LGarXiv official recent/list pagesAbs date 2026/05/13
一句话结论:系统比较 coding agents 的八种 memory condensation 策略,为长任务代码/科学发现 agent 提供直接工程选择。
作者/机构:Chintalapati, Renuka, Raskar, Sid, Acharya, Anurag, Willard, Jared, Emami, Patrick, Horawalavithana, Sameera
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
上下文窗口固定导致长程 coding agent 必须做压缩;该文把 sliding window、LLM summaries 等策略放到 DiscoveryBench 上比较,对工程决策很直接。
Method
评估 8 种记忆压缩策略,在 GPT-4o + 60 个 DiscoveryBench 科学发现任务上比较任务成功与上下文效率。
Data
60 个 DiscoveryBench tasks,覆盖 6 个科学领域;场景比普通代码补丁更接近 autonomous discovery。
Evaluation
摘要给出清晰实验设置;需要读表格找出哪种 condensation 在成本/成功率上最优。
Contribution
系统比较 coding agents 的八种 memory condensation 策略,为长任务代码/科学发现 agent 提供直接工程选择。
Quality note
非常 actionable,适合作为后续工程实验起点。
局限/风险:仅 GPT-4o 和 DiscoveryBench 未必代表所有 coding agents;压缩策略可能与模型长度/工具链强耦合。
Lucian 下一步:为 Hermes/Codex coding agent 做同样小规模 ablation:raw history vs summary vs episodic notes vs retrieval memory。
摘要摘录:Coding agents accumulate extensive context during long-running tasks, yet fixed context windows force practitioners to choose between truncation and task failure. While numerous memory condensation strategies have been proposed, from simple sliding windows to LLM-generated summaries, no systematic comparison exists to guide strategy selection, especially in scientific discovery tasks. We evaluate eight memory condens…
Top Pick 6 · 2605.17637
WebGameBench: Requirement-to-Application Evaluation for Coding Agents via Browser-Native Games
Total 22/25cs.AIarXiv official recent/list pagesAbs date 2026/05/17
一句话结论:WebGameBench 从“代码是否对”转向“浏览器里应用是否真的可玩”,更贴近 coding agent 交付质量。
作者/机构:Zhang, Wenyu, You, Guoliang, Tianlun, Zhao, Haotian, Zhu, Tianshu, Wang, Haoran 等 11 人
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
现有 coding benchmark 常停在源码/单元测试;真实 app 需要输入处理、状态同步、渲染和行为闭环。WebGameBench 用 browser-native games 压缩出高密度交付评测。
Method
冻结 Structured WebGame Specification,让 coding agent 生成可浏览器访问的游戏,再从需求到应用进行行为评估。
Data
一组浏览器原生游戏规格;任务小但行为密度高,适合检测端到端应用构建能力。
Evaluation
摘要给出 benchmark 设计动机;需查看 evaluator 是否自动化、是否覆盖视觉/交互/规则一致性。
Contribution
WebGameBench 从“代码是否对”转向“浏览器里应用是否真的可玩”,更贴近 coding agent 交付质量。
Quality note
高工程价值;对 coding-agent evaluation 很实用。
局限/风险:游戏域与企业软件仍有差异;代理可能针对小游戏模式过拟合。
Lucian 下一步:把 fjai.pw 静态原型测试从 DOM token 扩展为 browser behavior specs:按钮、状态、任务完成路径。
摘要摘录:Coding agents are increasingly used as application builders, yet many evaluations still focus on source code, repository-level tests, or intermediate traces rather than the delivered application. We introduce WebGameBench, a requirement-to-application benchmark that evaluates whether coding agents can turn a frozen Structured WebGame Specification into a browser-accessible game. Browser-native games provide a compact…
Top Pick 7 · 2605.16679
CHI-Bench: Can AI Agents Automate End-to-End, Long-Horizon, Policy-Rich Healthcare Workflows?
Total 22/25cs.AI, cs.CLHugging Face Daily Papers + arXiv official recent/list pagesAbs date 2026/05/15
一句话结论:CHI-Bench 用医疗政策密集、多角色、长流程任务评估 agent,是 long-horizon workflow benchmark 的强代表。
作者/机构:Chen, Haolin, Metelski, Deon, Qi, Leon, Xia, Tao, Lee, Joonyul, Brown, Steve 等 33 人
Relevance5/5Novelty4/5Substance5/5Evidence4/5Actionability4/5Total22/25
Motivation
真实工作流不是单步 QA,而是规则库、多角色交接、多轮沟通。CHI-Bench 的 policy-rich healthcare operations 可以启发“规则密集型 agent”评测。
Method
构建端到端医疗运营任务,要求 agent 在政策约束、多角色组合和多边互动中完成长流程。
Data
Healthcare workflow tasks;大规模作者团队,Hugging Face Daily Papers 收录。
Evaluation
摘要列出三类现有 benchmark 缺失能力;需查成功率、人类基线、错误分类。
Contribution
CHI-Bench 用医疗政策密集、多角色、长流程任务评估 agent,是 long-horizon workflow benchmark 的强代表。
Quality note
问题设置强,适合作为 long-horizon agent benchmark 参考。
局限/风险:医疗规则专门性强;非医疗迁移要抽象成 policy library + role handoff。
Lucian 下一步:把 benchmark 结构迁移到“研究助理合规流程”:引用规则、任务分工、审稿/复核 handoff。
摘要摘录:End-to-end automation of realistic healthcare operations stresses three capabilities underrepresented in current benchmarks: policy density, decisions must be grounded in a large library of medical, insurance, and operational rules; Multi-role composition: a single task requires the agent to play multiple roles with handoffs; and multilateral interaction: intermediate workflow steps are multi-turn dialogs, such as pe…
Top Pick 8 · 2605.16909
TOBench: A Task-Oriented Omni-Modal Benchmark for Real-World Tool-Using Agents
Total 22/25cs.AIHugging Face Daily Papers + arXiv official recent/list pagesAbs date 2026/05/16
一句话结论:TOBench/MM-TOBench 评估真实世界全模态工具使用,把 tool use、computer use、多模态推理放到同一端到端任务里。
作者/机构:Liu, Zhiqiang, Dong, Wenhui, Tan, Yilang, Qu, Yuwen, Yin, Haochen, Si, Chenyang
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
工具 agent 的难点在多模态输入、外部工具、中间 artifact 检查与修正的组合。该 benchmark 能帮助避免只测单一 tool-call accuracy。
Method
设计 task-oriented omni-modal tasks,要求 agent 解释多模态输入、调用工具、检查中间结果并迭代行动。
Data
Real-world tool-using agent tasks;HF Daily Papers 收录。
Evaluation
摘要说明 benchmark 填补 isolation evaluation 与端到端真实工具使用的 gap;需看任务数量和自动评分方式。
Contribution
TOBench/MM-TOBench 评估真实世界全模态工具使用,把 tool use、computer use、多模态推理放到同一端到端任务里。
Quality note
很适合设计下一代 agent evaluation。
局限/风险:全模态环境成本高,复现可能依赖具体工具和 artifact 格式。
Lucian 下一步:在 Hermes agent 评测中加入 artifact inspection:不只看 API 调用成功,还看中间文件/网页/图像是否满足任务。
摘要摘录:Tool-using agents are increasingly expected to operate across realistic professional workflows, where they must interpret multimodal inputs, coordinate external tools, inspect intermediate artifacts, and revise their actions before producing a final result. Existing benchmarks, however, often evaluate tool use, computer use, and multimodal reasoning in isolation, leaving a gap between benchmark settings and end-to-en…
Top Pick 9 · 2605.18703
EnvFactory: Scaling Tool-Use Agents via Executable Environments Synthesis and Robust RL
Total 23/25cs.CL, cs.LGHugging Face Daily Papers + arXiv official recent/list pagesAbs date 2026/05/18
一句话结论:EnvFactory 试图合成可执行环境来扩展 tool-use agent RL,抓住了 Agentic RL 的数据/环境瓶颈。
作者/机构:Xu, Minrui, Wang, Zilin, DENG, Mengyi, Li, Zhiwei, Yang, Zhicheng, Zhu, Xiao 等 15 人
Relevance5/5Novelty4/5Substance5/5Evidence4/5Actionability5/5Total23/25
Motivation
Agentic RL 需要大量可执行环境和真实轨迹;真实 API 贵,LLM simulator 容易幻觉。环境合成如果可靠,会显著降低 tool-use 训练门槛。
Method
合成 executable environments,并用 robust RL 训练工具使用 agent;目标是替代昂贵真实 API 或脆弱模拟器。
Data
面向 tool-use agents 的可执行环境与轨迹;HF Daily Papers 收录。
Evaluation
摘要明确指出现有模拟/合成轨迹的 failure mode;需读实验确认环境多样性与 transfer。
Contribution
EnvFactory 试图合成可执行环境来扩展 tool-use agent RL,抓住了 Agentic RL 的数据/环境瓶颈。
Quality note
方法实质强,和 RL for LLM/tool use 高度相关。
局限/风险:合成环境与真实 API 行为差异可能导致 sim-to-real gap;环境验证是核心风险。
Lucian 下一步:为内部工具 agent 生成“可执行假环境 + verifier”,先用于离线 RL/偏好数据,再迁移真实 API。
摘要摘录:Equipping LLMs with tool-use capabilities via Agentic Reinforcement Learning (Agentic RL) is bottlenecked by two challenges: the lack of scalable, robust execution environments and the scarcity of realistic training data that captures implicit human reasoning. Existing approaches depend on costly real-world APIs, hallucination-prone LLM simulators, or synthetic environments that are often single-turn or depend on pre…
Top Pick 10 · 2605.17554
Evaluating Deep Research Agents on Expert Consulting Work: A Benchmark with Verifiers, Rubrics, and Cognitive Traps
Total 22/25cs.AI, cs.LGarXiv official recent/list pagesAbs date 2026/05/17
一句话结论:把 deep research agents 放到专家咨询交付物上评测,比一般 QA 更接近 Lucian 的自动研究目标。
作者/机构:Asthana, Tanmay, Saksena, Aman, Sahu, Divyansh
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
DRAs 已用于企业调研,但常缺少 decision-grade deliverable 评测。该 benchmark 加入 verifiers、rubrics 和 cognitive traps,可直接借鉴到论文日报/研究报告质量控制。
Method
设计管理咨询式结构化分析任务,用 rubrics/verifiers 检查多文档综合、陷阱识别和最终交付质量。
Data
Expert consulting work benchmark;面向 frontier deep research agents。
Evaluation
摘要指出现有 factual recall/single-hop QA 不足;需看 rubrics 是否可自动化与模型排行。
Contribution
把 deep research agents 放到专家咨询交付物上评测,比一般 QA 更接近 Lucian 的自动研究目标。
Quality note
对 Auto Research 产品化很有参考价值。
局限/风险:咨询任务主观性强;rubric 设计会影响模型排名。
Lucian 下一步:把 AI 论文日报生成流程拆成 research agent benchmark:候选召回、去重、评分、Top 3 选择、引用正确性。
摘要摘录:Frontier deep research agents (DRAs) plan a research task, synthesize across documents, and return a structured deliverable on demand. They are being deployed in enterprise workflows faster than they are being evaluated. Existing benchmarks measure factual recall, single-hop QA, or generic agentic skill, missing the multi-document, decision-grade work DRAs are deployed to produce. We introduce a benchmark targeting t…
Top Pick 11 · 2605.17526
SaaSBench: Exploring the Boundaries of Coding Agents in Long-Horizon Enterprise SaaS Engineering
Total 23/25cs.AI, cs.SEarXiv official recent/list pagesAbs date 2026/05/17
一句话结论:SaaSBench 将 coding agents 拉到长程企业 SaaS 工程,比单仓库 patch 更接近真实软件交付。
作者/机构:Ren, Qingnan, Zou, Shun, Huang, Shiting, Zhang, Ziao, Shi, Kou, Fang, Zhen 等 14 人
Relevance5/5Novelty4/5Substance5/5Evidence4/5Actionability5/5Total23/25
Motivation
从 scratch 项目生成需要 full-stack orchestration、异构环境、系统级复杂度;这是 coding agent 的下一阶段评测。
Method
构建长程企业 SaaS engineering tasks,评估 autonomous coding agents 的端到端系统实现能力。
Data
Enterprise SaaS engineering benchmark;cs.SE + cs.AI。
Evaluation
摘要对比现有 localized code editing / simplified single-stack generation;需读任务规格、验收测试、模型结果。
Contribution
SaaSBench 将 coding agents 拉到长程企业 SaaS 工程,比单仓库 patch 更接近真实软件交付。
Quality note
对 coding-agent roadmap 很重要。
局限/风险:SaaS 任务可能评测成本高;自动验收能否覆盖 UX、安全、数据一致性仍是挑战。
Lucian 下一步:为 fjai.pw product prototypes 定义 SaaSBench-like acceptance:API、持久化、隐私、前端交互、部署验证。
摘要摘录:As autonomous coding agents become capable of handling increasingly long-horizon tasks, they have gradually demonstrated the potential to complete end-to-end software development. Although existing benchmarks have recently evolved from localized code editing to from-scratch project generation, they remain confined to structurally simplified, single-stack applications. Consequently, they fail to capture the heterogene…
Top Pick 12 · 2605.19932
PEEK: Context Map as an Orientation Cache for Long-Context LLM Agents
Total 21/25cs.AI, cs.CL, cs.LGHugging Face Daily Papers + arXiv official recent/list pagesAbs date 2026/05/19
一句话结论:PEEK 把长上下文中的“方向感”缓存为 context map,适合反复访问同一代码库/文档库的 agent。
作者/机构:Gu, Zhuohan, Zhang, Qizheng, Khattab, Omar, Madden, Samuel
Relevance5/5Novelty4/5Substance4/5Evidence3/5Actionability5/5Total21/25
Motivation
长上下文 agent 反复处理同一外部环境时,最缺的不是轨迹,而是“这里有什么、怎么组织、关键实体在哪里”的 orientation knowledge。
Method
构建 Context Map / Orientation Cache,保存可复用的上下文结构知识,供后续 invocation 快速定位。
Data
面向 document corpora 和 code repositories 等 recurring external contexts;HF Daily Papers 收录。
Evaluation
摘要提出了与 raw material / trajectory / strategies 不同的 memory 类型;实验需看跨调用收益。
Contribution
PEEK 把长上下文中的“方向感”缓存为 context map,适合反复访问同一代码库/文档库的 agent。
Quality note
概念非常有用,适合和 memory condensation 结合。
局限/风险:context map 需要更新与失效机制;环境变化后旧地图可能误导。
Lucian 下一步:给代码库 agent 自动生成 repo context map:目录职责、核心 schema、重要常量、常见任务入口。
摘要摘录:Large language model (LLM) agents increasingly operate over long and recurring external contexts, like document corpora and code repositories. Across invocations, existing approaches preserve either the agent's trajectory, passive access to raw material, or task-level strategies. None of them preserves what we argue is most needed for repeated same-context workloads: reusable orientation knowledge (e.g., what the con…
Top Pick 13 · 2605.19769
OpenComputer: Verifiable Software Worlds for Computer-Use Agents
Total 23/25cs.AI, cs.SEHugging Face Daily Papers + arXiv official recent/list pagesAbs date 2026/05/19
一句话结论:OpenComputer 用 verifier-grounded software worlds 评估 computer-use agents,强调可验证真实应用任务。
作者/机构:Wei, Jinbiao, Ma, Qianran, Zhao, Yilun, Zhou, Xiao, Ni, Kangqi, Gan, Guo 等 7 人
Relevance5/5Novelty4/5Substance5/5Evidence4/5Actionability5/5Total23/25
Motivation
Computer-use agent 最大难点是环境真实但可评分。OpenComputer 用应用状态 verifier 和自进化 verification layer,解决“看起来完成但无法机器确认”的问题。
Method
构建软件世界、app-specific state verifiers、task generation pipeline 与 evaluation harness;verifier 通过执行反馈自我改进。
Data
Real applications / desktop software tasks;HF Daily Papers 收录。
Evaluation
摘要提供完整框架组件;需查看任务规模、verifier 可靠性和 agent 排名。
Contribution
OpenComputer 用 verifier-grounded software worlds 评估 computer-use agents,强调可验证真实应用任务。
Quality note
强烈推荐,适合作为 agent eval infrastructure 参考。
局限/风险:为每个 app 写 verifier 有工程成本;self-evolving verifier 也需要防止错误奖励。
Lucian 下一步:在浏览器/网页 agent benchmark 中加入可查询 state verifier,而不只依赖截图或最终文本。
摘要摘录:We present OpenComputer, a verifier-grounded framework for constructing verifiable software worlds for computer-use agents. OpenComputer integrates four components: (1) app-specific state verifiers that expose structured inspection endpoints over real applications, (2) a self-evolving verification layer that improves verifier reliability using execution-grounded feedback, (3) a task-generation pipeline that synthesiz…
Top Pick 14 · 2605.19782
Prior Knowledge or Search? A Study of LLM Agents in Hardware-Aware Code Optimization
Total 21/25cs.AI, cs.LG, cs.SEarXiv official recent/list pagesAbs date 2026/05/19
一句话结论:研究 LLM agents 在硬件感知代码优化中依赖先验知识还是搜索,有助于拆解 propose-evaluate-revise loop。
作者/机构:Redko, Dmitry, Fazlyev, Albert, Sozykin, Konstantin, Ivanova, Maria, Burnaev, Evgeny, Shvetsov, Egor
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability4/5Total21/25
Motivation
Agent optimization 系统常被当成黑盒循环;这篇用 controlled experiments 分析哪些组件贡献最大、何时搜索失败,适合指导 code optimization/coding agent 设计。
Method
围绕 hardware-aware code optimization 设计三组 controlled experiments,区分先验知识、搜索与反馈环境的贡献。
Data
硬件感知代码优化任务;cs.AI/cs.LG/cs.SE。
Evaluation
摘要明确提出 controlled experiments;需看优化指标、硬件平台和失败案例。
Contribution
研究 LLM agents 在硬件感知代码优化中依赖先验知识还是搜索,有助于拆解 propose-evaluate-revise loop。
Quality note
适合做 agent component attribution 的参考。
局限/风险:硬件优化是特殊域;结论迁移到通用软件工程需谨慎。
Lucian 下一步:在 coding agent benchmark 中记录“搜索行为 vs 先验模板”的贡献,避免只看最终性能。
摘要摘录:LLM discovery and optimization systems are increasingly applied across domains, implementing a common propose-evaluate-revise loop. Such optimization or discovery progresses via context conditioning on received feedback from an environment. However, as modern LLM agents are increasingly complex in their structure, it is difficult to evaluate which components contribute the most, and when and how this exploration may …
Top Pick 15 · 2605.19099
DecisionBench: A Benchmark for Emergent Delegation in Long-Horizon Agentic Workflows
Total 22/25cs.AI, cs.CL, cs.MAarXiv official recent/list pagesAbs date 2026/05/18
一句话结论:DecisionBench 专门评估长程 agentic workflows 中的 delegation/routing,是 multi-model agent orchestration 的好基准。
作者/机构:Gao, Yuxuan, Wang, Megan, Yu, Yi Ling, Ma, Zijian Carl, Qu, Ao
Relevance5/5Novelty4/5Substance4/5Evidence4/5Actionability5/5Total22/25
Motivation
未来 agent 不一定单模型完成任务,而是把子任务委派给不同模型/供应商。DecisionBench 固定 peer-model pool、delegation interface 和多轴指标,很贴近生产 orchestration。
Method
基于 GAIA、tau-bench、BFCL multi-turn,提供 call_model/read_profile interface、skill annotations 和质量/成本/延迟/路由忠实度等指标。
Data
11 models、7 vendor families,覆盖多任务 suite。
Evaluation
摘要给出清晰 substrate 与 metrics;需要读模型结果看 delegation 是否真的提升质量/成本。
Contribution
DecisionBench 专门评估长程 agentic workflows 中的 delegation/routing,是 multi-model agent orchestration 的好基准。
Quality note
高 actionability,适合 orchestration 研究。
局限/风险:固定模型池会随时间过期;供应商自偏好与成本结构变化快。
Lucian 下一步:为 Hermes 多代理任务加入 delegation benchmark:什么时候委托 Claude/Codex/Hermes 子 agent,如何评估成本收益。
摘要摘录:We introduce DecisionBench, a benchmark substrate for emergent delegation in long-horizon agentic workflows. The substrate fixes a task suite (GAIA, tau-bench, BFCL multi-turn), a peer-model pool (11 models, 7 vendor families), a delegation interface (call_model plus an optional read_profile channel), a deterministic skill-annotation layer, and a multi-axis metric suite covering quality, cost, latency, delegation rat…