# 12.2 推荐论文与阅读清单

本节提供一份“按主题组织”的阅读清单，帮助你从论文与工程资料中有重点地学习智能体。考虑到框架与产品更迭很快，本书避免维护“某平台官方文档链接合集”与“某年最新文章清单”，而是给出更稳定的筛选与阅读路径。

> **阅读建议**：
>
> 建议的阅读顺序：先从 **必读论文**(12.2.1) 和 **推理技术**(12.2.2) 建立基础概念，再根据你的专注方向选读相应主题。如果关注生产环境部署，**优先阅读安全与对齐部分** 的论文，特别是提示词注入与供应链安全相关的工作，这些直接影响系统稳定性。**工程资料选择**(12.2.8) 提供的优先级建议可帮助你避免被快速迭代的工具细节淹没。

## 12.2.1 必读论文

建议优先阅读这几类“奠基性工作”，它们对后续工程实践影响最大：

| 论文                                                                                                | 主题            | 年份   |
| ------------------------------------------------------------------------------------------------- | ------------- | ---- |
| [Chain-of-Thought Prompting Elicits Reasoning](https://arxiv.org/abs/2201.11903)                  | 逐步推理与提示词设计    | 2022 |
| [ReAct: Synergizing Reasoning and Acting](https://arxiv.org/abs/2210.03629)                       | 推理与行动结合       | 2022 |
| [Toolformer: Language Models Can Teach Themselves to Use Tools](https://arxiv.org/abs/2302.04761) | 工具学习          | 2023 |
| [Reflexion: Language Agents with Verbal Reinforcement Learning](https://arxiv.org/abs/2303.11366) | 反思与自我改进       | 2023 |
| [Generative Agents: Interactive Simulacra of Human Behavior](https://arxiv.org/abs/2304.03442)    | 生成式智能体与记忆驱动行为 | 2023 |

## 12.2.2 推理技术

| 论文                                                                               | 主题        | 年份   |
| -------------------------------------------------------------------------------- | --------- | ---- |
| [Tree of Thoughts: Deliberate Problem Solving](https://arxiv.org/abs/2305.10601) | 树搜索式思考    | 2023 |
| [Self-Consistency Improves CoT Reasoning](https://arxiv.org/abs/2203.11171)      | 多样采样与自一致性 | 2022 |

## 12.2.3 多智能体

| 论文                                                                                                           | 主题          | 年份   |
| ------------------------------------------------------------------------------------------------------------ | ----------- | ---- |
| [AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation](https://arxiv.org/abs/2308.08155) | 对话式多智能体     | 2023 |
| [MetaGPT: Meta Programming for Multi-Agent](https://arxiv.org/abs/2308.00352)                                | 面向软件开发的多智能体 | 2023 |
| [CAMEL: Communicative Agents for Mind Exploration](https://arxiv.org/abs/2303.17760)                         | 角色扮演与协作     | 2023 |

## 12.2.4 检索增强

| 论文                                                                                         | 主题         | 年份   |
| ------------------------------------------------------------------------------------------ | ---------- | ---- |
| [RAG: Retrieval-Augmented Generation](https://arxiv.org/abs/2005.11401)                    | 检索增强生成     | 2020 |
| [Self-RAG: Learning to Retrieve, Generate, and Critique](https://arxiv.org/abs/2310.11511) | 自适应检索与自我批判 | 2023 |

## 12.2.5 安全与对齐

| 论文                                                                                                                                            | 主题          | 年份   |
| --------------------------------------------------------------------------------------------------------------------------------------------- | ----------- | ---- |
| [Ignore This Title and HackAPrompt](https://arxiv.org/abs/2311.16119)                                                                         | 提示词注入攻防     | 2023 |
| [Constitutional AI: Harmlessness from AI Feedback](https://arxiv.org/abs/2212.08073)                                                          | 价值对齐与无害性    | 2022 |
| [Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned](https://arxiv.org/abs/2209.07858)              | 模型安全评估与对齐实践 | 2022 |
| [Give Us the Facts: Enhancing Large Language Models with Knowledge Graphs for Fact-aware Language Modeling](https://arxiv.org/abs/2306.11489) | 事实锚定与幻觉减少   | 2023 |
| [Large Language Model Supply Chain: Open Problems From the Security Perspective](https://arxiv.org/abs/2411.01604)                            | 供应链安全与依赖管理  | 2024 |
| [IH-Challenge: A Training Dataset to Improve Instruction Hierarchy on Frontier LLMs](https://arxiv.org/abs/2603.10521)                        | 指令层级鲁棒性训练   | 2026 |
| [Many-Tier Instruction Hierarchy in LLM Agents](https://arxiv.org/abs/2604.09443)                                                             | 多层级指令优先级建模  | 2026 |

## 12.2.6 智能体编程

| 论文                                                                                                    | 主题             | 年份   |
| ----------------------------------------------------------------------------------------------------- | -------------- | ---- |
| [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://arxiv.org/abs/2310.06770)  | 软件工程智能体评估      | 2023 |
| [AutoCoder: Enhancing Code Large Language Model with AIEV-Instruct](https://arxiv.org/abs/2405.14906) | 自动化代码生成增强      | 2024 |
| [SkillOpt: Executive Strategy for Self-Evolving Agent Skills](https://arxiv.org/abs/2605.23904)       | 可验证的 Skill 自进化 | 2026 |

## 12.2.7 综述论文

| 论文                                                                                           | 主题    | 年份   |
| -------------------------------------------------------------------------------------------- | ----- | ---- |
| [A Survey on Large Language Model based Autonomous Agents](https://arxiv.org/abs/2308.11432) | 智能体综述 | 2023 |
| [The Rise and Potential of LLM Based Agents: A Survey](https://arxiv.org/abs/2309.07864)     | 智能体发展 | 2023 |

## 12.2.8 工程资料如何选

建议把工程材料按“可迁移性”排序：

1. **方法论与流程**：评测设计、可观测性、权限边界、回归样例集、事故复盘模板。
2. **系统能力结构**：工具调用、结构化输出、缓存、检索、沙箱、审批与治理。
3. **工具/框架手册**：具体配置文件、命令行参数、SDK 示例。

这样阅读能避免被短周期的产品变化牵着走。

***

*注：论文链接多为 arXiv，版本可能随作者更新。*

***

**下一节**: [AGENTS.md 规范指南](/agentic_ai_guide/di-wu-bu-fen-fu-lu/12_appendix/12.3_agents_md.md)


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://yeasy.gitbook.io/agentic_ai_guide/di-wu-bu-fen-fu-lu/12_appendix/12.2_reading_list.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.