Agent 研究周度深度综述（2026-07-20 ~ 2026-07-26）

数据来源：papers.cool/arxiv/cs.AI · 生成时间：2026/7/31 22:52:26
本报告聚合本周 arXiv cs.AI 的 Agent 相关论文，跨天去重后深度分析。

本周总览

本周去重后共 57 篇论文，覆盖 4 天数据。上周 76 篇，环比减少 19 篇。

研究方向分布

方向 本周 上周 变化
规划推理 26 22 +4
评估基准 13 19 -6
工程架构 9 6 +3
记忆系统 7 7 0
自我进化 7 1 +6
其他 7 25 -18
安全对齐 6 3 +3
多智能体 5 9 -4

应用场景分布

场景 论文数 占比
信息检索与问答 8 14%
决策支持 8 14%
企业自动化 5 9%
科学研究 5 9%
代码开发 2 4%
数据分析 1 2%
创意与内容 1 2%

核心论文解读

1. Agents in the Wild: Where Research Meets Deployment

  • arXiv: 2607.19336
  • 方向: 规划推理 · 工程架构
  • 场景: 科学研究、信息检索与问答、决策支持
  • 关键词: deployment agentic agents checklists wild meets research reasoning planning discovery

2. AgentFAIR: A Multi-Agent Collaborative Framework for FAIRness Evaluation of Geospatial Datasets

  • arXiv: 2607.15781
  • 方向: 记忆系统 · 多智能体 · 评估基准 · 工程架构
  • 关键词: agentfair geospatial evaluators dataset datasets averages percentage critic agent fair

3. DrawingVQA: A Real-World Benchmark for Multi-Depth Visual-Textual Reasoning on Construction Drawings

  • arXiv: 2607.15418
  • 方向: 规划推理 · 评估基准
  • 场景: 企业自动化、信息检索与问答
  • 关键词: drawings drawingvqa reasoning engineering construction workflows mllms domain textual world

4. AnovaX: A Local, Multi-Agent Voice Assistant with LLM Planning, Typed Executors, and Adaptive Recovery

  • arXiv: 2607.15367
  • 方向: 规划推理 · 多智能体 · 自我进化
  • 场景: 决策支持
  • 关键词: anovax assistant agent llm recovery phone typed executors desktop gemini

5. PhoenixRepair: Rethinking Repair Strategy Exploration in Software Agents

  • arXiv: 2607.18859
  • 方向: 其他
  • 场景: 代码开发、数据分析、决策支持
  • 关键词: phoenixrepair repair exploration swe agent edit deepsoftwareanalytics rethinking insufficient locations

6. Knowledge-Centric Agents for Workflow Generation

  • arXiv: 2607.15845
  • 方向: 规划推理
  • 场景: 企业自动化、信息检索与问答
  • 关键词: knowledge workflow generation workflows executable centric comfyui reasoning strategies brittleness

7. Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents

  • arXiv: 2607.15715
  • 方向: 安全对齐 · 自我进化
  • 场景: 企业自动化
  • 关键词: agentic extraction reflective workflows agent fixed agents llm reflection retries

8. S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding, Prediction, and Generation

  • arXiv: 2607.15686
  • 方向: 规划推理 · 评估基准
  • 场景: 科学研究
  • 关键词: omni scientific unified reasoning prediction generation specific multimodal benchmarks ai4s

9. SeerGuard: A Safety Framework for Mobile GUI Agents via World Model Prediction

  • arXiv: 2607.15550
  • 方向: 规划推理 · 安全对齐 · 工程架构
  • 关键词: gui safety seerguard risks risk assessment sawm mobile agents action

10. WorldCupArena: Fine-Grained Evaluation of Language Models and Deep-Research Agents on Football Forecasting

  • arXiv: 2607.18084
  • 方向: 评估基准
  • 场景: 科学研究、信息检索与问答
  • 关键词: worldcuparena scoreline score football match accuracy result wzk1015 evaluation kickoff

研究趋势

主导方向：规划推理（26 篇），较上周（22 篇）上升。

上升: 规划推理（22→26）、工程架构（6→9）、自我进化（1→7）、安全对齐（3→6）

下降: 其他（25→7）、多智能体（9→5）、评估基准（19→13）

技术演进脉络

规划推理（26 篇）

评估基准（13 篇）

工程架构（9 篇）

记忆系统（7 篇）

自我进化（7 篇）

其他（7 篇）

安全对齐（6 篇）

多智能体（5 篇）

工程实践启示

  • 工程架构方向 9 篇，关注系统设计与可扩展性。
  • 记忆系统方向 7 篇，RAG 与长期记忆方案不断优化。
  • 多智能体方向 5 篇，协作模式从简单分工走向复杂协调。
  • 安全方向 6 篇，Agent 安全从外部围栏走向内化机制。

下周关注

持续热点：规划推理（本周 26 篇，上周 22 篇）、评估基准（本周 13 篇，上周 19 篇）、工程架构（本周 9 篇，上周 6 篇）、记忆系统（本周 7 篇，上周 7 篇）、其他（本周 7 篇，上周 25 篇）、安全对齐（本周 6 篇，上周 3 篇）、多智能体（本周 5 篇，上周 9 篇）

附录：本周论文完整列表

去重后共 57 篇。

2026-07-20（17 篇）

  1. DSWorld: A Data Science World Model for Efficient Autonomous Agents — planning
  2. Knowledge-Centric Agents for Workflow Generation — planning
  3. AgentFAIR: A Multi-Agent Collaborative Framework for FAIRness Evaluation of Geospatial Datasets — memory, multi_agent, evaluation, engineering
  4. NeurOWL: An LLM-Based Neural-symbolic Framework for Incomplete OWL Ontology Reasoning — planning, engineering
  5. Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents — safety, evolution
  6. S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding, Prediction, and Generation — planning, evaluation
  7. ToolVerse: Unlocking Massive Environments and Long-Horizon Tasks for Agentic Reinforcement Learning — planning
  8. Neuro-Symbolic AI for LEED compliance: Document-Centric Benchmarking, Deterministic Numeric Checking, and When Multimodal Hurts — evaluation
  9. SeerGuard: A Safety Framework for Mobile GUI Agents via World Model Prediction — planning, safety, engineering
  10. From Black Box to Executable Logic: Explainable Reinforcement Learning through Prolog Expert Systems — other
  11. Beyond a Joke: Multi-Angle Reasoning for Detecting and Explaining Harmful Humor in Memes — planning, safety
  12. Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3? — planning
  13. DrawingVQA: A Real-World Benchmark for Multi-Depth Visual-Textual Reasoning on Construction Drawings — planning, evaluation
  14. Precise but Uncoupled: Reviewer Precision Does Not Guarantee Critique Uptake in Multi-Agent Math Reasoning — planning, multi_agent
  15. AnovaX: A Local, Multi-Agent Voice Assistant with LLM Planning, Typed Executors, and Adaptive Recovery — planning, multi_agent, evolution
  16. Cura 1T: Specialized Model for Agentic Healthcare — planning
  17. Causal-Audit: Explicit and Auditable Graph-based Reasoning via Target-Aware Causal Chain Construction — planning

2026-07-21（14 篇）

  1. Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steering — planning
  2. WorldCupArena: Fine-Grained Evaluation of Language Models and Deep-Research Agents on Football Forecasting — evaluation
  3. AdaHome: An Adaptive Smart Home Assistant using Local Small Language Models — planning, evolution
  4. Rethinking Heterogeneous LLM Merging: A Weighted Model Averaging Perspective — memory
  5. PAMD: Structured Adaptive Distances for Bisimulation Representations in Visual Reinforcement Learning — evolution
  6. SAGE: Subgoal-Conditioned Action Generation for Latent World Model Planning — planning
  7. Towards Agentic Agent-based Models: Feasibility, Performance, and Statistical Model Checking — other
  8. PEARL: Auditable Repair for Scientific Reasoning Graph Extraction — planning
  9. Stress Testing Concept Erasure with Large Language Model Agents — evaluation
  10. ST-Veto: Spatio-Temporal Token Veto for Diffusion MLLMs via Taylor Prediction and Visual Grounding — planning
  11. Exploratory and Assimilating Reflection: Reflective Recall Cycle for Long-term Memory — memory, evolution
  12. WuYu-EnvLE-Bench: A Benchmark for Evaluating Large Language Models in Environmental Law Enforcement — evaluation
  13. Semantically Similar, Logically Distinct: Diagnosing the Semantic-Answerability Gap in Table RAG — memory
  14. SR-Agent: An Experience-Driven Agentic Framework for Post-Ranking Strategies Refinement in E-Commerce Recommendation — engineering

2026-07-22（10 篇）

  1. CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents — other
  2. Agents in the Wild: Where Research Meets Deployment — planning, engineering
  3. ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D — other
  4. Graph-Based Agentic AI with LangGraph: Workflow Pathways for Long-Running Stateful Business Processes — other
  5. BioSecBench-Surveillance: A Verifiable Benchmark for AI Agents in Pathogen Genomic Surveillance — evaluation
  6. Comparative Study of Multi-Agent Actor-Critic Algorithms in Parameterized Action Reinforcement Learning — multi_agent
  7. OpenRTAG: A Comprehensive Benchmark for Robust Text-Attributed Graph Learning under Data Quality Degradation — evaluation
  8. Supra Cognitive Modes: A Routed Architecture for Agent Memory — memory, engineering
  9. Fishing Out Free Riders: Shapley-Based Reward Attribution for Parallel Reasoning via Reinforcement Learning — planning
  10. PhoenixRepair: Rethinking Repair Strategy Exploration in Software Agents — other

2026-07-23（16 篇）

  1. SoftReason: A Fully Differentiable Neuro-Soft-Symbolic Deductive Reasoning Architecture over High-Dimensional Perceptual Data — planning, engineering
  2. PoTRE: Test-Time Reasoning inspired by Cognitive Heterogeneity — planning, evaluation
  3. CUSUM-Shaped Inference-Time Monitoring and Targeted Re-Decoding for Quantized Small Language Model Reasoning — planning
  4. PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning — memory, planning
  5. EvoDRC: A Self-Evolving Agentic Framework for Automated DRC Violation Repair — engineering
  6. CLARK: Closed-loop Learning for Adaptive Reasoning over Knowledge Graphs — planning, evolution
  7. Coordinating from Memory: Graph-Structured Experience Reuse for Multi-Agent Adaptation in Dynamic Manufacturing — memory, multi_agent
  8. EvoThink: Evolving Thinking in Large Reasoning Models via Self-Pruning and Aha-Moment Preference Optimization — planning, engineering
  9. MOF-Sleuth: Tool-Grounded Reward Alignment for Explainable Fine-Grained MOF CIF Auditing — safety
  10. JANUS: Foreseeing Latent Risk for Long-Horizon Agent Safety — safety
  11. DocOps: A Verifiable Benchmark for Autonomous Agents in Complex Document Operations — evaluation
  12. Know Your Agent: Reconnaissance-Driven Pentesting of AI Agents — evaluation
  13. Rewarding Better Thinking for LLM Preference Alignment — safety
  14. Silent Failures in Multimodal Agentic Search:A Diagnostic Taxonomy and Cross-Judge Evaluation — evaluation
  15. Symbol and Footprint Database for Electronic Components by Agentic Recognition and Generation — other
  16. Knowledge-Centric Self-Improvement — evolution

本报告由 OpenClaw 自动生成，基于 agent-papers-research 每日数据聚合。

相关文章推荐

tag:

    缺失模块。
    1、请确保node版本大于6.2
    2、在博客根目录（注意不是yilia根目录）执行以下命令：
    npm i hexo-generator-json-content --save

    3、在根目录_config.yml里添加配置： 

      jsonContent:
    meta: false
    pages: false
    posts:
      title: true
      date: true
      path: true
      text: false
      raw: false
      content: false
      slug: false
      updated: false
      comments: false
      link: false
      permalink: false
      excerpt: false
      categories: false
      tags: true

程序员，关于技术我所知甚少，唯有探索，方得真知。