[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-CharlesQ9--Self-Evolving-Agents":3,"tool-CharlesQ9--Self-Evolving-Agents":64},[4,17,27,35,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",140436,2,"2026-04-05T23:32:43",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,43,44,45,15,46,26,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":10,"last_commit_at":54,"category_tags":55,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,46],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},519,"PaddleOCR","PaddlePaddle\u002FPaddleOCR","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。",74939,"2026-04-05T23:16:38",[26,14,13,46],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":80,"owner_email":81,"owner_twitter":68,"owner_website":68,"owner_url":82,"languages":68,"stars":83,"forks":84,"last_commit_at":85,"license":86,"difficulty_score":87,"env_os":88,"env_gpu":88,"env_ram":88,"env_deps":89,"category_tags":92,"github_topics":68,"view_count":10,"oss_zip_url":68,"oss_zip_packed_at":68,"status":16,"created_at":93,"updated_at":94,"faqs":95,"releases":101},455,"CharlesQ9\u002FSelf-Evolving-Agents","Self-Evolving-Agents",null,"Self-Evolving-Agents 是一个专注于“自进化智能体”领域的学术资源库，旨在系统梳理通往人工超级智能（ASI）的研究路径。它本质上是一份高质量的文献综述合集，而非单一的可执行软件。\n\n它主要解决了当前 AI 智能体难以在动态环境中自主适应、缺乏持续学习能力的问题。通过整合模型优化、记忆演化、工具调用及架构调整等核心内容，Self-Evolving-Agents 为构建能够像生物一样不断自我完善的 AI 系统提供了理论框架。\n\n这份资源非常适合 AI 领域的研究人员、算法工程师以及对通用人工智能（AGI）感兴趣的开发者深入研读。其独特价值在于构建了一个全景式的进化方法论体系，详细探讨了“进化什么”（如模型、上下文）、“何时进化”以及“如何进化”（如基于奖励、模仿学习、群体进化）等关键维度，为研发具备终身学习能力的下一代智能体提供了宝贵的导航图。","# \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FCharlesQ9_Self-Evolving-Agents_readme_6e592f5bdbb5.png\" alt=\"Example Figure\" width=\"50\" height=\"50\" \u002F> A Survey of Self-Evolving Agents: On Path to Artificial Super Intelligence\n\n# \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FCharlesQ9_Self-Evolving-Agents_readme_e5646a3a595b.jpg\" alt=\"Example Figure\" width=\"800\" height=\"140\" \u002F>\n\n\u003C!-- omit in toc -->\n## 📢 Updates\n\n- **2025.07**: We released a github repo to record papers related with reasoning economy. Feel free to cite or open pull requests.\n\n\n---\n\n## 📜 Table of Contents\n- [A Survey of Self-Evolving Agents: On Path to Artificial Super Intelligence](#a-survey-of-self-evolving-agents-on-path-to-artificial-super-intelligence)\n    - [1. Introduction](#1-introduction)\n    - [2. Definitions and Foundations](#2-definitions-and-foundations)\n    - [3. What to Evolve?](#3-what-to-evolve)\n        - [3.1 Models](#31-models)\n        - [3.2 Context](#32-context)\n            - [3.2.1 Memory Evolution](#321-memory-evolution)\n            - [3.2.2 Prompt Optimization](#322-prompt-optimization)\n        - [3.3 Tools](#33-tools)\n        - [3.4 Architecture](#34-architecture)\n            - [3.4.1 Single-Agent System Optimization](#341-single-agent-system-optimization)\n            - [3.4.2 Multi-Agent System Optimization](#342-multi-agent-system-optimization)\n    - [4. When to Evolve?](#4-when-to-evolve)\n        - [4.1 Intra-test-Time Self-Evolution](#41-intra-test-time-self-evolution)\n        - [4.2 Inter-test-Time Self-evolution](#42-inter-test-time-self-evolution)\n    - [5. How to Evolve](#5-how-to-evolve)\n        - [5.1 Reward-based Self-Evolution](#51-reward-based-self-evolution)\n        - [5.2 Imitation and Demonstration Learning](#52-imitation-and-demonstration-learning)\n            - [5.2.1 Self-generated Demonstration Learning](#521-self-generated-demonstration-learning)\n            - [5.2.2 Cross-Agent Demonstration Learning](#522-cross-agent-demonstration-learning)\n            - [5.2.3 Hybrid Demonstration Learning](#523-hybrid-demonstration-learning)\n        - [5.3 Population-based and Evolutionary Methods](#53-population-based-and-evolutionary-methods)\n        - [5.4 Cross-cutting Evolutionary Dimensions](#54-cross-cutting-evolutionary-dimensions)\n        - [5.5 Other Dimensions of Self-Evolution Methods](#55-other-dimensions-of-self-evolution-methods)\n    - [6. Where to Evolve?](#6-where-to-evolve)\n        - [6.1 General Domain Evolution](#61-general-domain-evolution)\n        - [6.2 Specialized Domain Evolution](#62-specialized-domain-evolution)\n    - [7. Evaluation of Self-evolving Agents](#7-evaluation-of-self-evolving-agents)\n        \u003C!-- - [7.1 Evaluation Goal and Metrics](#71-evaluation-goal-and-metrics)\n        - [7.2 Evaluation Paradigm](#72-evaluation-paradigm)\n            - [7.2.1 Static Assessment](#721-static-assessment)\n            - [7.2.2 Short-Horizon Adaptive Assessment](#722-short-horizon-adaptive-assessment)\n            - [7.2.3 Long-Horizon Lifelong Learning Ability Assessment](#723-long-horizon-lifelong-learning-ability-assessment) -->\n    - [8. Future Directions](#8-future-directions)\n\n\n---\n\n### 1. Introduction\n\n- [Large language model agent: A survey on methodology, applications and challenges](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.21460.pdf)\n\n- [Advances and challenges in foundation agents: From brain-inspired intelligence to evolutionary, collaborative, and safe systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.01990.pdf)\n\n- [Toward a Theory of Agents as Tool-Use Decision-Makers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.00886.pdf)\n\n- [A survey on self-evolution of large language models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14387.pdf)\n\n### 2. Definitions and Foundations\n\n- [Curriculum Learning for Cooperation in Multi-Agent Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.11768.pdf)\n\n- [A Survey on Curriculum Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2010.13166)\n\n- [Self-Evolving Curriculum for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.14970)\n\n- [Continual lifelong learning with neural networks: A review](https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.07569)\n\n- [Lifelong Learning of Large Language Model-based Agents: A Roadmap](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.07278)\n\n### 3. What to Evolve?\n\n#### 3.1 Models\n\n- [Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2509.25541)\n\n- [Self-Challenging Language Model Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.01716)\n\n- [Self Rewarding Self Improving](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08827.pdf)\n\n- [SELF: Self-Evolution with Language Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.00533.pdf)\n\n- [Self-reasoning Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.14116)\n\n- [AgentGen: Enhancing Planning Abilities for Large Language Model based Agent via Environment and Task Generation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.00764)\n\n- [Reflexion: Language Agents with Verbal Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.11366.pdf)\n\n- [AdaPlanner: Adaptive Planning from Feedback with Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16653.pdf)\n\n- [Self-Refine: Iterative Refinement with Self-Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.17651.pdf)\n\n- [Learn-by-interact: A Data-Centric Framework for Self-Adaptive Agents in Realistic Environments](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.10893)\n\n- [DYSTIL: Dynamic Strategy Induction with Large Language Models for Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.03209.pdf)\n\n#### 3.2 Context\n\n#### 3.2.1 Memory Evolution\n\n- [Self-evolving Agents with reflective and memory-augmented abilities](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.00872)\n\n- [Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.19413)\n\n- [MemInsight: Autonomous Memory Augmentation for LLM Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.21760)\n\n- [Large Language Models Are Semi-Parametric Reinforcement Learning Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.07929)\n\n- [Agent Workflow Memory](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.07429)\n\n- [Richelieu: Self-Evolving LLM-Based Agents for AI Diplomacy](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.06813)\n\n- [Memory OS of AI Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.06326)\n\n#### 3.2.2 Prompt Optimization\n\n- [Large Language Models Are Human-Level Prompt Engineers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2211.01910)\n\n- [Large Language Models as Optimizers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.03409)\n\n- [Automatic Prompt Optimization with \"Gradient Descent\" and Beam Search](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.03495)\n\n- [PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.16427)\n\n- [REVOLVE: Optimizing AI Systems by Tracking Response Evolution in Textual Optimization](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.03092)\n\n- [DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.03714)\n\n#### 3.3 Tools\n\n- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16291)\n\n- [Alita: Generalist Agent Enabling Scalable Agentic Reasoning with Minimal Predefinition and Maximal Self-Evolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.20286)\n\n- [Advanced Tool Learning and Selection System (ATLASS): A Closed-Loop Framework Using LLM](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.10071)\n\n- [From Exploration to Mastery: Enabling LLMs to Master Tools via Self-Driven Interactions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.08197)\n\n- [ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.16789)\n\n- [ToolGen: Unified Tool Retrieval and Calling via Generation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.03439)\n\n- [AgentSquare: Automatic LLM Agent Search in Modular Design Space](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.06153.pdf)\n\n- [Retrieval Models Aren't Tool-Savvy: Benchmarking Tool Retrieval for Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01763)\n\n- [Towards Completeness-Oriented Tool Retrieval for Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.16089)\n\n#### 3.4 Architecture\n\n- [AgentSquare: Automatic LLM Agent Search in Modular Design Space](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.06153.pdf)\n\n- [Gödel Agent: A Self-Referential Framework for Agents Recursively Self-Improvement](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.04444.pdf)\n\n- [AlphaEvolve: A coding agent for scientific and algorithmic discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.13131.pdf)\n\n- [TextGrad: Automatic \"Differentiation\" via Text](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.07496.pdf)\n\n- [EvoFlow: Evolving Diverse Agentic Workflows On The Fly](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.07373.pdf)\n\n- [Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.02533.pdf)\n\n- [AFlow: Automating Agentic Workflow Generation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.10762.pdf)\n\n- [Automated Design of Agentic Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.08435.pdf)\n\n- [AutoFlow: Automated Workflow Generation for Large Language Model Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.12821.pdf)\n\n- [Language Agents as Optimizable Graphs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.16823.pdf)\n\n- [ScoreFlow: Mastering LLM Agent Workflows via Score-based Preference Optimization](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04306.pdf)\n\n- [FlowReasoner: Reinforcing Query-Level Meta-Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.15257.pdf)\n\n- [ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.09501.pdf)\n\n### 4. When to Evolve?\n\n#### 4.1 Intra-test-Time Self-Evolution\n\n- [AdaPlanner: Adaptive Planning from Feedback with Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16653.pdf)\n\n- [Reflexion: Language Agents with Verbal Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.11366.pdf)\n\n- [Self-Refine: Iterative Refinement with Self-Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.17651.pdf)\n\n- [TrustAgent: Towards Safe and Trustworthy LLM-based Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.01586.pdf)\n\n- [Self-Adapting Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.10943.pdf)\n\n- [Test-Time Training on Nearest Neighbors for Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.18466.pdf)\n\n- [Efficiently Learning at Test-Time: Active Fine-Tuning of LLMs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.08020.pdf)\n\n- [LADDER: Self-Improving LLMs Through Recursive Problem Decomposition](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.00735.pdf)\n\n- [TTRL: Test-Time Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.16084.pdf)\n\n#### 4.2 Inter-test-Time Self-evolution\n\n- [SELF: Self-Evolution with Language Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.00533.pdf)\n\n- [STaR: Bootstrapping Reasoning With Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2203.14465.pdf)\n\n- [Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.09629.pdf)\n\n- [SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04780.pdf)\n\n- [RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.20073.pdf)\n\n- [DYSTIL: Dynamic Strategy Induction with Large Language Models for Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.03209.pdf)\n\n- [Learning Like Humans: Advancing LLM Reasoning Capabilities via Adaptive Difficulty Curriculum Learning and Expert-Guided Self-Reformulation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08364.pdf)\n\n- [WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.02337.pdf)\n\n- [DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.11896.pdf)\n\n### 5. How to Evolve?\n\n#### 5.1 Reward-based Self-Evolution\n\n- [Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2509.25541)\n\n- [Self-Refine: Iterative Refinement with Self-Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.17651.pdf)\n\n- [Training Language Models to Self-Correct via Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.12917.pdf)\n\n- [PAG: Multi-Turn Reinforced LLM Self-Correction with Policy as Generative Verifier](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.10406.pdf)\n\n- [Confidence Improves Self-Consistency in LLMs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.06233.pdf)\n\n- [Self-ensemble: Mitigating Confidence Distortion for Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.01951.pdf)\n\n- [Self Rewarding Self Improving](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08827.pdf)\n\n- [Scalable Best-of-N Selection for Large Language Models via Self-Certainty](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.18581.pdf)\n\n- [Can Large Reasoning Models Self-Train?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.21444.pdf)\n\n- [Unsupervised Post-Training for Multi-Modal LLM Reasoning via GRPO](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.22453.pdf)\n\n- [Consistent Paths Lead to Truth: Self-Rewarding Reinforcement Learning for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.08745.pdf)\n\n- [SWE-Dev: Evaluating and Training Autonomous Feature-Driven Software Development](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16975.pdf)\n\n- [A Self-Improving Coding Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.15228.pdf)\n\n- [Feedback Friction: LLMs Struggle to Fully Incorporate External Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.11930.pdf)\n\n- [Unified Software Engineering agent as AI Software Engineer](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.14683.pdf)\n\n- [OTC: Optimal Tool Calls via Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.14870v1.pdf)\n\n- [AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.15651.pdf)\n\n- [SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.24119.pdf)\n\n- [Reward Is Enough: LLMs Are In-Context Reinforcement Learners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.06303.pdf)\n\n- [Generalist Reward Models: Found Inside Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.23235)\n\n#### 5.2 Imitation and Demonstration Learning\n\n- [STaR: Bootstrapping Reasoning With Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2203.14465.pdf)\n\n- [V-STaR: Training Verifiers for Self-Taught Reasoners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.06457.pdf)\n\n- [AdaSTaR: Adaptive Data Sampling for Training Self-Taught Reasoners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16322.pdf)\n\n- [Enhancing Large Vision Language Models with Self-Training on Image Comprehension](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.19716.pdf)\n\n- [Genixer: Empowering Multimodal Large Language Models as a Powerful Data Generator](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.06731.pdf)\n\n- [SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04780.pdf)\n\n- [Bridging the Gap: Self-Optimized Fine-Tuning for LLM-based Recommender Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.20771.pdf)\n\n- [Recursive Introspection: Teaching Language Model Agents How to Self-Improve](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.18219.pdf)\n\n- [Confidence Matters: Revisiting Intrinsic Self-Correction Capabilities of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.12563.pdf)\n\n#### 5.3 Population-based and Evolutionary Methods\n\n- [Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.22954.pdf)\n\n- [Nature-Inspired Population-Based Evolution of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01155.pdf)\n\n- [Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.01335)\n\n- [SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.19162)\n\n- [Language Models can Self-Improve at State-Value Estimation for Better Search](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.02878)\n\n- [elf-Evolving Multi-Agent Collaboration Networks for Software Development](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.16946)\n\n- [Multi-Agent Collaboration via Evolving Orchestration](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.19591)\n\n- [MDTeamGPT: A Self-Evolving LLM-based Multi-Agent Framework for Multi-Disciplinary Team Medical Consultation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.13856)\n\n- [Self-Evolving Multi-Agent Simulations for Realistic Clinical Interactions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.22678)\n\n\n### 6. Where to Evolve?\n\n### 6.1 General Domain Evolution\n\n- [Mobile-Agent-E: Self-Evolving Mobile Assistant for Complex Tasks](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.11733)\n\n- [WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.02337)\n\n- [WebEvolver: Enhancing Web Agent Self-Improvement with Coevolving World Model](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.21024)\n\n- [MobileSteward: Integrating Multiple App‑Oriented Agents with Self‑Evolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.16796)\n\n- [Generative Agents: Interactive Simulacra of Human Behavior](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.03442)\n\n- [Intelligent Virtual Assistants with LLM-based Process Automation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.06677)\n\n- [UI-Genie: A Self-Improving Approach for Iteratively Boosting MLLM-based Mobile GUI Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.21496)\n\n### 6.2 Specialized Domain Evolution\n\n- [Paper Copilot: A Self-Evolving and Efficient LLM System for Personalized Academic Assistance](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.04593)\n\n- [Richelieu: Self-Evolving LLM-Based Agents for AI Diplomacy](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.06813)\n\n- [AlphaEvolve: A Learning Framework to Discover Novel Alphas in Quantitative Investment](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.16196)\n\n- [MDTeamGPT: A Self-Evolving LLM-based Multi-Agent Framework for Multi-Disciplinary Team Medical Consultation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.13856)\n\n- [Self-Evolving Multi-Agent Simulations for Realistic Clinical Interactions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.22678)\n\n- [LLMs Can Simulate Standardized Patients via Agent Coevolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.11716)\n\n- [SEW: Self-Evolving Agentic Workflows for Automated Code Generation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.18646)\n\n- [AgentCoder: Multi-Agent-based Code Generation with Iterative Testing and Optimisation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.13010)\n\n- [A Self-Improving Coding Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.15228)\n\n- [QuantAgent: Seeking Holy Grail in Trading by Self-Improving Large Language Model](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.03755)\n\n- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16291)\n\n- [Learning to Be A Doctor: Searching for Effective Medical Agent Architectures](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.11301)\n\n- [Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.02957)\n\n- [Simulating Classroom Education with LLM-Empowered Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.19226)\n\n- [One Size Doesn’t Fit All: A Personalized Conversational Tutoring Agent for Mathematics Instruction](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.12633)\n\n### 7. Evaluation of Self-evolving Agents\n\n- [Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.21506.pdf)\n\n- [MCP-Universe: Benchmarking Large Language Models with Real-World Model Context Protocol Servers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.14704.pdf)\n\n- [DSBENCH: HOW FAR ARE DATA SCIENCE AGENTS FROM BECOMING DATA SCIENCE EXPERTS?](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2409.07703) `ICLR 2025`\n\n- [ScienceAgentBench: Toward Rigorous Assessment of Language Agents for Data-Driven Scientific Discovery](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.05080) `ICLR 2025` [Code](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002FScienceAgentBench)\n\n- [AppBench: Planning of Multiple APIs from Various APPs for Complex User Instruction](https:\u002F\u002Faclanthology.org\u002F2024.emnlp-main.856.pdf) `EMNLP 2025`\n\n- [MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.07095)\n\n- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.06770.pdf)\n\n- [OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.07972)\n\n- [Mobile-Agent-E: Self-Evolving Mobile Assistant for Complex Tasks](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.11733.pdf)\n\n- [WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2207.01206.pdf)\n\n- [WebArena: A Realistic Web Environment for Building Autonomous Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.13854.pdf)\n\n- [WebWalker: Benchmarking LLMs in Web Traversal](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.07572.pdf)\n\n- [ST-WebAgentBench: A Benchmark for Evaluating Safety and Trustworthiness in Web Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.06703.pdf)\n\n- [xbench: Tracking Agents Productivity Scaling with Profession-Aligned Real-World Evaluations](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2506.13651.pdf)\n\n- [BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.12516.pdf)\n\n- [Agent-SafetyBench: Evaluating the Safety of LLM Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.14470.pdf)\n\n- [LifelongAgentBench: Evaluating LLM Agents as Lifelong Learners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.11942.pdf)\n\n- [AgentBench: Evaluating LLMs as Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2308.03688.pdf)\n\n- [GAIA: a benchmark for General AI Assistants](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.12983.pdf)\n\n- [TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.14161.pdf)\n\n- [ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.16789.pdf) [code](https:\u002F\u002Fgithub.com\u002FOpenBMB\u002FToolBench)\n\n- [Seal-Tools: Self-Instruct Tool Learning Dataset for Agent Tuning and Detailed Benchmark](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.08355)\n\n- [API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs](https:\u002F\u002Faclanthology.org\u002F2023.emnlp-main.187\u002F)\n\n- [T-Eval: Evaluating the Tool Utilization Capability of Large Language Models Step by Step](https:\u002F\u002Faclanthology.org\u002F2024.acl-long.515\u002F) `ACL 2024`\n\n- [ACEBench: Who Wins the Match Point in Tool Usage?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.12851.pdf)\n\n- [StoryBench: A Multifaceted Benchmark for Continuous Story Visualization](https:\u002F\u002Farxiv.org\u002Fabs\u002F2308.11606.pdf)\n\n- [MultiAgentBench: Evaluating the Collaboration and Competition of LLM agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01935.pdf) `ACL 2025`\n\n- [Benchmarking LLMs' Swarm intelligence](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.04364.pdf)\n\n- [Beyond Prompts: Dynamic Conversational Benchmarking of Large Language Models](https:\u002F\u002Farxiv.org\u002Fhtml\u002F2409.20222v2.pdf)\n\n- [ACPBench: Reasoning about Action, Change, and Planning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.05669.pdf)\n\n- [PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Change](https:\u002F\u002Farxiv.org\u002Fabs\u002F2206.10498.pdf)\n\n\n\u003C!-- #### 7.1 Evaluation Goal and Metrics\n\n#### 7.2 Evaluation Benchmarks -->\n\n\n### 8. Future Directions\n\n#### 8.1 Personalize AI Agents\n\n\n- [Personalization of Large Language Models: A Survey](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.00027)\n\n- [AutoPal: Autonomous Adaptation to Users for Personal AI Companionship](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.13960)\n\n- [Personalize Your LLM: Fake it then Align it](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01048)\n\n- [A Survey of Personalization: From RAG to Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.10147)\n\n- [ROUGE: A Package for Automatic Evaluation of Summaries](https:\u002F\u002Faclanthology.org\u002FW04-1013\u002F)\n\n- [Bleu: a Method for Automatic Evaluation of Machine Translation](https:\u002F\u002Faclanthology.org\u002FP02-1040\u002F)\n\n#### 8.2 Generalization\n\n- [Position: Scaling LLM Agents Requires Asymptotic Analysis with LLM Primitives](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04358)\n\n\n- [Automating Safety Enhancement for LLM-based Agents with Synthetic Risk Scenarios](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.17735v1)\n\n- [TrustAgent: Towards Safe and Trustworthy LLM-based Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.01586)\n\n- [Foundational Challenges in Assuring Alignment and Safety of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.09932)\n\n#### 8.3 Safe and Controllable Agents\n\n- [Automating Safety Enhancement for LLM-based Agents with Synthetic Risk Scenarios](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.17735v1)\n\n- [AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.09780)\n\n- [Foundational Challenges in Assuring Alignment and Safety of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.09932)\n\n- [Unveiling Privacy Risks in LLM Agent Memory](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.13172)\n\n#### 8.4 Ecosystems of Multi-Agents\n\n- [MDTeamGPT: A Self-Evolving LLM-based Multi-Agent Framework for Multi-Disciplinary Team Medical Consultation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.13856)\n\n- [Collab-Overcooked: Benchmarking and Evaluating Large Language Models as Collaborative Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.20073)\n\n- [Self-Evolving Multi-Agent Collaboration Networks for Software Development](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.16946)\n\n- [MultiAgentBench: Evaluating the Collaboration and Competition of LLM agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01935)\n\n### Others\n\n- https:\u002F\u002Fgithub.com\u002FEvoAgentX\u002FEvoAgentX\n\n- https:\u002F\u002Fdeepmind.google\u002Fdiscover\u002Fblog\u002Falphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms\u002F\n    - https:\u002F\u002Fgithub.com\u002Fcodelion\u002Fopenevolve\n \n## 🔎 Citation\n\nTo cite the research [paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.21046), you could use the following BibTeX entries. \n\n```bibtex\n@misc{gao2025surveyselfevolvingagentspath,\n      title={A Survey of Self-Evolving Agents: On Path to Artificial Super Intelligence}, \n      author={Huan-ang Gao and Jiayi Geng and Wenyue Hua and Mengkang Hu and Xinzhe Juan and Hongzhang Liu and Shilong Liu and Jiahao Qiu and Xuan Qi and Yiran Wu and Hongru Wang and Han Xiao and Yuhang Zhou and Shaokun Zhang and Jiayi Zhang and Jinyu Xiang and Yixiong Fang and Qiwen Zhao and Dongrui Liu and Qihan Ren and Cheng Qian and Zhenghailong Wang and Minda Hu and Huazheng Wang and Qingyun Wu and Heng Ji and Mengdi Wang},\n      year={2025},\n      eprint={2507.21046},\n      archivePrefix={arXiv},\n      primaryClass={cs.AI},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.21046}, \n}\n```\n","# \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FCharlesQ9_Self-Evolving-Agents_readme_6e592f5bdbb5.png\" alt=\"Example Figure\" width=\"50\" height=\"50\" \u002F> 自进化智能体综述：通往人工超级智能之路\n\n# \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FCharlesQ9_Self-Evolving-Agents_readme_e5646a3a595b.jpg\" alt=\"Example Figure\" width=\"800\" height=\"140\" \u002F>\n\n\u003C!-- omit in toc -->\n## 📢 更新\n\n- **2025.07**: 我们发布了一个 GitHub 仓库，用于记录与推理经济（Reasoning Economy）相关的论文。欢迎引用或提交 Pull Request。\n\n\n---\n\n## 📜 目录\n- [自进化智能体综述：通往人工超级智能之路](#自进化智能体综述通往人工超级智能之路)\n    - [1. 引言](#1-引言)\n    - [2. 定义与基础](#2-定义与基础)\n    - [3. 进化什么？](#3-进化什么)\n        - [3.1 模型](#31-模型)\n        - [3.2 上下文](#32-上下文)\n            - [3.2.1 记忆进化](#321-记忆进化)\n            - [3.2.2 提示词优化](#322-提示词优化)\n        - [3.3 工具](#33-工具)\n        - [3.4 架构](#34-架构)\n            - [3.4.1 单智能体系统优化](#341-单智能体系统优化)\n            - [3.4.2 多智能体系统优化](#342-多智能体系统优化)\n    - [4. 何时进化？](#4-何时进化)\n        - [4.1 测试时间内自我进化](#41-测试时间内自我进化)\n        - [4.2 测试时间间自我进化](#42-测试时间间自我进化)\n    - [5. 如何进化](#5-如何进化)\n        - [5.1 基于奖励的自我进化](#51-基于奖励的自我进化)\n        - [5.2 模仿与示范学习](#52-模仿与示范学习)\n            - [5.2.1 自生成示范学习](#521-自生成示范学习)\n            - [5.2.2 跨智能体示范学习](#522-跨智能体示范学习)\n            - [5.2.3 混合示范学习](#523-混合示范学习)\n        - [5.3 基于种群与进化的方法](#53-基于种群与进化的方法)\n        - [5.4 交叉进化维度](#54-交叉进化维度)\n        - [5.5 自我进化方法的其他维度](#55-自我进化方法的其他维度)\n    - [6. 在哪里进化？](#6-在哪里进化)\n        - [6.1 通用领域进化](#61-通用领域进化)\n        - [6.2 专用领域进化](#62-专用领域进化)\n    - [7. 自进化智能体的评估](#7-自进化智能体的评估)\n        \u003C!-- - [7.1 评估目标与指标](#71-评估目标与指标)\n        - [7.2 评估范式](#72-评估范式)\n            - [7.2.1 静态评估](#721-静态评估)\n            - [7.2.2 短期自适应评估](#722-短期自适应评估)\n            - [7.2.3 长期终身学习能力评估](#723-长期终身学习能力评估) -->\n    - [8. 未来方向](#8-未来方向)\n\n\n---\n\n### 1. 引言\n\n- [Large language model agent: A survey on methodology, applications and challenges](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.21460.pdf)\n\n- [Advances and challenges in foundation agents: From brain-inspired intelligence to evolutionary, collaborative, and safe systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.01990.pdf)\n\n- [Toward a Theory of Agents as Tool-Use Decision-Makers](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.00886.pdf)\n\n- [A survey on self-evolution of large language models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14387.pdf)\n\n### 2. 定义与基础\n\n- [Curriculum Learning for Cooperation in Multi-Agent Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.11768.pdf)\n\n- [A Survey on Curriculum Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2010.13166)\n\n- [Self-Evolving Curriculum for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.14970)\n\n- [Continual lifelong learning with neural networks: A review](https:\u002F\u002Farxiv.org\u002Fabs\u002F1802.07569)\n\n- [Lifelong Learning of Large Language Model-based Agents: A Roadmap](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.07278)\n\n### 3. 演化什么？\n\n#### 3.1 模型\n\n- [Vision-Zero：通过战略性博弈自我对弈实现可扩展 VLM (视觉语言模型) 自我改进](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2509.25541)\n\n- [自我挑战的语言模型智能体 (Agent)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.01716)\n\n- [自我奖励自我改进](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08827.pdf)\n\n- [SELF：基于语言反馈的自我演化](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.00533.pdf)\n\n- [自我推理语言模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.14116)\n\n- [AgentGen：通过环境与任务生成增强基于大语言模型 (LLM) 的智能体规划能力](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.00764)\n\n- [Reflexion：具备语言强化学习能力的语言智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.11366.pdf)\n\n- [AdaPlanner：基于语言模型反馈的自适应规划](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16653.pdf)\n\n- [Self-Refine：基于自我反馈的迭代式优化](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.17651.pdf)\n\n- [Learn-by-interact：面向现实环境中自适应智能体的以数据为中心框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.10893)\n\n- [DYSTIL：利用大语言模型进行强化学习的动态策略归纳](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.03209.pdf)\n\n#### 3.2 上下文\n\n#### 3.2.1 记忆演化\n\n- [具备反思与记忆增强能力的自我演化智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.00872)\n\n- [Mem0：利用可扩展长期记忆构建生产级 AI 智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.19413)\n\n- [MemInsight：LLM 智能体的自主记忆增强](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.21760)\n\n- [大语言模型是半参数化强化学习智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.07929)\n\n- [智能体工作流记忆](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.07429)\n\n- [Richelieu：用于 AI 外交的自我演化 LLM 智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.06813)\n\n- [AI 智能体的记忆操作系统](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.06326)\n\n#### 3.2.2 提示词 优化\n\n- [大语言模型是人类水平的提示词工程师](https:\u002F\u002Farxiv.org\u002Fabs\u002F2211.01910)\n\n- [大语言模型作为优化器](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.03409)\n\n- [基于“梯度下降”和束搜索的自动提示词优化](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.03495)\n\n- [PromptAgent：利用语言模型进行战略规划实现专家级提示词优化](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.16427)\n\n- [REVOLVE：通过在文本优化中追踪响应演化来优化 AI 系统](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.03092)\n\n- [DSPy：将声明式语言模型调用编译为自我改进流水线](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.03714)\n\n#### 3.3 工具\n\n- [Voyager：基于大语言模型的开放式具身智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16291)\n\n- [Alita：以最小预定义和最大自我演化实现可扩展智能体推理的通用智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.20286)\n\n- [高级工具学习与选择系统 (ATLASS)：一种使用 LLM 的闭环框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.10071)\n\n- [从探索到精通：通过自主驱动交互使 LLM 掌握工具](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.08197)\n\n- [ToolLLM：促进大语言模型掌握 16000+ 真实世界 API](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.16789)\n\n- [ToolGen：通过生成实现统一的工具检索与调用](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.03439)\n\n- [AgentSquare：模块化设计空间中的自动 LLM 智能体搜索](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.06153.pdf)\n\n- [检索模型不善用工具：大语言模型的工具检索基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01763)\n\n- [面向大语言模型的完整性导向工具检索](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.16089)\n\n#### 3.4 架构\n\n- [AgentSquare：模块化设计空间中的自动 LLM 智能体搜索](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.06153.pdf)\n\n- [Gödel Agent：智能体递归自我改进的自指框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.04444.pdf)\n\n- [AlphaEvolve：用于科学和算法发现的代码智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.13131.pdf)\n\n- [TextGrad：基于文本的自动“微分”](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.07496.pdf)\n\n- [EvoFlow：即时演化多样化的智能体工作流](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.07373.pdf)\n\n- [多智能体设计：通过更优提示词和拓扑结构优化智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.02533.pdf)\n\n- [AFlow：自动化智能体工作流生成](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.10762.pdf)\n\n- [智能体系统的自动化设计](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.08435.pdf)\n\n- [AutoFlow：大语言模型智能体的自动化工作流生成](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.12821.pdf)\n\n- [将语言智能体视为可优化图](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.16823.pdf)\n\n- [ScoreFlow：通过基于分数的偏好优化掌握 LLM 智能体工作流](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04306.pdf)\n\n- [FlowReasoner：强化查询级元智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.15257.pdf)\n\n- [ReMA：通过多智能体强化学习学习 LLM 的元思维](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.09501.pdf)\n\n### 4. 何时进化？\n\n#### 4.1 测试时内自进化\n\n- [AdaPlanner: Adaptive Planning from Feedback with Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16653.pdf)\n\n- [Reflexion: Language Agents with Verbal Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.11366.pdf)\n\n- [Self-Refine: Iterative Refinement with Self-Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.17651.pdf)\n\n- [TrustAgent: Towards Safe and Trustworthy LLM-based Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.01586.pdf)\n\n- [Self-Adapting Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.10943.pdf)\n\n- [Test-Time Training on Nearest Neighbors for Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.18466.pdf)\n\n- [Efficiently Learning at Test-Time: Active Fine-Tuning of LLMs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.08020.pdf)\n\n- [LADDER: Self-Improving LLMs Through Recursive Problem Decomposition](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.00735.pdf)\n\n- [TTRL: Test-Time Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.16084.pdf)\n\n#### 4.2 测试间自进化\n\n- [SELF: Self-Evolution with Language Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.00533.pdf)\n\n- [STaR: Bootstrapping Reasoning With Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2203.14465.pdf)\n\n- [Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.09629.pdf)\n\n- [SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04780.pdf)\n\n- [RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.20073.pdf)\n\n- [DYSTIL: Dynamic Strategy Induction with Large Language Models for Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.03209.pdf)\n\n- [Learning Like Humans: Advancing LLM Reasoning Capabilities via Adaptive Difficulty Curriculum Learning and Expert-Guided Self-Reformulation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08364.pdf)\n\n- [WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.02337.pdf)\n\n- [DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.11896.pdf)\n\n### 5. 如何进化？\n\n#### 5.1 基于奖励的自进化\n\n- [Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2509.25541)\n\n- [Self-Refine: Iterative Refinement with Self-Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2303.17651.pdf)\n\n- [Training Language Models to Self-Correct via Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.12917.pdf)\n\n- [PAG: Multi-Turn Reinforced LLM Self-Correction with Policy as Generative Verifier](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.10406.pdf)\n\n- [Confidence Improves Self-Consistency in LLMs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.06233.pdf)\n\n- [Self-ensemble: Mitigating Confidence Distortion for Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.01951.pdf)\n\n- [Self Rewarding Self Improving](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08827.pdf)\n\n- [Scalable Best-of-N Selection for Large Language Models via Self-Certainty](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.18581.pdf)\n\n- [Can Large Reasoning Models Self-Train?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.21444.pdf)\n\n- [Unsupervised Post-Training for Multi-Modal LLM Reasoning via GRPO](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.22453.pdf)\n\n- [Consistent Paths Lead to Truth: Self-Rewarding Reinforcement Learning for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.08745.pdf)\n\n- [SWE-Dev: Evaluating and Training Autonomous Feature-Driven Software Development](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16975.pdf)\n\n- [A Self-Improving Coding Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.15228.pdf)\n\n- [Feedback Friction: LLMs Struggle to Fully Incorporate External Feedback](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.11930.pdf)\n\n- [Unified Software Engineering agent as AI Software Engineer](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.14683.pdf)\n\n- [OTC: Optimal Tool Calls via Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.14870v1.pdf)\n\n- [AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.15651.pdf)\n\n- [SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.24119.pdf)\n\n- [Reward Is Enough: LLMs Are In-Context Reinforcement Learners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.06303.pdf)\n\n- [Generalist Reward Models: Found Inside Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.23235)\n\n#### 5.2 模仿与示范学习\n\n- [STaR: Bootstrapping Reasoning With Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2203.14465.pdf)\n\n- [V-STaR: Training Verifiers for Self-Taught Reasoners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.06457.pdf)\n\n- [AdaSTaR: Adaptive Data Sampling for Training Self-Taught Reasoners](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.16322.pdf)\n\n- [Enhancing Large Vision Language Models with Self-Training on Image Comprehension](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.19716.pdf)\n\n- [Genixer: Empowering Multimodal Large Language Models as a Powerful Data Generator](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.06731.pdf)\n\n- [SiriuS: Self-improving Multi-agent Systems via Bootstrapped Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04780.pdf)\n\n- [Bridging the Gap: Self-Optimized Fine-Tuning for LLM-based Recommender Systems](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.20771.pdf)\n\n- [Recursive Introspection: Teaching Language Model Agents How to Self-Improve](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.18219.pdf)\n\n- [Confidence Matters: Revisiting Intrinsic Self-Correction Capabilities of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.12563.pdf)\n\n#### 5.3 基于群体与进化的方法\n\n- [Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.22954.pdf)\n\n- [Nature-Inspired Population-Based Evolution of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01155.pdf)\n\n- [Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.01335)\n\n- [SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.19162)\n\n- [Language Models can Self-Improve at State-Value Estimation for Better Search](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.02878)\n\n- [elf-Evolving Multi-Agent Collaboration Networks for Software Development](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.16946)\n\n- [Multi-Agent Collaboration via Evolving Orchestration](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.19591)\n\n- [MDTeamGPT: A Self-Evolving LLM-based Multi-Agent Framework for Multi-Disciplinary Team Medical Consultation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.13856)\n\n- [Self-Evolving Multi-Agent Simulations for Realistic Clinical Interactions](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.22678)\n\n\n### 6. 何处进化？\n\n### 6.1 通用领域进化\n\n- [Mobile-Agent-E：面向复杂任务的自进化移动助手](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.11733)\n\n- [WebRL：通过自进化在线课程强化学习训练 LLM Web 智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.02337)\n\n- [WebEvolver：利用协同进化的世界模型增强 Web 智能体的自我改进](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.21024)\n\n- [MobileSteward：集成多个面向 App 的智能体与自进化能力](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.16796)\n\n- [生成式智能体：人类行为的交互式模拟](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.03442)\n\n- [基于 LLM 流程自动化的智能虚拟助手](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.06677)\n\n- [UI-Genie：一种迭代提升基于 MLLM 的移动 GUI 智能体的自我改进方法](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.21496)\n\n### 6.2 特定领域进化\n\n- [Paper Copilot：一个用于个性化学术辅助的自进化高效 LLM 系统](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.04593)\n\n- [Richelieu：用于 AI 外交的基于 LLM 的自进化智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.06813)\n\n- [AlphaEvolve：一个在量化投资中发现新 Alpha 的学习框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F2103.16196)\n\n- [MDTeamGPT：一个用于多学科团队医疗会诊的基于 LLM 的自进化多智能体框架](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.13856)\n\n- [用于真实临床交互的自进化多智能体模拟](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.22678)\n\n- [LLM 可通过智能体协同进化模拟标准化病人](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.11716)\n\n- [SEW：用于自动化代码生成的自进化智能体工作流](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.18646)\n\n- [AgentCoder：基于多智能体的代码生成与迭代测试及优化](https:\u002F\u002Farxiv.org\u002Fabs\u002F2312.13010)\n\n- [一个自我改进的编程智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.15228)\n\n- [QuantAgent：通过自我改进的大语言模型在交易中寻找圣杯](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.03755)\n\n- [Voyager：一个基于大语言模型的开放式具身智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.16291)\n\n- [学习成为一名医生：探索有效的医疗智能体架构](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.11301)\n\n- [Agent Hospital：一个拥有可进化医疗智能体的医院模拟环境](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.02957)\n\n- [使用 LLM 赋能的智能体模拟课堂教育](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.19226)\n\n- [一刀切不可行：一种用于数学教学的个性化对话辅导智能体](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.12633)\n\n### 7. 自进化智能体的评估\n\n- [Mind2Web 2：使用“智能体即裁判”评估智能体搜索](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.21506.pdf)\n\n- [MCP-Universe：利用真实世界模型上下文协议服务器对大语言模型进行基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.14704.pdf)\n\n- [DSBENCH：数据科学智能体距离成为数据科学专家还有多远？](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2409.07703) `ICLR 2025`\n\n- [ScienceAgentBench：迈向对数据驱动科学发现的语言智能体进行严格评估](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.05080) `ICLR 2025` [代码](https:\u002F\u002Fgithub.com\u002FOSU-NLP-Group\u002FScienceAgentBench)\n\n- [AppBench：针对复杂用户指令规划来自不同 App 的多个 API](https:\u002F\u002Faclanthology.org\u002F2024.emnlp-main.856.pdf) `EMNLP 2025`\n\n- [MLE-bench：评估机器学习智能体在机器学习工程上的表现](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.07095)\n\n- [SWE-bench：语言模型能否解决真实世界的 GitHub Issues？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.06770.pdf)\n\n- [OSWorld：在真实计算机环境中对开放式任务的多模态智能体进行基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.07972)\n\n- [Mobile-Agent-E：面向复杂任务的自进化移动助手](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.11733.pdf)\n\n- [WebShop：迈向与基于现实环境的语言智能体进行可扩展的真实世界 Web 交互](https:\u002F\u002Farxiv.org\u002Fabs\u002F2207.01206.pdf)\n\n- [WebArena：一个用于构建自主智能体的真实 Web 环境](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.13854.pdf)\n\n- [WebWalker：LLM 在 Web 遍历中的基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.07572.pdf)\n\n- [ST-WebAgentBench：评估 Web 智能体安全性与可信度的基准](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.06703.pdf)\n\n- [xbench：通过职业对齐的真实世界评估追踪智能体生产力扩展](https:\u002F\u002Fwww.arxiv.org\u002Fabs\u002F2506.13651.pdf)\n\n- [BrowseComp：一个简单但具有挑战性的浏览智能体基准](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.12516.pdf)\n\n- [Agent-SafetyBench：评估 LLM 智能体的安全性](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.14470.pdf)\n\n- [LifelongAgentBench：将 LLM 智能体作为终身学习者进行评估](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.11942.pdf)\n\n- [AgentBench：将 LLM 作为智能体进行评估](https:\u002F\u002Farxiv.org\u002Fabs\u002F2308.03688.pdf)\n\n- [GAIA：通用 AI 助手的基准](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.12983.pdf)\n\n- [TheAgentCompany：在重要的真实世界任务上对 LLM 智能体进行基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.14161.pdf)\n\n- [ToolLLM：促进大语言模型掌握 16000+ 个真实世界 API](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.16789.pdf) [代码](https:\u002F\u002Fgithub.com\u002FOpenBMB\u002FToolBench)\n\n- [Seal-Tools：用于智能体微调和详细基准测试的自指令工具学习数据集](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.08355)\n\n- [API-Bank：工具增强型 LLM 的综合基准](https:\u002F\u002Faclanthology.org\u002F2023.emnlp-main.187\u002F)\n\n- [T-Eval：逐步评估大语言模型的工具利用能力](https:\u002F\u002Faclanthology.org\u002F2024.acl-long.515\u002F) `ACL 2024`\n\n- [ACEBench：谁在工具使用中赢得赛点？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.12851.pdf)\n\n- [StoryBench：连续故事可视化的多面基准](https:\u002F\u002Farxiv.org\u002Fabs\u002F2308.11606.pdf)\n\n- [MultiAgentBench：评估 LLM 智能体的协作与竞争](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01935.pdf) `ACL 2025`\n\n- [LLM 群体智能的基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.04364.pdf)\n\n- [超越提示词：大语言模型的动态对话基准测试](https:\u002F\u002Farxiv.org\u002Fhtml\u002F2409.20222v2.pdf)\n\n- [ACPBench：关于动作、变化和规划的推理](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.05669.pdf)\n\n- [PlanBench：评估大语言模型在规划及变化推理方面的可扩展基准](https:\u002F\u002Farxiv.org\u002Fabs\u002F2206.10498.pdf)\n\n\n\u003C!-- #### 7.1 评估目标与指标\n\n#### 7.2 评估基准 -->\n\n### 8. 未来方向\n\n#### 8.1 个性化 AI 智能体\n\n- [Personalization of Large Language Models: A Survey](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.00027)\n\n- [AutoPal: Autonomous Adaptation to Users for Personal AI Companionship](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.13960)\n\n- [Personalize Your LLM: Fake it then Align it](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01048)\n\n- [A Survey of Personalization: From RAG to Agent](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.10147)\n\n- [ROUGE: A Package for Automatic Evaluation of Summaries](https:\u002F\u002Faclanthology.org\u002FW04-1013\u002F)\n\n- [Bleu: a Method for Automatic Evaluation of Machine Translation](https:\u002F\u002Faclanthology.org\u002FP02-1040\u002F)\n\n#### 8.2 泛化能力\n\n- [Position: Scaling LLM Agents Requires Asymptotic Analysis with LLM Primitives](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.04358)\n\n\n- [Automating Safety Enhancement for LLM-based Agents with Synthetic Risk Scenarios](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.17735v1)\n\n- [TrustAgent: Towards Safe and Trustworthy LLM-based Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.01586)\n\n- [Foundational Challenges in Assuring Alignment and Safety of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.09932)\n\n#### 8.3 安全可控的智能体\n\n- [Automating Safety Enhancement for LLM-based Agents with Synthetic Risk Scenarios](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.17735v1)\n\n- [AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.09780)\n\n- [Foundational Challenges in Assuring Alignment and Safety of Large Language Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.09932)\n\n- [Unveiling Privacy Risks in LLM Agent Memory](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.13172)\n\n#### 8.4 多智能体生态系统\n\n- [MDTeamGPT: A Self-Evolving LLM-based Multi-Agent Framework for Multi-Disciplinary Team Medical Consultation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.13856)\n\n- [Collab-Overcooked: Benchmarking and Evaluating Large Language Models as Collaborative Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.20073)\n\n- [Self-Evolving Multi-Agent Collaboration Networks for Software Development](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.16946)\n\n- [MultiAgentBench: Evaluating the Collaboration and Competition of LLM agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2503.01935)\n\n### 其他\n\n- https:\u002F\u002Fgithub.com\u002FEvoAgentX\u002FEvoAgentX\n\n- https:\u002F\u002Fdeepmind.google\u002Fdiscover\u002Fblog\u002Falphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms\u002F\n    - https:\u002F\u002Fgithub.com\u002Fcodelion\u002Fopenevolve\n \n## 🔎 引用\n\n如需引用该研究[论文](https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.21046)，您可以使用以下 BibTeX 条目。\n\n```bibtex\n@misc{gao2025surveyselfevolvingagentspath,\n      title={A Survey of Self-Evolving Agents: On Path to Artificial Super Intelligence}, \n      author={Huan-ang Gao and Jiayi Geng and Wenyue Hua and Mengkang Hu and Xinzhe Juan and Hongzhang Liu and Shilong Liu and Jiahao Qiu and Xuan Qi and Yiran Wu and Hongru Wang and Han Xiao and Yuhang Zhou and Shaokun Zhang and Jiayi Zhang and Jinyu Xiang and Yixiong Fang and Qiwen Zhao and Dongrui Liu and Qihan Ren and Cheng Qian and Zhenghailong Wang and Minda Hu and Huazheng Wang and Qingyun Wu and Heng Ji and Mengdi Wang},\n      year={2025},\n      eprint={2507.21046},\n      archivePrefix={arXiv},\n      primaryClass={cs.AI},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.21046}, \n}\n```","# Self-Evolving-Agents 快速上手指南\n\n本项目是一个专注于 **自进化智能体** 的研究资源库，收录了通往人工超级智能（ASI）路径上的相关前沿学术论文，并按照演化维度进行了系统分类。\n\n## 环境准备\n\n本项目主要为文献索引与综述资源，无需复杂的编程环境或 GPU 依赖。\n\n- **系统要求**：Windows \u002F macOS \u002F Linux 均可\n- **前置依赖**：\n  - Git（用于克隆仓库）\n  - PDF 阅读器（用于阅读论文）\n  - 稳定的网络连接（用于访问 arXiv 等论文源）\n\n## 安装步骤\n\n通过 Git 将资源库克隆到本地即可使用：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fyour-username\u002FSelf-Evolving-Agents.git\n# 请将 your-username 替换为实际的仓库所有者名称\ncd Self-Evolving-Agents\n```\n\n## 基本使用\n\n本仓库通过 `README.md` 组织了大量论文链接，您可以按照以下结构快速定位感兴趣的研究方向：\n\n1. **查阅核心框架**\n   阅读 Introduction 与 Definitions 章节，建立对自进化智能体基础概念的理解。\n\n2. **按维度检索论文**\n   项目将研究内容划分为四大核心维度，方便针对性查阅：\n   - **What to Evolve (演化对象)**：涵盖模型层、上下文（记忆与提示词优化）、工具层及架构层的进化论文。\n   - **When to Evolve (演化时机)**：包含测试时与测试间两个时间维度的自我进化机制研究。\n   - **How to Evolve (演化方法)**：收录基于奖励、模仿学习及群体进化等具体技术实现的论文。\n   - **Where to Evolve (演化场景)**：区分通用领域与特定领域的进化应用。\n\n3. **追踪前沿动态**\n   查看 `Updates` 板块获取最新收录的论文信息，或访问 `Future Directions` 章节了解未来研究方向。","某金融科技团队正在开发一款智能投研助手，旨在自动分析市场动态并生成投资报告，需要处理海量非结构化数据和瞬息万变的市场环境。\n\n### 没有 Self-Evolving-Agents 时\n- 智能体表现僵化，遇到新出现的金融术语或突发市场事件时，往往因缺乏先验知识而给出过时或错误的分析。\n- 缺乏记忆进化机制，同样的数据提取错误（如财报指标计算失误）会在不同用户的对话中反复出现，无法自动修正。\n- 提示词和工具调用逻辑固定，当数据源 API 变更或出现更优的分析工具时，系统无法自适应调整，需工程师频繁手动介入维护。\n- 面对复杂的多步骤推理任务，无法根据任务难度动态调整思考链，导致长程任务经常半途而废或逻辑断裂。\n\n### 使用 Self-Evolving-Agents 后\n- 智能体具备“自我进化”能力，能通过上下文学习和自我反思，快速适应新的市场术语和突发情况，保持分析结果的时效性。\n- 系统建立了记忆进化机制，能从历史交互中汲取教训并优化内部知识库，确保同样的错误只犯一次，实现越用越聪明。\n- 智能体能根据反馈自动优化提示词和工具组合，当发现某个数据源不稳定时，能自主尝试切换备用工具，大幅降低了人工维护成本。\n- 通过架构层面的自我优化，智能体能根据任务复杂度动态调整推理路径，在处理长周期投研任务时的成功率与逻辑连贯性显著提升。\n\nSelf-Evolving-Agents 将 AI 从被动执行指令的静态程序转变为能自主学习、持续优化的动态系统，真正实现了智能体在复杂业务场景下的终身成长。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FCharlesQ9_Self-Evolving-Agents_7488001f.png","CharlesQ9","Jiahao Qiu","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FCharlesQ9_7e7ae868.png","\r\n\r\n","Princeton University","Princeton","jq3984@princeton.edu","https:\u002F\u002Fgithub.com\u002FCharlesQ9",1027,93,"2026-04-05T16:18:31","Apache-2.0",1,"未说明",{"notes":90,"python":88,"dependencies":91},"该项目为一个学术综述仓库，主要内容为‘自进化智能体’相关领域的论文列表和链接，并非可执行的软件工具或代码库。因此，README 中未提供具体的运行环境、硬件需求或依赖库信息。",[],[15,46],"2026-03-27T02:49:30.150509","2026-04-06T08:46:28.545796",[96],{"id":97,"question_zh":98,"answer_zh":99,"source_url":100},1763,"调查综述是否会收录 PiFlow 等新提出的代表性方法？","是的，维护者确认将在下一版本中加入 PiFlow。PiFlow 引入了一种根本不同的方法，通过原则感知的科学发现将领域专业知识与计算探索相结合，被视为自进化智能体领域的一种范式转变，适合作为“如何进化”分类下的新颖信息论方法。","https:\u002F\u002Fgithub.com\u002FCharlesQ9\u002FSelf-Evolving-Agents\u002Fissues\u002F2",[]]