覆盖北京时间 2026-07-02 / 2026-07-03。candidate count 2000；new included count 560；selected count 15。按 Relevance、Novelty、Substance、Evidence、Actionability 评分排序。
Top Picks
#1 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Yunhao Feng, Ruixiao Lin, Ming Wen, Qinqin He 等 15 人 · 2026-07-02 · arXiv:cs.AI
一句话：LLM agents increasingly perform autonomous actions through external tools, leading to complex and evolving safety risks.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：LLM agents increasingly perform autonomous actions through external tools, leading to complex and evolving safety risks. However, existing safety testing targets expert-designed safety violations, and the corresponding outcomes are evaluated by hard-coded rules, making them costly to extend as agents evolve. To this end, we present Vera, an end-to-end automated safety testing framework that instantiates software engineering testing principles for non-deterministic agents through a three-stage, self-reinforcing pipe…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#2 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Xinyi Hou, Shenao Wang, Yanjie Zhao, Haoyu Wang · 2026-07-02 · arXiv:cs.SE
一句话：LLM agents increasingly rely on iterative execution to solve tasks through planning, tool use, state updates, and agent collaboration.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：LLM agents increasingly rely on iterative execution to solve tasks through planning, tool use, state updates, and agent collaboration. While this design enables flexible automation, it also creates a new class of failures: an agent may repeatedly execute model calls, tools, workflow transitions, or agent handoffs when the feedback path is not effectively bounded. We call this problem Infinite Agentic Loops (IALs). IALs are not ordinary programming loops; they arise from the interaction between agent logic, framewor…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#3 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Shenao Wang, Xinyi Hou, Yanjie Zhao, Xiao Cheng 等 5 人 · 2026-07-02 · arXiv:cs.SE
一句话：LLM agents are increasingly developed as source-code applications built on agent frameworks.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：LLM agents are increasingly developed as source-code applications built on agent frameworks. These agent programs combine conventional host-language code with framework-defined semantics for models, prompts, tools, memory, and multi-agent orchestration logic. As a result, their behavior depends not only on traditional control and data flows, but also on a new class of agent dependencies. Such dependencies are often expressed as framework-induced semantics, such as agent constructors, tool decorators, and agent hand…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#4 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Aseel Mohamed, Rama AlHamidi, Mohamed Rayan Barhdadi, Rasul Khanbayov 等 6 人 · 2026-07-02 · arXiv:cs.CV
一句话：Agentic Video Question Answering (VideoQA) systems invoke tools during inference, but their tool libraries are fixed, so recurring procedures are rebuilt from primitives on every question.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Agentic Video Question Answering (VideoQA) systems invoke tools during inference, but their tool libraries are fixed, so recurring procedures are rebuilt from primitives on every question. Synthesizing composite tools could remove this overhead, but whether such expansion helps is hard to assess: final-answer accuracy, the standard metric, ignores inference effort, so it cannot reveal how a system shifts cost. We propose a cost-aware, paired protocol for auditing tool-augmented video agents. The protocol pairs two …
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#5 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Chunjiang Liu, Xiaoyuan Wang, Haoyu Chen, Yizhou Zhao 等 6 人 · 2026-07-02 · arXiv:cs.AI
一句话：LLM agents are increasingly used to translate natural language into 3D scenes in a procedural way, but existing systems focus on static output.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：LLM agents are increasingly used to translate natural language into 3D scenes in a procedural way, but existing systems focus on static output. Dynamic 4D scenes from text alone, in which liquids flow, particles emit, rigid bodies cascade, and articulated mechanisms move, remain largely unexplored despite their value as editable content and as physics-grounded training data for video generation and embodied AI. Two challenges set the dynamic case apart from static text-to-scene work: an agent must jointly coordinat…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#6 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Zhiyun Zhang, Liwen Sun, Xiang Qian, Chenyan Xiong · 2026-07-02 · arXiv:cs.CL
一句话：Faithful reasoning is essential in medicine, where clinical decisions require transparent justification grounded in reliable evidence.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Faithful reasoning is essential in medicine, where clinical decisions require transparent justification grounded in reliable evidence. Current medical LLMs either lack active access to evidence or use retrieved evidence without supervising how it should be appraised and applied during reasoning. To address this, we formalize evidence-based medicine principles as process-level criteria and introduce FaithMed, a framework that combines clinician-designed, automatically refined rubrics with reinforcement learning usin…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#7 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Temitayo Olamilekan Ogunsusi, Lijun Qian, Xishuang Dong · 2026-07-02 · arXiv:cs.AI
一句话：Software development is a complex task that demands cooperation among agents with diverse roles.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Software development is a complex task that demands cooperation among agents with diverse roles. Large language models (LLMs) have enabled autonomous multi-agent software development frameworks that leverage role-based collaboration to automate requirements analysis, coding, testing, and refinement. However, existing approaches typically assume that intermediate agent outputs are equally reliable, leaving them vulnerable to hallucination propagation, where incorrect decisions generated in early development phases a…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#8 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Yueqi Song, Lintang Sutawika, Jiarui Liu, Lindia Tjuatja 等 11 人 · 2026-07-02 · arXiv:cs.AI
一句话：Evaluating LLM agents on benchmarks like SWE-Bench and GAIA can be expensive, time-consuming, and requires complex infrastructure.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Evaluating LLM agents on benchmarks like SWE-Bench and GAIA can be expensive, time-consuming, and requires complex infrastructure. A single evaluation can cost thousands of dollars and take days to complete. In contrast, non-agentic LLM benchmarks that test individual capabilities (e.g., reasoning, code generation) are fast and cheap to run. In this paper, we investigate whether performance on expensive agentic benchmarks can be accurately predicted by the performance on a small, carefully selected subset of atomic…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#9 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Jiayi Zhang, Kai Huang, Yang Liu, Chunyang Chen · 2026-07-02 · arXiv:cs.SE
一句话：Recent advances in agentic program repair have significantly improved issue resolution by enabling iterative repository exploration.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Recent advances in agentic program repair have significantly improved issue resolution by enabling iterative repository exploration. However, existing approaches predominantly rely on sequential, text-based code navigation, which fundamentally limits their ability to reason over large-scale long-horizon repositories with complex and long-range dependencies. As issue-resolution agents traverse repositories through fragmented textual observations, structural information such as module organization, call relationships…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#10 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Yuanzhi Liu, Shousheng Zhao, Bo Zhou, Kongming Liang 等 5 人 · 2026-07-02 · arXiv:cs.AI
一句话：Evaluation benchmarks are essential for assessing vision-language models (VLMs), but most multimodal benchmarks are static, making them vulnerable to temporal staleness, data contamination, and costly maintenance.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Evaluation benchmarks are essential for assessing vision-language models (VLMs), but most multimodal benchmarks are static, making them vulnerable to temporal staleness, data contamination, and costly maintenance. We present MMBench-Live, a continuously evolving multimodal benchmark built by a multi-agent-driven automated pipeline. Our framework treats benchmark evolution as task-guided dataset construction, integrating structured benchmark specification, feedback-controlled real-time data acquisition, and verifiab…
Lucian 下一步：把指标拆成可自动运行的 smoke benchmark，并记录最容易失败的 case。
arXiv · PDF
#11 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Kaiwen Xiong, Haonian Ji, Shi Qiu, Zeyu Zheng 等 7 人 · 2026-06-30 · arXiv query:all:"LLM agent"
一句话：Production large language-model (LLM) agents are increasingly deployed not as lone problem-solvers but as managers: a main model creates specialized subagents, delegates work, and orchestrates their parallel, asynchronous returns through dynamic workflows.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Production large language-model (LLM) agents are increasingly deployed not as lone problem-solvers but as managers: a main model creates specialized subagents, delegates work, and orchestrates their parallel, asynchronous returns through dynamic workflows. Whether one model can actually run such a team is largely unmeasured: existing benchmarks score a policy's own task-solving or a fixed multi-agent system's emergent behavior, but none isolate the management ability of the single LLM acting as leader. We introduce…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#12 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Ganlin Xu, Linghao Zhang, Zhitao Yin, Hongda Xi 等 10 人 · 2026-07-01 · arXiv:cs.IR
一句话：Retrieval-Augmented Generation (RAG) effectively grounds large language models (LLMs) in external knowledge but struggles with \textbf{exploratory reasoning problems (ERPs)} that are the complex queries involving high uncertainty and ambiguity.
为什么重要：聚焦 LLM 推理能力与评测，可用于设计更可靠的 planner / critic / verifier。
方法要点：Retrieval-Augmented Generation (RAG) effectively grounds large language models (LLMs) in external knowledge but struggles with \textbf{exploratory reasoning problems (ERPs)} that are the complex queries involving high uncertainty and ambiguity. Resolving ERPs requires complex reasoning with unclear paths, tending to result in retrieval noise and error accumulation. Furthermore, the absence of an end-to-end planning mechanism makes it difficult to generate effective trajectories for ERPs. Motivated by database query…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#13 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Carlos Baquero, Luís Brito · 2026-07-02 · arXiv:cs.MA
一句话：Multi-agent latent reasoning composes agents' KV-caches into one context for a final agent.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：Multi-agent latent reasoning composes agents' KV-caches into one context for a final agent. Prior work (Agent Primitives) does this by concatenating caches along the sequence axis with RoPE re-encoding, which we call BagMerge. BagMerge is non-commutative, and the best input ordering is unpredictable, shifting with the regime, the latent-step budget, and the model scale. We make this exchange a convergent replicated state. First, CanonicalMerge fixes the layout by content: ordering caches by mean K-norm at a middle …
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#14 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Michele Armillotta, Nicolò Romandini, Rebecca Montanari, Lorenzo Cavallaro · 2026-07-02 · arXiv query:all:"LLM reasoning"
一句话：LLM-based vulnerability detectors have shown promising results in identifying memory-safety bugs and vulnerability classes whose violations can often be expressed through established security properties.
为什么重要：直接关系到 Agent 的工具使用、任务分解或开放环境泛化，是 Lucian 当前 Agent/Auto Research 方向的高优先级素材。
方法要点：LLM-based vulnerability detectors have shown promising results in identifying memory-safety bugs and vulnerability classes whose violations can often be expressed through established security properties. Logic vulnerabilities, however, pose a different challenge, as their identification requires inferring application-specific security invariants and implicit assumptions about intended behavior. Even frontier agentic models struggle because these invariants are often implicit and buried among unrelated code. Motivat…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF
#15 · Relevance 5/5 | Novelty 4/5 | Substance 4/5 | Evidence 4/5 | Actionability 5/5 | Total 22/25
Hamed Babaei Giglou, Jennifer D'Souza, Andrei Aioanei, Nandana Mihindukulasooriya 等 5 人 · 2026-07-02 · arXiv:cs.AI
一句话：Ontology learning (OL) aims to automatically construct structured knowledge models from text, yet progress remains fragmented across methods, domains, and evaluation practices.
为什么重要：对检索增强、记忆系统和知识库工程有直接参考价值。
方法要点：Ontology learning (OL) aims to automatically construct structured knowledge models from text, yet progress remains fragmented across methods, domains, and evaluation practices. Despite decades of research, OL lacks a shared infrastructure for systematic evaluation and ontology access. This absence has hindered progress and fragmented research, leaving the central challenges of OL largely unaddressed. We introduce OntoLearner, a modular, cross-domain, and first-of-its-kind framework that unifies ontology access, lar…
Lucian 下一步：复现其任务定义/benchmark rubric，映射到 fjai.pw Code Agent 或 novel-feedback 工程流。
arXiv · PDF