paper-reading

Read academic papers from local PDF files, arXiv URLs, or paper titles and generate structured reading notes in Chinese. Use when the user provides a PDF fil...

Security Scan

The skill appears to implement a straightforward paper-reading workflow, but the runtime instructions hardcode absolute local script paths that don't match the bundled files and allow writing arbitrary absolute output paths — these inconsistencies and broad filesystem write instructions warrant caution.

Purpose & Capability

Name/description match the included code: scripts fetch PDFs from arXiv or local paths and extract text with pdfplumber. Requested capabilities (downloading from arXiv, reading local PDFs, writing note files) are appropriate for a 'paper-reading' skill. However, the SKILL.md references an absolute installation path (/Users/yihe/.config/...) for executing the scripts while the bundle contains scripts relative to the skill; this mismatch is incoherent and should be fixed.

Instruction Scope

Instructions ask the agent to accept a local PDF path, download to a user-specified absolute PDF path, and save notes to a user-specified absolute path. Allowing arbitrary absolute output paths means the skill (when run) can overwrite arbitrary files if the user supplies such paths. The SKILL.md also hardcodes exact interpreter invocations pointing to /Users/yihe/... which is unlikely to exist for end users and could cause the agent to attempt to execute code from unintended locations.

Install Mechanism

No install spec is provided (instruction-only install); code files are bundled with the skill. There are no remote install URLs or archive extraction steps in the manifest. The only external network activity is downloading PDFs from arXiv (expected). Runtime may request installing pdfplumber via pip, which is normal for PDF parsing.

Credentials

The skill declares no environment variables, credentials, or config paths. The scripts only use local file I/O and HTTP(S) requests to arXiv — proportional to the stated purpose. No secrets are requested.

Persistence & Privilege

The skill is not always-enabled and does not request elevated platform-level privileges. It writes files only where the user-specified absolute paths point, which is normal for a note-generation tool but requires the user to choose safe paths.

What to consider before installing

This skill mostly does what it says: fetch arXiv PDFs, extract text with pdfplumber, and generate notes. Before installing or running it: 1) Fix the hardcoded script paths in SKILL.md — replace /Users/yihe/.config/... with relative or packaged script invocations (e.g., python ./scripts/fetch_pdf.py) so the agent runs the bundled code, not an arbitrary absolute path. 2) Be careful with PDF output and note output absolute paths you supply — the skill will write files to those exact paths and could overwrite sensitive files if given a path like /etc/hosts or ~/private/*. 3) Review the bundled scripts (fetch_pdf.py, read_pdf.py) yourself or run them in an isolated environment (container/VM) before granting access to your real filesystem. 4) Expect the skill to install pdfplumber via pip if missing; consider installing dependencies into a sandboxed environment. 5) If you need stronger assurance, ask the developer to remove hardcoded paths from SKILL.md and to add explicit safeguards (e.g., restrict writable directories, validate/deny dangerous output paths).

Read a paper and produce structured notes following the standard template.

User Input Required

Before starting, the user must provide:

1. 论文信息: 本地 PDF 路径、arXiv URL 或论文标题（三选一）
2. PDF 存储路径: 下载/保存 PDF 的绝对路径（如 /tmp/paper.pdf）
3. 笔记输出路径: 最终生成的笔记文件的绝对路径（如 /Users/xxx/notes/Paper Title.md）

Workflow

1. Fetch the PDF

Given a local path, arXiv URL, or paper title from the user:

python /Users/yihe/.config/opencode/skills/paper-reading/scripts/fetch_pdf.py <path_or_url_or_title> -o <pdf_output_path>

• Local path: validates existence, returns absolute path
• arXiv URL: downloads PDF (supports both /abs/ and /pdf/ URL formats)
• Paper title: searches arXiv API by title, downloads the most relevant result
• -o 参数指定 PDF 存储的绝对路径（由用户提供）
• Prints the fetched PDF path to stdout

2. Extract PDF Content

pip install pdfplumber  # if not already installed
python /Users/yihe/.config/opencode/skills/paper-reading/scripts/read_pdf.py <pdf_path> [-p <page_range>] [-o <extracted_text_path>]

• Extracts text and tables from the PDF
• Optionally specify page range (e.g. -p 1-10 or -p 3)
• -o 可选，指定提取文本的输出路径；不指定则输出到 stdout
• Outputs markdown-formatted text

3. Read and Analyze

Read the extracted text. For papers with complex layouts (columns, figures, equations), supplement extraction with direct PDF reading using the pdf tool if available, or ask the user to clarify specific sections.

4. Generate Notes

Produce notes following the standard template.

Note Structure (summary)

# 一、基本信息
1.paper：《标题》
2.github：链接或未知
3.会议：会议名或未知

# 二、文章理解
## 1. 研究背景与动机 (Motivation)
## 2. 核心问题 (Problem Statement)
## 3. 解决方法 (Methodology)
## 4. 实验结果 (Experiments)

Key Requirements

• Language: Chinese with English technical terms preserved
• Formulas: LaTeX format ($...$ inline, $$...$$ block)
• Images: preserve original image URLs with ![](url) format
• Key concepts: bold with **...**
• Deep dives / proofs / interview-level knowledge: wrap in :::info blocks
• PyTorch code: use python code blocks
• Strict 4-level heading hierarchy: # → ## → ### → ####
• Numerical results: use specific numbers to show improvements

5. Save Notes

Save the generated notes to the path specified by the user (e.g. /Users/xxx/notes/Paper Title.md).