Skip to content

docs: 补充 Paper Agent 蓝图的社区实现 #154

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
11XuX wants to merge 1 commit into from
Closed

docs: 补充 Paper Agent 蓝图的社区实现 #154

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions projects/01-paper-agent/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,6 +159,15 @@ outputs/paper-agent/<run-id>/

> 构建论文研读 Agent，接入 arXiv、Semantic Scholar 和 PDF 解析工具，使用分层 context builder 管理 query、候选论文、证据片段和生成草稿；为每条综述结论保留 paper/page/section 引用，设计 20 条主题检索和单篇精读 eval case，统计相关性、引用准确率、幻觉率和平均成本。

## 社区实现

- [agent-interview](https://github.com/11XuX/agent-interview)：按本蓝图实现的完整项目，同一业务用三种 Agent 形态各写一遍（写死的 workflow / 裸 ReAct / deepagents harness），附三形态对照跑分数据。

- 架构与本蓝图一一对应：Planner → Paper Search Tools → Paper Ranker → PDF Reader → Evidence Store → Synthesis Agent → Review Agent
- 数据源用 Europe PMC 的 `fullTextXML` 接口，直接返回带 `<sec><title>` 的结构化全文，**免去 PDF 解析**（PyMuPDF / Docling 那一层可以整块跳过）
- 21 个 commit 一步一个 LangGraph 概念，可 `git checkout` 逐步走，适合当 LangGraph 入门材料
- 实测结论：文献密集的题上写死的 workflow 引用密度是 agent 形态的 2 倍以上；交叉领域的题上 workflow 会撞死循环上限并**静默降级**（交出看似正常实则敷衍的短报告），agent 形态则能自适应但 token 成本随轮次二次增长

## 参考项目与资料

- [GPT Researcher](https://github.com/assafelovic/gpt-researcher)：开源深度研究 Agent，适合学习 research report pipeline。
Expand Down
16 changes: 16 additions & 0 deletions projects/04-end-to-end-projects/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -151,6 +151,22 @@
- RAG 与 tool use 如何结合。
- 文档/数据库/知识库连接器生态。

### [agent-interview](https://github.com/11XuX/agent-interview)

**定位**：同一个文献综述业务的三种 Agent 形态实现与对照，可当 LangGraph 入门材料。

**适合学习**：

- 同一业务下 workflow / ReAct / agent harness 三种形态的代码规模与成本差异
- LangGraph 的 Send 扇出、双回边、`interrupt` 与 checkpointer 恢复语义
- 引用可追溯的设计：证据同时保留 paper_id / section / quote，使"引文是否逐字来自原文"成为可用代码 100% 判定的硬约束

**重点看**：

- 三形态并列的目录结构 —— 同一业务一个要 18 个文件、一个 33 行，不对称本身是结论
- `nodes/review.py` 的两层校验：能用代码查死的（引用真伪、引文是否为原文子串）不交给模型
- `docs/eval-design.md` 的评测方法论：先枚举失败模式再倒推指标

## Web / Computer-Use Agent

### [BrowserGym](https://github.com/ServiceNow/BrowserGym)
Expand Down