How to use the Nature Literature Downloader agent skill

What the skill makes the agent do

An agent skill is a folder of instructions and scripts that an AI agent reads when a request matches the skill's description. This one covers academic full-text retrieval. The skill's own description says it is for lawful academic full text, CNKI institutional access, English open-access retrieval, publisher API access, institutional browser fallback, and supporting information downloads.

The skill routes each paper through a fixed order. Chinese literature always goes to CNKI. For English papers from Elsevier, Springer Nature, or IEEE with a configured API key, the publisher API is tried first. If that fails, or the publisher is something else, the agent checks legitimate open-access sources such as PMC, Unpaywall, publisher OA pages, and arXiv before any institutional browser route is considered.

• Open-access copies are checked first for most English papers.
• Publisher APIs are used with your own keys, if you have them.
• CNKI is used for Chinese papers, reusing your existing library login.
• The agent asks before downloading supplementary information (SI).
• It stops at login pages, CAPTCHAs, and paywalls instead of bypassing them.

Install the nature-downloader skill

Download the skill ZIP from this page and unzip it. There are two ways to make it available to your agent.

For a personal install, unzip the folder into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder instead. In the Claude apps you can also upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up requirements before the first download

The skill runs Node.js and Python code and needs Node.js 22 or newer and Python 3. Python packages are listed in requirements.txt in the skill folder.

CNKI, Web Access, and SI routes also need Chrome running with remote debugging enabled and an authenticated library session, plus the web-access CDP proxy. Publisher API keys for Elsevier, Springer Nature, or IEEE are optional, and you only need them if you want those publishers' APIs.

Library setup is only asked for when a route needs it. Give the agent the library resource URL you normally use, such as your library's database portal or a database page that redirects to login. The agent infers the access route from that URL and saves it. The skill ships with no school presets, so it will not guess your institution's domain.

• Node.js 22 or newer and Python 3
• Python packages from requirements.txt
• Chrome with remote debugging and an authenticated library session, for CNKI, Web Access, and SI routes
• Optional API keys: Elsevier, Springer Nature, IEEE

Add publisher API keys only when you need them

Credentials are configured lazily, the first time a route needs them. Keys are handled in the browser flow and the skill does not export cookies or passwords. You can enter a key interactively or pipe it in with --stdin, and check what is stored with show.

Springer Nature has separate services (openaccess, fulltext, metadata), so set the one you have access to. IEEE uses a full-text endpoint that your IEEE account provides.

Trigger the skill with natural requests

You do not need to type commands. Ask in plain language, naming the paper by DOI, exact title, or a clear topic. The agent will be asked first whether you want Supporting Information, and you should answer explicitly.

When you already have a legitimate open-access PDF link, you can give it to the agent. For an exact title where you only want open-access copies, the agent uses an OA-only search with no SI.

• "Download the full text PDF for DOI 10.xxxx/xxxxx and save it in my thesis folder."
• "Find an open-access PDF for the paper titled '<exact title>'. Do not include supplementary files."
• "Get the CNKI full text for this Chinese paper title and save the PDF."
• "Batch download these 12 DOIs and tell me where each file came from."

Understand the SI and format rules

Before any PDF, CAJ, HTML, XML, archive, or attachment is downloaded, the agent must know whether you want Supporting Information. The downloader runs with exactly one of --si or --no-si. Without a flag it stops with si_confirmation_required and creates no output folder.

If you ask for PDF only, the agent needs a real PDF link or a PDF response. If none exists it reports no_authorized_pdf_found or pdf_fetch_failed. It will not save a CAJ file, an HTML page, or a login page under a .pdf name.

• Use --si to include supporting files, --no-si to exclude them.
• Successful PDFs are marked open_access_downloaded in the manifest.
• Each download records its source, so batch collections stay auditable.

Safety notes: scripts and network access

This skill runs scripts and uses the network. It executes Node.js and Python code on your machine and can drive a Chrome session where you are logged in to your library. Only use it on a machine where you are comfortable running local scripts.

The skill is built to avoid bypassing access controls. It does not get around paywalls, logins, CAPTCHAs, or two-factor checks, and it does not read or export browser cookies, passwords, or session files. Papers that need an institutional login you do not have will be reported as missing, not worked around.

Avoid this skill on sensitive machines or in environments where a browser carrying institutional credentials should not be automated. Also note that a logged-in browser profile matters: if the controlled browser is a fresh profile with no login, the agent should ask you to authenticate in that browser rather than treat it as a permission problem.

• Runs scripts and uses the network
• Does not bypass paywalls, logins, CAPTCHAs, or two-factor verification
• Does not export cookies, passwords, localStorage, or session files

Tips for better results

Give exact titles or DOIs where you can, since topic searches return more ambiguous matches. If a publisher API key is configured but returns no full text, the agent will try open-access sources first and will ask once before using institutional Web Access. Keep the output folder per project so the manifest for each batch stays together.

If the config is stored elsewhere, for tests or isolated profiles you can point the skill at another folder with the LIT_DL_CONFIG_DIR variable. The default config file is ~/.config/lit-dl/school.json.

• Prefer exact DOIs or titles over vague topics.
• Keep one output folder per batch so manifests stay together.
• Check school configuration with the health command after changing library URLs.