zachary709/autopaperskill

AutoPaperSkill is a paper organization skill designed for AI agents. Its goal is not simply to “save a PDF” for you, but to turn a paper into a local resource package that can be searched, read, and reviewed over the long term.

It mainly does three things:

1. Find papers: Find reliable sources based on title, DOI, arXiv ID, conference year, or research direction.
2. Archive the paper: Put the PDF, metadata, source code links, project pages, images, and analysis results into a unified directory.
3. Reading Papers: Generate structured Chinese reports and refresh the homepage of the paper library papers.html.

The best way to use it is: clearly explain the goal to the agent, and have the agent call AutoPaperSkill to search, deduplicate, save, analyze, render, and refresh the homepage.

Who is it suitable for

If you often encounter these situations, AutoPaperSkill can be quite useful:

• The number of papers increases, and file names and folders start to spiral out of control.
• Check whether a paper has been saved to avoid duplicate downloads.
• Hope each paper will have a unified Chinese reading report.
• Want to search and preview papers on a local HTML homepage.
• You want an agent to help you organize your paper, but you want the results to be in a local file that can be reviewed.

Install the Agent

You don’t need to memorize the installation script yourself. Just send the following passage to the agent that supports skill installation:

If you have already cloned your repository locally, you can say:

If your agent environment does not support skill installation, you can also ask it to specify what capabilities or permissions are missing, then decide whether to switch to manual installation.

Preparing a thesis library

AutoPaperSkill requires a thesis library directory. This directory stores all thesis data packages and generates the homepage papers.html.

It is recommended that you prepare a dedicated directory first, for example:

Then he said to the agent:

If you don’t have a table of contents yet, you can also have the agent create the following:

Most common usage

Analyze a paper

If your agent supports sub-agents or isolation tasks, you can add a sentence:

Batch analysis of multiple papers

Only refresh the homepage

Only re-render the report

If the report content already exists, but the formula or style in the PDF is not displayed correctly, you can say:

Plagiarism Check but Not Save

What does the generated result look like?

Each paper usually forms an independent data package, structured similarly:

Among them:

• paper.pdf is the original text of the paper.
• metadata.json is information such as title, author, year, DOI, arXiv, source link, etc.
• analysis.json is the result of structured analysis.
• report.pdf is a Chinese reading report.
• images/ Save images extracted or rendered from the paper.
• sources/ Save the webpage, project page, or supplementary materials for review.

The root directory of the paper database also includes:

This is the local thesis homepage, which can be opened in a browser to search for and select papers, and preview the original text or report.

Recommended Work Habits