SkillPartly tested (needs an account)

Turn a research paper into deep notes

DeepPaperNote

Reads one research paper from a title, DOI, link or PDF and writes a long Obsidian note with method, results, figures and limits, not a summary.

Works with
Claude Code, Codex CLI
Good for
Daily work and research

What it does

DeepPaperNote helps you study one research paper properly. You give your assistant the paper (a title, a DOI, which is a paper's permanent ID, a web link, an arXiv ID or a PDF on your computer), and the skill gathers the paper, its details and its figures, then writes one long, structured note.

The skill first sorts the paper into one of ten types (for example a new method, a benchmark, a survey or a clinical study) and uses that type's set of questions. The note explains the problem, how the work answers it, what the evidence shows, what the paper does not prove and why it is worth keeping. Figures and tables are cut from the PDF and placed where the note discusses them. If only the abstract is available, the skill stops and asks you for the full paper instead of writing a note that only looks complete.

Notes are Markdown files made for Obsidian, a free note app that links notes together, so each paper becomes a page in your own research library. Each note ends with Takeaways, links to related papers already in your vault and a My Notes section that stays as you wrote it when the note is made again.

Three optional companion skills come with it: research-memory records your research directions so Takeaways fit your work, organize-wiki tidies the domain folders of your paper library, and paper-glossary makes notes for key terms.

Example prompts

  • "Generate a deep-reading note for this paper: <paper>"
  • "Turn this paper into an Obsidian note: <paper>"
  • "Build my research memory"
  • "Organize my paper library"
  • "Build glossary notes for this paper"

How to install

claude plugin marketplace add 917Dhj/DeepPaperNote
claude plugin install deeppapernote@deeppapernote
npx skills add 917Dhj/DeepPaperNote
python3 -m pip install PyMuPDF

How to use it

Give your assistant a paper title, DOI, URL, arXiv ID or local PDF and ask for a deep-reading note, or type /deeppapernote <paper>. On the first run it asks for your note language (English or Chinese) and where to save.

Check it works

The README says to ask your assistant: "Check DeepPaperNote's environment." It reports Python and PyMuPDF, your Obsidian settings and the optional add-ons. If a note stops partway, ask: "Where did this DeepPaperNote run stop?"

Common problems

Other assistants

  • Codex: install with npx skills add 917Dhj/DeepPaperNote and pick Codex; the repo also ships a .codex-plugin manifest. The README names only Claude Code and Codex. The repo description also names Cursor, Copilot and Gemini CLI, with no test status — https://github.com/917Dhj/DeepPaperNote

What to watch out for

  • Runs bundled Python scripts on your computer (paper lookup, PDF download, text and figure extraction, note checks, saving)
  • Looks up papers online: arxiv.org, export.arxiv.org, doi.org, api.crossref.org, api.openalex.org and api.semanticscholar.org, and downloads PDFs from arXiv, open-access or publisher links. The paper text goes to your assistant's AI provider
  • Loads by itself when you mention a paper (no disable-model-invocation). Before saving, the skill asks for an independent review of each note, by a fresh subagent where the assistant has one, so each note uses more model time
  • Writes each note and its figure images into your workspace or Obsidian vault, ~/.deeppapernote/config.json and work files in ~/.deeppapernote/runs/, which are never cleaned up. It does not silently overwrite a note, but a regenerated note replaces images of the same name
  • When no settings file exists, it reads your shell startup files (~/.zshrc, ~/.bashrc and others) for DEEPPAPERNOTE_ settings and asks before it saves them. An optional SEMANTIC_SCHOLAR_API_KEY improves lookups
  • Read-only Zotero lookup on 127.0.0.1:23119 in the default auto mode; set the Zotero mode to off to skip it
  • The plugin installs four skills. organize-wiki moves misfiled paper folders and rewrites links in your vault after you confirm a preview; it logs each change and can undo the last one
  • Needs Python 3.10 or newer with PyMuPDF (the macOS built-in python3 is 3.9)
  • npx skills add without -a asks which agents to install to; the skills CLI sends anonymous usage data unless DISABLE_TELEMETRY=1 or DO_NOT_TRACK=1 is set

Licence and author

Licence
MIT
Author
917Dhj
Source code
github.com/917Dhj/DeepPaperNote

How we tested it

We installed it on an empty test computer on 9 Nov 2026. Partly tested: installed in a clean test system, but its main features need an account or key we did not use.

Test level

Not run on a real paper. npx skills add skipped organize-wiki because its SKILL.md header is not valid YAML. python3 -m pip install PyMuPDF failed on Debian's system Python (externally managed); a virtual environment or uv avoids this.

Technical details of the test
Date
9 Nov 2026
Method
Reviewed by our team at commit 6cf41a6 (v3.0.0): README, CHANGELOG, the four SKILL.md files, plugin manifests and the scripts' network and file code. In a clean test machine (Claude Code 2.1.296, Python 3.11) the two claude plugin lines installed version 3.0.0, and npx skills add -a claude-code installed three of the four skills.

Checked again on 9 Nov 2026

Alternatives

  • co-researcher — literature reviews across many papers with checked citations; DeepPaperNote goes deep on one paper at a time
  • markitdown-mcp — only converts files such as PDFs to Markdown; it does not analyse the paper

FAQ

  • Q: Do I need Obsidian? A: No. You can save notes in the current folder; Obsidian just makes them easier to browse and link.
  • Q: Can it read many papers at once? A: No. The skill is built for one paper per run.
  • Q: Which languages can the notes use? A: English or Chinese, chosen on the first run and changeable per paper.
  • Q: Does it cost money? A: The skill is free (MIT). It uses your assistant's normal model time, and each note also gets a review pass. Paper lookups use free public services; a Semantic Scholar API key is optional.

Sources

  • github.com/917Dhj/DeepPaperNote
  • github.com/917Dhj/DeepPaperNote/blob/6cf41a6/skills/deeppapernote/SKILL.md
  • github.com/917Dhj/DeepPaperNote/blob/6cf41a6/CHANGELOG.md
  • github.com/917Dhj/DeepPaperNote/blob/6cf41a6/skills/deeppapernote/references/user-configuration.md
  • github.com/917Dhj/DeepPaperNote/releases/tag/v3.0.0
  • github.com/vercel-labs/skills