Skip to content

AlphaLab-USTC/ResearchClaw

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🦞 ResearchClaw

你的 AI 全能论文助手 | Your Complete AI Research Companion

OpenClaw 驱动 · Powered by OpenClaw

OpenClaw Skill Version arXiv License: MIT Demo


6 大核心能力,零 API Key,开箱即用的 AI 科研全流程助手。

6 core capabilities. No API keys. End-to-end AI research companion, out of the box.


🌐 在线演示 · 📖 安装 · ⚙️ 配置 · 🐛 Issues


✨ 6 大核心能力 | 6 Core Capabilities

📡 论文雷达

Paper Scout

每天从 arXiv 自动发现最新论文,按你的研究画像智能排序,推送 Top 5–10。

  • 🔍 覆盖 LLM/RL/多模态等方向
  • 🎯 个性化相关度评分
  • 📈 从你的反馈中自动学习
  • ⏰ 支持每日定时推送

触发: 推荐今日论文 ▶ 查看演示

📝 深度速读

Paper Reader

发一个 arXiv 链接,AI 读完论文,基于 Deep Note 7-section 框架 生成结构化阅读笔记。

  • 📖 Deep Note 框架:Metadata → Why-read → CRGP → Figures → Experiments → Insights → Next steps
  • 📊 含图表 URL、具体实验数字、消融分析
  • 📝 默认输出 Markdown(git-friendly,低 token 消耗)
  • 🎨 可选生成 HTML 页面(--html
  • 🇨🇳 中英双语混合

触发: arXiv 链接 / 帮我读一下 / DNL ▶ 查看演示

📋 论文列表

Reading List

维护你的论文待读/正在读/已读列表,一键生成 HTML 看板。

  • 📊 三栏看板:To Read / Reading / Done
  • ➕ 一句话加入待读
  • ✅ 一句话标记已读
  • 🔗 每篇都链接到 HTML 笔记

触发: 我的论文列表 / 加入待读 ▶ 查看演示

🧠 研究画像

Research Taste Profile

学习你的研究偏好,生成可视化研究画像页面,个性化一切推荐。

  • 📝 研究方向描述
  • 🌱 种子论文 & 关键词
  • 👤 关注作者白名单
  • 🔄 从对话反馈自动更新

触发: 我的研究画像 / 更新我的研究画像 ▶ 查看演示

💡 灵感生成器

Idea Generator

分析你已读的论文,找出研究空白,生成 3–5 个有可行性的研究灵感。

  • 🔗 跨论文连接分析
  • 🕳️ 自动识别研究缺口
  • 📋 每个灵感含问题陈述 + 方法 + 相关论文 + 可行性评估
  • ⭐ 时机分析

触发: 给我一些研究灵感 ▶ 查看演示

📄 论文写作

Paper Writer

从灵感到草稿到回复审稿,覆盖论文写作全流程。

  • 🗂️ 结构化大纲生成
  • ✍️ 逐章节起草
  • 🔍 自我审稿 & 找弱点
  • 💬 Reviewer 回复起草

触发: 论文大纲 / 写引言 / rebuttal ▶ 查看演示


🚀 一句话安装

把这句话发给你的 OpenClaw 代理:

帮我安装 ResearchClaw:https://raw.githubusercontent.com/AlphaLab-USTC/ResearchClaw/main/docs/install.md

就这一句话。 代理会自动下载、注册技能,无需任何手动操作。

还没有 OpenClaw? 先访问 openclaw.ai 安装,再运行上面这句话。


💬 快速上手示例 | Quick Start

📡 每日论文推荐

推荐今日论文

📡 今日论文推荐 | Daily Paper Scout ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🗓️ 2026-03-26 | 匹配兴趣: LLM · RL · Agentic AI

1️⃣ RLVR: Reinforcement Learning with Verifiable Rewards 👤 Zhang et al. (2026) 🏷️ cs.LG · cs.AI 💡 通过可验证奖励信号显著提升 LLM 数学推理能力 🎯 相关原因: 直接结合 RL + LLM reasoning,命中核心方向 🔗 https://arxiv.org/abs/2503.XXXXX ...


📝 深度速读 + Deep Note 笔记

帮我读一下这篇:https://arxiv.org/abs/2503.19823

📝 Deep Note 完成 | AutoRefine ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📌 AutoRefine: Search and Refine During Think 👤 Shi et al. | 📅 2025 | ⭐ 4/5 🔗 https://arxiv.org/abs/2503.19823

💡 核心发现: 在推理链中插入知识精炼步骤比单纯增加检索更有效 📊 关键数据: 2WikiMHQA EM 67.7, HotpotQA EM 55.3 ✨ 启发: 精炼检索到的知识比检索本身更重要

💾 笔记 → papers/2026-04-01_autorefine.md

想同时生成 HTML 页面?加 --html

DNL https://arxiv.org/abs/2503.19823 --html

📋 管理论文列表

加入待读 https://arxiv.org/abs/2503.19823

✅ 已加入待读:AutoRefine | 共 7 篇待读

我的论文列表

📋 阅读列表已更新 → 打开 reading-list.html 查看看板 To Read: 5 篇 · Reading: 2 篇 · Done: 12 篇


💡 灵感生成

给我一些研究灵感

💡 Research Idea 1: Adaptive Memory Retrieval for RL Agents 📋 Problem: 当前 RL agent 的记忆检索策略是静态的,无法根据任务难度动态调整... 🔧 Approach: 在 rollout 过程中,用 uncertainty estimation 决定何时检索记忆... 📚 Related: AutoRefine (推理时检索) · MemOCR (记忆布局感知) 🎯 Feasibility: 数据集现成,4×A100 × 3天,风险:检索延迟影响 rollout 速度


📄 论文写作

论文大纲:我想做一个 RL + Memory 的工作,针对 long-horizon agentic 任务

📄 论文大纲 | Paper Outline 📌 Working Title: MemRL: Memory-Augmented Reinforcement Learning for Long-Horizon Agents 🎯 Core Claim: 动态记忆检索机制显著提升 RL agent 在长程任务上的成功率 📍 Target Venue: NeurIPS 2026

1. Introduction ...

2. Related Work ...

(完整大纲展开)

rebuttal
[Reviewer 2: The comparison with baseline X is unfair because...]

📝 Rebuttal Draft Reviewer 2

R2.1: "The comparison with baseline X is unfair..." Response: We thank the reviewer for this concern. The comparison is fair because...


🎬 在线演示 | Live Showcase

访问 alphalab-ustc.github.io/ResearchClaw 查看完整演示:

案例 描述
每日信号简报 Paper Scout 多源聚合推送示例
论文阅读笔记 Paper Reader HTML 笔记示例
研究方向分析 Idea Generator 灵感生成示例
多智能体协作 Paper Writer 写作辅助示例

⚙️ 配置研究画像

安装后,直接告诉代理:

更新我的研究画像:我主要研究强化学习和多模态 agent,关注 RLVR 方向

或手动编辑 ~/.openclaw/workspace/research-claw-config.md

research_direction: "PhD in LLM reasoning and agentic memory systems"
seed_papers:
  - 2503.19823   # AutoRefine
keywords:
  - reinforcement learning
  - agentic memory
  - long-context reasoning
whitelist_authors:
  - Yaorui Shi
  - An Zhang

详见 配置指南


📦 每日自动推送

帮我设置每天早上 9 点自动推荐论文

代理会自动创建 OpenClaw cron 任务。✅


🔧 工作原理 | How It Works

用户: 推荐今日论文
  ↓
读取 research-claw-config.md(研究画像)
  ↓
调用 arXiv 公开 API(免费,无需 Key)
  ↓
按个性化偏好评分排序
  ↓
返回 Top 5–10 论文速览
  ↓
用户反馈 → 自动更新研究画像
用户: 帮我读一下 arxiv.org/abs/XXXXX
  ↓
提取 arXiv ID → 获取元数据 + HTML 版本
  ↓
分析论文内容(Introduction, Method, Experiments)
  ↓
按 Deep Note 7-section 框架提取结构化信息
  ↓
输出 Markdown 笔记文件(默认)+ 聊天摘要
  ↓
(可选)生成 HTML 页面

零第三方依赖:只需 OpenClaw + 网络,不需要任何额外 API Key。


🤝 Credits

  • 🦞 Built with OpenClaw — AI assistant platform
  • 📄 Paper data from arXiv (free open-access API)
  • 🧠 Created by AlphaLab@USTC (Yaorui Shi et al.)

🌍 English Quick Reference

Install (one line — send to your OpenClaw agent):

帮我安装 ResearchClaw:https://raw.githubusercontent.com/AlphaLab-USTC/ResearchClaw/main/docs/install.md

6 capabilities at a glance:

Say this What happens
推荐今日论文 Daily arXiv paper recommendations, ranked by your profile
帮我读一下 [arXiv link] Deep read → markdown deep note (+ optional HTML with --html)
加入待读 [link] Add paper to your reading list
我的论文列表 View & regenerate your HTML reading dashboard
更新我的研究画像 Update research taste profile + visual HTML page
给我一些研究灵感 Cross-paper analysis → 3–5 research idea proposals
论文大纲 [idea] Generate structured paper outline
写引言 / 写方法 Draft specific paper sections
rebuttal [comments] Draft point-by-point reviewer responses

No API key needed. Uses arXiv's free public API + OpenClaw's built-in tools. Zero third-party dependencies.



🔗 More from AlphaLab

Check out our other projects:

  • 🦀 PaperClaw — Automated paper collection and analysis toolkit
  • 🛠️ OhMyCode — Code generation and repair benchmark
  • 📚 Must-Read-LLM-Papers — Curated reading list for LLM research

Made with ❤️ for researchers · MIT License · alphalab-ustc.github.io/ResearchClaw

About

上朝式科研:AI-powered research workflow showcase

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors