Guide · Agent skill
How to use the arXiv Paper Search & Download agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The arXiv Paper Search & Download skill lets an AI agent search the arXiv preprint archive by topic or paper ID, list matching papers, save the PDFs into a local folder and write short summaries based on each abstract. It is aimed at students and researchers building a local reading list, and at teams keeping a shared literature folder for a project. By the end of this guide you will know what the skill does, how to install it, how to trigger it with plain requests and what to check for safety and limits.
arXiv Paper Search & Downloadwanshuiyin/Auto-claude-code-research-in-sleep · MIT Download the skill Quick start
- 1Download the skill ZIP. Get the arxiv skill ZIP from this page and unzip it on your computer.
- 2Place the folder. Move the arxiv folder into ~/.claude/skills/ for personal use, or into your project's .claude/skills/ folder.
- 3Or upload in the Claude apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP.
- 4Check Python 3 and network. Make sure python3 runs on your machine and that you can reach export.arxiv.org and arxiv.org.
- 5Ask for a search. Make a natural request such as search arxiv for a topic, or give a paper ID to fetch.
- 6Review the results table. Check the arXiv IDs, authors, dates and categories before choosing which papers to keep.
- 7Download the papers you need. Ask for a download of one paper or all results, and check the reported path and file size.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. This skill is named arxiv and is described as searching, downloading and summarizing academic papers from arXiv.
Once installed, the agent can search arXiv by topic or by paper ID, list the matches with authors, dates and categories, and save PDFs into a local folder such as papers/. It writes a short summary of each paper from its abstract and can add accepted papers to a research wiki if one is active in the project.
- Search arXiv by topic or by a bare arXiv ID such as 2301.07041 or cs/0601001
- Show results with arXiv ID, title, authors, date and category
- Save PDFs to papers/ by default, or to a folder you specify
- Skip files that already exist rather than overwriting them
- Update a research wiki if one is active in the project
Install the arXiv skill in Claude Code or the Claude apps
Download the ZIP of the skill from this page and unzip it. For a personal install, place the arxiv folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder instead.
In the Claude apps, you can upload the skill instead. Open Settings, go to Capabilities, then Skills, and upload the ZIP there.
- Personal install: ~/.claude/skills/arxiv/
- Project install: .claude/skills/arxiv/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need a special command. The skill is meant to trigger when your request matches its description, which includes phrases such as "search arxiv", "download paper", "fetch arxiv", "arxiv search", "get paper pdf", or a request to find and save papers from arXiv to the local paper library.
The source also documents argument-style overrides you can append to a request. Use them when you want a specific limit, folder or download behaviour.
- "Search arxiv for diffusion models for protein design and save the top 5 PDFs"
- "Fetch arxiv paper 2301.07041 and summarize it"
- "Get paper pdf for cs/0601001"
- Overrides: - max: 20 returns up to 20 results
- Overrides: - dir: literature/ saves PDFs to a custom folder
- Overrides: - download downloads the first result, and - download: all downloads every result
/arxiv "attention mechanism" - max: 20Examples of what the agent does
A typical search returns a numbered table. The agent then lists the full details for a chosen paper, including its full abstract, PDF link and abstract link. Each downloaded file is reported with its path and size. Downloads are checked so that a file under 10 KB is treated as an error and removed.
- Search results table with columns for #, arXiv ID, title, authors, date and category
- Downloaded: papers/2301.07041.pdf (842 KB)
- A short summary block with arXiv ID, authors, date, categories, abstract and local PDF path
- Final report such as: Found N papers for "query"
What the skill needs and how it works
The skill needs Python 3 and network access to export.arxiv.org and arxiv.org. It can use a bundled fetcher script, arxiv_fetch.py, if one is found in the ARIS repository's tools folder, in a local tools folder or in ~/.codex/skills/arxiv/. If none is found, it runs inline Python that queries arXiv's public API directly, so the fetcher is optional.
Downloads wait one second between consecutive PDFs. If arXiv returns HTTP 429, the skill retries once after five seconds.
- Python 3 installed and available as python3
- Network access to export.arxiv.org and arxiv.org
- Optional: the ARIS repository's tools/arxiv_fetch.py fetcher
Safety notes: scripts and network use
This skill runs scripts and uses the network. The inline fallback runs Python code that the agent writes into a shell command, and the fetcher script runs when present. Read the SKILL.md file and any script before you install, and run the skill only in folders where you are comfortable seeing new PDFs appear.
Respect copyright. The skill does not check licences, and terms vary by paper, so check the terms before redistributing any downloaded PDF. Avoid very large bulk downloads, because the skill adds only a one-second delay and arXiv may rate-limit heavy use.
- Review SKILL.md and any bundled scripts before first use
- The skill only queries arXiv; it does not reach paywalled or other journal sites
- Existing PDFs at the same path are not overwritten
- If the arXiv API is unreachable, the skill reports the error and suggests a fallback
Tips for a clean literature folder
Keep one folder per project and pass it with the dir override so that papers from different topics do not mix. Start with a small max value, check the table, then download only the papers you need. If you use a research wiki, the skill adds each accepted paper with its arXiv ID, title, authors, abstract URL, PDF URL, local path and source query.
When you want a broader review across sources, the skill suggests follow-up skills such as /research-lit for multi-source review and /novelty-check to test an idea against the papers you have found.
- Use - dir: to separate projects
- Start with a small - max: value and download selectively
- Keep the arXiv ID with each note, since it is needed for citations
- Ask for a specific paper by ID when you already know it
What people use it for
- Building a local reading list on a new research topic, saved into a project papers/ folder
- Fetching a specific arXiv paper by ID and getting a short summary based on its abstract
- Keeping a shared literature folder for a research team, with each paper's arXiv ID recorded
- Adding accepted papers to an active research wiki in the project
Frequently asked questions
Is the arXiv agent skill free?
The skill is released under the MIT licence, as stated on its listing. The arXiv papers themselves are governed by their own copyright and licence terms, which the skill does not check.
What do I need to run the arXiv skill?
You need Python 3 and network access to export.arxiv.org and arxiv.org. The bundled fetcher script arxiv_fetch.py is optional, because the skill falls back to inline Python that calls arXiv's public API.
How do I install the arXiv skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings, Capabilities, Skills.
Can the skill download papers from journals other than arXiv?
No. The skill only queries arXiv's API and PDF links, so paywalled or non-arXiv journals are outside its scope.
Why did a download fail or get skipped?
The skill rejects files under 10 KB because they are usually error pages, and it never overwrites an existing PDF at the same path. If arXiv returns HTTP 429, it retries once after five seconds. Heavy downloading can trigger rate limits.
How does this compare with searching arXiv directly?
The skill lists results in a table, saves PDFs into a named folder and writes abstract-based summaries in one request. Searching arXiv directly gives you the site's full interface and features that the skill does not describe.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits students and researchers who want a local, organised reading list from arXiv and a quick abstract-based summary of each paper. It is a poor fit if you need paywalled or non-arXiv journals, want to redistribute PDFs without checking licences, or plan very large bulk downloads.
Download the skill