SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillspaper-researcher

paper-researcher

Writing & research

View repository

Updated 10-16 09:52

About this skill

Paper Researcher — Full-Process Paper Assistant

Pre-check

The first step of starting this Skill is to check whether wpsnote-cli is available:

wpsnote-cli status --json
  • Available → All subsequent WPS note operations use the wpsnote-cli command
  • Unavailable → Fall back to using MCP tools (theuser-wpsnote service) to operate WPS notes

CLI Tool Conventions (when wpsnote-cli is available)

wpsnote-cli <子命令> [参数]
# 复杂参数一律通过 --json-args 传入
wpsnote-cli <子命令> --json-args '{"key": "value"}'
Operation CLI Command
Search notes wpsnote-cli search-notes --query "关键词" --json
List notes wpsnote-cli list --json
Read full note text wpsnote-cli read --note_id <id> --json
Create note wpsnote-cli create --json-args '{"title":"标题","content":"<h1>...</h1>"}'
Insert block wpsnote-cli edit --json-args '{"note_id":"...","op":"insert","anchor_id":"...","content":"<p>...</p>"}'
Get tag list wpsnote-cli tags --json

Core Script

Search and download operations are completed through scripts/paper.py , path:<skill目录>/scripts/paper.py。

# 首次使用安装依赖
pip install arxiv markitdown pymupdf

# 搜索（默认 arXiv，推荐英文关键词）
python3 /path/to/scripts/paper.py search "transformer attention" --top 5
python3 /path/to/scripts/paper.py search "LoRA fine-tuning" --source openalex --top 5
python3 /path/to/scripts/paper.py search "diffusion model" --category cs.CV

# 下载 PDF 并转 Markdown
python3 /path/to/scripts/paper.py get 1706.03762 --markdown
python3 /path/to/scripts/paper.py get 1706.03762 --output ~/papers --markdown

Actual script path: read the scripts/paper.pyunder the directory of this SKILL.md, and before executing, first use ls to confirm the path exists.

PDF → Markdown Conversion Instructions

After adding the --markdown parameter, the script tries conversion in the following order and stops on success:

  1. markitdown(preferred): calls MarkItDown().convert(pdf_path), highest conversion quality, preserves structure
  2. pymupdf(alternative): extracts text page by page and concatenates it into Markdown in ## Page N format
  3. Both methods fail → print a warning and return an empty string (usually a scanned PDF)

The conversion result is saved as a file with the same name and .md extension, at the same path as the PDF, such as ~/papers/1706.03762.md。

After conversion is complete, use the Shell cat command to read the contents of the .md file, then perform the analysis.

Workflow

Scenario A: Search + Close Reading (most common)

1. 运行 paper.py search  →  展示结果，请用户选择
2. 用户确认后，运行 paper.py get <ID> --markdown
3. Shell 读取生成的 .md 文件：cat ~/papers/<ID>.md
4. 按分析框架输出结构化摘要
5. 存入 WPS 笔记（CLI 优先，见存档流程）

Scenario B: User directly provides a PDF path

1. 用 Read 工具读取 PDF（Cursor 内置支持）
2. 若内容为空 → 运行 paper.py get <id> --markdown 转换后再读
3. 分析并存档

Scenario C: Batch Literature Review

1. 多次搜索，收集论文 ID 列表
2. 批量执行 get --markdown 下载
3. 逐篇分析，最后输出横向对比表
4. 汇总写入 WPS 笔记的"文献综述"笔记

Analysis Framework

When close-reading a paper, output according to the following dimensions:

Dimension Content
Metadata Title, authors, institutions, year, journal/conference
Research problem What problem it solves, why it matters
Method Core technical approach, key innovations
Experiments Datasets, evaluation metrics, key numerical results
Conclusion Main findings, contributions to the field
Limitations Shortcomings acknowledged by the authors or future work
Personal evaluation Innovativeness, practicality, reproducibility

WPS Note Archiving Process

Use wpsnote-cli (preferred)

# 1. 先查是否已有同名笔记
wpsnote-cli search-notes --query "论文标题" --json

# 2a. 无同名笔记 → 新建
wpsnote-cli create --json-args '{
  "title": "论文标题",
  "content": "<h1>论文标题</h1><p><tag>#论文/NLP</tag></p>"
}'

# 2b. 有同名笔记 → 追加内容到末尾
wpsnote-cli edit --json-args '{
  "note_id": "<id>",
  "op": "insert",
  "anchor_id": "<last_block_id>",
  "content": "<p>追加内容</p>"
}'

Use MCP (when wpsnote-cli is unavailable)

Call the user-wpsnote MCP service's search_notes、create_note、edit_block and other tools to complete the same operations.

Note Content Template

For the note XML structure template, see references/note-template.md。

  • Single paper: tags #论文/[领域], such as #论文/NLP、#论文/CV
  • Overview: Tag #论文/综述 + #论文/[领域]

Data Source Selection

Scenario Recommended Data Source
AI/ML Papers arxiv(default)
Medicine, Biology, Social Sciences openalex(high open access rate)
Uncertain Start with arxiv; if few results, switch to openalex

arXiv Category Filter Reference:cs.AI cs.LG cs.CV cs.CL cs.NE stat.ML

Troubleshooting

Issue Solution
arxiv Package Missing pip3 install arxiv
PDF Download Failed The script automatically retries 3 times; if it still fails, directly visit https://arxiv.org/pdf/<ID> to download manually
Markdown Conversion Empty The PDF may be a scanned document; confirm that pip install pymupdf is installed and retry
No OpenAlex Results Switch to arxiv; or use more general English keywords
wpsnote-cli Unavailable Fall back to using MCP user-wpsnote service