paper-notion-summarizer
by lococaeco
Browser & AutomationBrief
Fetch paper metadata by title or arXiv/DOI link, create a deep structured summary, and post it as a Notion page.
Description
What This Skill Does
Fetches a research paper by arXiv link, DOI, or title, reads the full text, and writes a structured seminar-quality summary to a Notion page. The summary follows a fixed eight-section template covering method, experiments, ablations, and critical assessment. Output language matches the language of the user's request.
Unlike extractive summarizers, the agent reads the full PDF and writes an original analysis at seminar depth, including key equations and critical limitations.
When to Use It
- Archiving papers read during a literature review into Notion
- Preparing for a lab meeting by summarizing a new arXiv preprint
- Building a personal research database with structured paper notes
- Quickly getting up to speed on a paper outside your subfield
- Sharing a formatted paper breakdown with collaborators via Notion
Example Workflow
User asks: Summarize https://arxiv.org/abs/2301.12345 and save it to my Notion reading list
- 1
Runs extract_paper.py with the arXiv URL to produce /tmp/paper_extract.json
- 2
Reads the extracted JSON section by section (Abstract, Introduction, Method, Experiments, Conclusion)
- 3
Writes a structured summary JSON to /tmp/paper_summary.json following the eight-section template
- 4
Runs push_to_notion.py with the summary file and the target Notion parent page ID
A Notion page titled with the paper name, containing structured sections for problem statement, method, experiments, ablations, limitations, and overall assessment
Requirements
- NOTION_API_KEY environment variable or ~/.config/notion/api_key file
- NOTION_PARENT_PAGE_ID environment variable or --parent-page-id flag value
- Notion integration with access to the target parent page
- pypdf Python package (pip install pypdf) for arXiv PDF extraction
Installation
clawhub install lococaeco/paper-notion-summarizerRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.