liming-ai/paper-to-note

liming-ai 资料

最近记录的来源活动: 2026年5月25日 22:33

SkillsMP 收录数据更新: 2026年10月26日 11:37

已收集 skills: 1

GitHub 星标: 4
GitHub Forks: 0

仓库 README

仓库 README 来源：README.md

查看 README 源文件

展开完整 README

# Paper-to-Note: AI Agent Skill for Academic Paper Reading Notes An AI agent skill that generates high-quality, structured reading notes for academic papers. Designed for use with Claude Code, Cursor, Codex, and other AI coding assistants that support the skill/agent framework. ## What It Does Given a paper (PDF, arXiv link, or title), the skill instructs an AI agent to: 1. **Download and read** the full paper 2. **Extract figures** from arXiv LaTeX source by default (PDF crops only as fallback, with automatic SVG/PNG conversion) 3. **Search for source code** on GitHub (mandatory — never assumes code is unavailable) 4. **Generate structured Chinese notes** with 5 sections: Motivation, Idea, Method, Experimental Setup, Results 5. **Write Python/PyTorch pseudocode** based on actual source code (not paper abstractions) 6. **Run parallel quality review** (Format + Content + Source Code reviewers) 7. **Save to Obsidian vault** with proper frontmatter, tags, and figure embeds ## Output Quality - All math formulas in LaTeX - Pseudocode reflects real implementation (verified against source code) - Code-to-paper mapping table with commit SHA anchoring - Figures extracted as original arXiv source assets when available (not blurry full-page screenshots) - Parallel 3-reviewer quality check catches factual errors and structural gaps ## Installation ### For Claude Code ```bash # Clone to your skills directory git clone https://github.com/liming-ai/paper-to-note.git ~/.claude/skills/paper-to-note # Copy shared resources (if not already present) cp -rn ~/.claude/skills/paper-to-note/_shared ~/.claude/skills/_shared ``` ### For Cursor / Codex / Other Runtimes ```bash # Clone to the shared agents skills directory git clone https://github.com/liming-ai/paper-to-note.git ~/.agents/skills/paper-to-note # Copy shared resources (if not already present) cp -rn ~/.agents/skills/paper-to-note/_shared ~/.agents/skills/_shared ``` ### Dependencies ```bash # Required: figure extraction from PDF pip install pymupdf Pillow # Required: Obsidian vault management # Install obsidian-cli: https://github.com/Yakitrak/obsidian-cli brew install yakitrak/yakitrak/obsidian # macOS # Or download from GitHub Releases for other platforms # Required: GitHub API access (for source code search & commit anchoring) # Install gh CLI: https://cli.github.com/ brew install gh # macOS # apt install gh # Linux # Optional: source PDF/EPS→SVG vector conversion (higher quality figures) brew install pdf2svg # macOS # apt install pdf2svg # Linux # Optional: EPS source figure conversion fallback brew install ghostscript imagemagick ``` ### Configuration (IMPORTANT) After installation, open `SKILL.md` and update the **User Configuration** section at the top: | Variable | What to set | |----------|-------------| | `VAULT_NAME` | Your Obsidian vault name (as shown in Obsidian app) | | `VAULT_PATH` | Absolute path to your vault (e.g. `~/Documents/my_papers`) | | `PAPERS_DIR` | Where to store downloaded PDFs (e.g. `~/papers`) | | `SKILLS_DIR` | Path to paper-to-skill output (optional, can ignore if not using paper-to-skill) | Then create the required directories: ```bash mkdir -p $PAPERS_DIR mkdir -p $VAULT_PATH/notes mkdir -p $VAULT_PATH/files ``` Verify obsidian-cli can find your vault: ```bash obsidian vault="YOUR_VAULT_NAME" folders ``` ## Directory Structure ``` paper-to-note/ ├── SKILL.md # Main skill definition (instructions for the AI agent) ├── agents/ │ └── paper-to-note-reviewer.md # Quality reviewer agent definition ├── scripts/ │ ├── extract_figures.py # Figure extraction tool (arxiv/PDF/compose/auto-width modes) │ └── calibrate_widths.py # Vault-wide figure width and centering audit/fix tool ├── _shared/ # Shared infrastructure with paper-to-skill │ ├── commit-anchor.md # Commit SHA anchoring format spec │ ├── pseudocode-rules.md # Pseudocode quality rules │ ├── known-categories.md # Obsidian category taxonomy │ └── reviewer-base.md # Reviewer output format & severity definitions └── README.md ``` ## Usage Once installed, the skill is automatically triggered when you ask the AI agent to read a paper: ``` > 帮我读一下这篇论文 https://arxiv.org/abs/2504.xxxxx > Read this paper and generate notes: [paper title] > 帮我做一下这篇论文的笔记 [PDF path] ``` ### Output Location (Customizable) By default, notes are saved to (paths configurable in `SKILL.md`): - Notes: `$VAULT_PATH/notes/<Category>/<PaperTitle>.md` - Figures: `$VAULT_PATH/files/<Category>/<PaperTitle>/` See the **Configuration** section below to set your paths. ## Figure Extraction Modes The `scripts/extract_figures.py` supports three modes: ```bash # 1. arxiv source (preferred — extracts original high-res figures) python scripts/extract_figures.py --arxiv 2504.12345 ./output_dir # 2. PDF crop (for non-arxiv papers) python scripts/extract_figures.py --pdf paper.pdf --crop ./output_dir \ --figures "fig1:4:72,48,540,370" "table1:5:100,490,520,610" # 3. Recommend adaptive embed widths python scripts/extract_figures.py --auto-width ./output_dir # 4. Compose grouped subfigures into one SVG python scripts/extract_figures.py ./output_dir \ --compose "fig3_group:row:panel_a.svg,panel_b.svg,panel_c.svg" ``` Use `scripts/calibrate_widths.py --auto-center --tolerance 0` to audit existing notes for width drift, broken figure references, and uncentered `<img>` embeds. ## Customization ### Change Language The skill generates Chinese notes by default. To change: - Modify the "Language Rules" section in `SKILL.md` ### Add Categories Edit `_shared/known-categories.md` to add your research categories. ### Skip Obsidian CLI If you don't use Obsidian or prefer plain markdown files, you can modify Step 5 in `SKILL.md` to use the `Write` tool directly instead of `obsidian` commands. The note content format remains the same. ## Related - **paper-to-skill**: A companion skill that generates implementation-ready engineering manuals from papers (code-level mapping, porting checklists, module interface contracts). Best workflow: use `paper-to-note` first to understand the paper, then `paper-to-skill` to extract engineering details. ## License MIT

正在加载仓库 README…

这个仓库中的 skills

高校图书馆学教师

1 个职业分类 · 已分类 100%

skill

职业分类

描述

更新

paper-to-note

高校图书馆学教师

Use when reading an academic paper or paper URL and saving structured Chinese notes with metadata/assets to Obsidian.

原文语言：英语

2026-05-25

已展示 1 / 1 个已收集 Skill。

用 AI 助手安装

复制这段 Prompt，发给你正在使用的 AI 助手。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/liming-ai/paper-to-note.