ljg-paper

lijigang/ljg-skills · updated Apr 20, 2026

MDX-style export adds YAML metadata + attribution linking explainx.ai and this canonical listing URL.

$npx skills add https://github.com/lijigang/ljg-skills --skill ljg-paper
0 commentsdiscussion
summary

读论文不是做学术,是猎取思想。把别人的发现拆解成自己能用的认知。

skill.md

ljg-paper: 读论文

读论文不是做学术,是猎取思想。把别人的发现拆解成自己能用的认知。

格式约束

Org-mode 语法

  • 加粗用 *bold*(单星号),禁止 **bold**
  • 标题层级从 * 开始,不跳级

ASCII Art

所有图表用纯 ASCII 字符。允许:+ - | / \ > < v ^ * = ~ . : # [ ] ( ) _ , ; ! ' " 和空格。禁止 Unicode 绘图符号。

模板权威性

输出结构依据 references/template.org。禁止参考 ~/Documents/notes/ 中已有论文文件的章节结构——旧文件可能使用过期模板。

Denote 文件规范

  • 时间戳:date +%Y%m%dT%H%M%S
  • 可读时间:date "+%Y-%m-%d %a %H:%M"
  • 文件名:{时间戳}--paper-{简短标题}__paper.org
  • 输出目录:~/Documents/notes/

Org 文件头

#+title:      paper-{简短标题}
#+date:       [{YYYY-MM-DD Day HH:MM}]
#+filetags:   :paper:
#+identifier: {YYYYMMDDTHHMMSS}
#+source:     {URL 或来源描述}
#+authors:    {作者列表}
#+venue:      {发表场所/年份}

文件写入后报告路径。

红线(每条必须过)

  1. 口语检验 — 你会这样跟朋友介绍一篇论文吗?不会→改。学术腔是默认敌人
  2. 零术语 — 先用大白话落地,再顺带提术语名。如果必须用原文术语才能解释,说明还没懂
  3. 短词优先 — 能用两个字说的不用四个字。「本文提出了一种新的框架」→「他们做了个东西」
  4. 一句一事 — 每句只推一步
  5. 具体 — 名词看得见,动词有力气。形容词能砍就砍
  6. 开头给理由 — 问题部分的第一句让人想知道答案
  7. 不填充 — 删学术套话(「近年来随着...的发展」「值得注意的是」)。每句干活
  8. 信任读者 — 说一遍够了。不重复结论
  9. 诚实 — 论文有硬伤就说有硬伤。看不懂的部分说看不懂

写作原则

四条核心原则,决定文章是"活人在说话"还是"机器在汇报":

  1. 一个锚点撑全文 — 找到一个具象的中心隐喻(一张图、一个场景、一个动作),让所有概念围绕它生长。不是并列罗列五个概念,是一根绳子串起来。锚点在「翻译」开头就要出现,后续章节可以反复回到它
  2. 推理外显 — 模拟"一个人想明白的过程",而非呈现"想明白之后的结果"。用"既然A是B,那能不能C也是D?"带读者一起推。让读者觉得结论差一步就是自己想到的
  3. 变形替代定义 — 解释两个概念的关系时,把A连续变形成B,不要说"A和B是XX关系"。「把LSTM变形→看起来像ResNet」比「LSTM和ResNet是对偶的」有力十倍
  4. 落点在能用 — 给出"这意味着你可以___",而非"这让我们重新思考___"。读者读完要带走一个能动手的东西,不是一个值得沉思的感慨

工具箱(选用)

讲解论文时可以拿的工具,没有哪个是必须的:

  • 类比 — 承重的,方法的关键组件都能映射上。沿着类比走一遍方法
  • ASCII 图 — 展示组件关系、数据流、结构对比。读者有概念脚手架后再画
  • 餐巾纸速写 — 「以前这么想,现在应该这么想」的并排对比
  • 好问题 — 把论文解决的困境变成一个让外行也好奇的问题
  • 递进例子 — 从简单到复杂,一步步搭建理解
  • 反问入链 — 遇到隐含假设,用问题打开

执行

1. 获取内容

  • arxiv URL → WebFetch
  • PDF → Read(注意 pages 参数限制)
  • 本地文件 → Read
  • 论文名称 → WebSearch

确保拿到:标题、作者、摘要、核心方法、结果。

如果论文有一张承载全文核心思路的总览图(overview / architecture diagram,通常是 Figure 1),提取并保存到 ~/Documents/notes/images/,文件名 {identifier}--paper-{简短标题}-overview.png

判断标准:这张图让人一看就抓住论文在做什么。不是所有论文都有——没有就跳过,不要硬找。

提取方法:

  • arxiv → 访问 HTML 版(arxiv.org/html/...),找到图片 URL,WebFetch 下载
  • PDF → 截取含图页面保存为图片

2. 定位:它在解决什么?

找到那个真实的困境——某件事做不到、某个现象解释不通、某条路走不下去。用一段话讲清来龙去脉。

不是「本文提出了一种新的 XXX 框架」,是「大模型明明很聪明,为什么一问具体事实就开始胡说?」

3. 费曼:让外行懂

把论文的核心想法讲到一个不懂这个领域的聪明人能跟上。形式自由——类比、图、例子、递进讲解,选最适合这篇论文的方式。

开头先立锚点:找到一个具象的中心隐喻或画面,在翻译的第一段就亮出来。后面所有概念围绕这个锚点生长,不是并列罗列。

推理带着读者走:不要直接给结论。模拟"一步步想明白"的过程——"既然X是这样,那Y能不能也这样?"让读者觉得结论差一步就是自己想到的。

需要覆盖:

  • 它怎么做的(核心机制/方法)
  • 做出来效果如何(挑最说明问题的两三个结果)
  • 理解全文需要的钥匙概念(如果有)

费曼翻译部分的子标题按内容需要组织,不必固定。

4. 核心概念:把术语变成直觉

挑出论文中最关键的 1 至 3 个概念(方法名、架构组件、数学对象、新定义……),逐个拆解。

每个概念:

  • 一句话:这东西是什么,干什么用的
  • 类比或例子:让没接触过的人秒懂。解释两个概念的关系时,优先用"把A变形成B"而非"A和B是XX关系"——变形比定义有力
  • 为什么重要:少了它论文的逻辑链断在哪里

选概念的标准:读者如果不懂这个,后面的洞见和审稿就跟不上。已经在「翻译」里讲透的不重复选。

5. 洞见:思想结晶

整篇论文最值钱的往往就一个点——作者真正找到的那颗新结晶。

用一句话把它说出来。这句话应该让读者觉得「这个想法我可以带走」,而不是「哦,论文说了这么个事」。

检验标准:把这句话单独抽出来,脱离论文上下文,它还有没有力量?如果只是在复述论文结论,那不是洞见。洞见是你读完之后自己看到的那个东西——论文里未必直说,但逻辑指向它。

说不出来就重读第三步。如果论文确实没有思想火花,直说「这篇论文是工程改进,没有认知层面的新发现」。不要硬挤。

6. 博导审稿

换身份:这个方向上带了二十年研究生的博导。学生拿着论文来找你,你判断这东西值不值得认真对待。

用白话说,像在办公室跟学生聊:

  • 选题眼光:问题值不值得做?真缺口还是人造缺口?
  • 方法成熟度:巧劲还是蛮力?有没有更自然的做法被忽略?
  • 实验诚意:baseline 公不公道?消融到位没?数字经不经得起追问?
  • 写作功力:最该说清楚的地方有没有偷懒?
  • 判决:strong accept / weak accept / borderline / weak reject / strong reject,一句话理由

好的说好,差的说差在哪儿。

7. 启发:对我的提醒

落点在"能用",不在"能想"。给出"这意味着你可以___",而非"这让我们重新思考___"。

用三个视角试探连接,命中展开,没命中跳过,全没命中说「没有」:

  • 迁移:论文的某个机制/视角能移植升级我体系的某个零件吗?具体怎么接?
  • 混搭:论文的某个组件和我已有的东西组合能产生新东西吗?产出什么?
  • 反转:论文的做法和我的默认假设相反吗?该停下什么、开始什么?

8. 过红线

逐条扫红线。额外检查:

  • 破公式——否定式排比全文不超过两处,三段式改两项或四项
  • 变节奏——长短句交替
  • 杀金句——听起来像可引用的,重写
  • 查跳跃——逻辑每步可追

列修改清单确认后生成文件。

9. 生成 Org 文件

按 Denote 规范获取时间戳,读 references/template.org,写入 ~/Documents/notes/

验收

  • 问题勾人:让不懂的人也想知道答案
  • 有锚点:翻译部分有一个具象的中心隐喻,后续概念围绕它生长
  • 带着推:读者能感受到"一步步想明白"的过程,而非接收打包好的结论
  • 外行能跟:不懂这个领域的聪明人读完能复述核心思路
  • 博导像博导:有判断力有分寸,最后一句判决
  • 启发能动手:启发部分的落点是"你可以___",不是"值得思考___"
  • 零割裂感:读完像一个人在跟你说「我读了篇论文,发现了个有意思的事」
how to use ljg-paper

How to use ljg-paper on Cursor

AI-first code editor with Composer

1

Prerequisites

Before installing skills in Cursor, ensure your development environment meets these requirements:

  • Cursor installed and configured on your development machine
  • Node.js version 16.0+ with npm package manager (verify with node --version)
  • Active project directory or workspace where you want to add ljg-paper
2

Execute installation command

Execute the skills CLI command in your project's root directory to begin installation:

$npx skills add https://github.com/lijigang/ljg-skills --skill ljg-paper

The skills CLI fetches ljg-paper from GitHub repository lijigang/ljg-skills and configures it for Cursor.

3

Select Cursor when prompted

The CLI will show a list of available agents. Use arrow keys to navigate and space to select Cursor:

◆ Which agents do you want to install to?
│ ── Universal (.agents/skills) ── always included ────
│ • Amp
│ • Antigravity
│ • Cline
│ • Codex
│ ●Cursor(selected)
│ • Cursor
│ • Windsurf
4

Verify installation

Confirm successful installation by checking the skill directory location:

.cursor/skills/ljg-paper

Reload or restart Cursor to activate ljg-paper. Access the skill through slash commands (e.g., /ljg-paper) or your agent's skill management interface.

Security & Verification Notice

We perform automated surface-level scans (Gen AI Scanner, Socket, Snyk) during installation. These checks detect common vulnerabilities but do not guarantee complete security. Always review skill source code and verify the publisher's reputation before production use.

Skills execute code in your development environment. Always verify the publisher's identity, review recent commits, and test in isolated environments before production deployment.

List & Monetize Your Skill

Submit your Claude Code skill and start earning

GET_STARTED →

Use Cases

User Story & Requirements Generation

Create detailed user stories, acceptance criteria, and feature specs

Example

Generate user stories for 'password reset feature' with acceptance criteria, edge cases, and test scenarios

Reduce spec writing time by 50%, ensure comprehensive coverage

Competitive Analysis

Research competitors, compare features, identify gaps

Example

Analyze 5 competitor products, create feature comparison matrix, suggest differentiation opportunities

Complete competitive research in 2 hours instead of 2 days

Roadmap Prioritization

Evaluate features using frameworks (RICE, ICE, Kano) and create prioritized backlogs

Example

Score 20 feature ideas using RICE framework, generate prioritized roadmap with rationale

Make data-driven prioritization decisions faster

Stakeholder Communication

Draft PRDs, status updates, and stakeholder presentations

Example

Create executive summary of Q3 roadmap, monthly progress report, feature launch announcement

Save 3-5 hours/week on communication overhead

Implementation Guide

Prerequisites

  • Claude Desktop or compatible AI client
  • Access to product documentation and roadmap tools (Jira, Notion, etc.)
  • Understanding of product management frameworks (RICE, Jobs-to-be-Done, etc.)
  • Stakeholder contact information and communication channels

Time Estimate

30-60 minutes to see productivity improvements

Installation Steps

  1. 1.Install product management skill
  2. 2.Start with user story generation for known feature
  3. 3.Progress to competitive analysis: research 2-3 competitors
  4. 4.Use for roadmap prioritization: apply RICE/ICE scoring
  5. 5.Draft stakeholder communications and refine based on feedback
  6. 6.Build template library for recurring PM tasks
  7. 7.Share effective prompts with product team

Common Pitfalls

  • Not validating competitive research—verify facts before sharing
  • Accepting user stories without involving engineering team
  • Over-relying on frameworks without qualitative judgment
  • Not customizing outputs to company culture and communication style
  • Skipping stakeholder validation of generated requirements

Best Practices

✓ Do

  • +Validate research and competitive analysis with real data
  • +Collaborate with engineering when generating technical requirements
  • +Customize frameworks and templates to your company context
  • +Use skill for first drafts, refine with stakeholder input
  • +Document successful prompt patterns for PM tasks
  • +Combine AI efficiency with human judgment and intuition

✗ Don't

  • Don't publish competitive analysis without fact-checking
  • Don't finalize user stories without engineering review
  • Don't make prioritization decisions solely on AI scoring
  • Don't skip customer validation of generated requirements
  • Don't ignore company-specific context and culture

💡 Pro Tips

  • Provide context: company goals, constraints, customer feedback
  • Ask for alternatives: 'Show 3 ways to prioritize this roadmap'
  • Request stakeholder-specific formatting: 'Executive summary vs. engineering spec'
  • Use skill for 70% generation + 30% customization to company needs

When to Use This

✓ Use When

Use for user story writing, competitive research, roadmap prioritization, stakeholder communication, and PRD drafting. Best for reducing repetitive documentation and research work.

✗ Avoid When

Avoid for strategic product vision (requires deep customer empathy), pricing decisions (needs market and financial expertise), or when face-to-face customer discovery is more valuable than speed.

Learning Path

  1. 1Basic: user stories, feature specs, status updates
  2. 2Intermediate: competitive analysis, prioritization frameworks, PRDs
  3. 3Advanced: product strategy, go-to-market planning, OKR setting
  4. 4Expert: product vision, market positioning, business model innovation

Discussion

Product Hunt–style comments (not star reviews)
  • No comments yet — start the thread.
general reviews

Ratings

4.543 reviews
  • Lucas Perez· Dec 24, 2024

    Useful defaults in ljg-paper — fewer surprises than typical one-off scripts, and it plays nicely with `npx skills` flows.

  • Omar Gupta· Dec 16, 2024

    I recommend ljg-paper for anyone iterating fast on agent tooling; clear intent and a small, reviewable surface area.

  • Aditi Jackson· Dec 4, 2024

    Registry listing for ljg-paper matched our evaluation — installs cleanly and behaves as described in the markdown.

  • Camila Rahman· Nov 15, 2024

    ljg-paper is among the better-maintained entries we tried; worth keeping pinned for repeat workflows.

  • Hiroshi Li· Nov 11, 2024

    ljg-paper reduced setup friction for our internal harness; good balance of opinion and flexibility.

  • Meera Lopez· Nov 7, 2024

    Keeps context tight: ljg-paper is the kind of skill you can hand to a new teammate without a long onboarding doc.

  • Yash Thakker· Nov 3, 2024

    Registry listing for ljg-paper matched our evaluation — installs cleanly and behaves as described in the markdown.

  • Soo Ndlovu· Oct 26, 2024

    ljg-paper is among the better-maintained entries we tried; worth keeping pinned for repeat workflows.

  • Dhruvi Jain· Oct 22, 2024

    ljg-paper reduced setup friction for our internal harness; good balance of opinion and flexibility.

  • Camila Zhang· Oct 6, 2024

    Keeps context tight: ljg-paper is the kind of skill you can hand to a new teammate without a long onboarding doc.

showing 1-10 of 43

1 / 5