📚 Academic Paper Downloader Skill

中文说明 (README.zh-CN.md)

A practical AI skill for searching papers from Elsevier/Scopus and then downloading full text by DOI/title with this strategy:

1. Use Scopus to get clean metadata (DOI, title, year, source, cited count).
2. Try download via Unpaywall first.
3. Optionally use Sci-Hub CLI as a fallback when Unpaywall is unavailable.

Designed for conversational AI use ("download a batch", "download latest papers") and automated agent workflows.

✨ Features

• 🔎 Scopus search by keywords/title/raw query
• 🧾 Structured metadata output (DOI/title/year/source/cited_by)
• ⬇️ Auto download by DOI
• 🟢 Unpaywall first (open-access priority)
• 🛟 Optional Sci-Hub fallback
• 🧠 Intent mapping for natural language:
  • few / "some" / "a few"
  • batch
  • as many as possible
  • latest / recent-year mode

🧩 Repository Structure

• SKILL.md - skill behavior and policy mapping
• agents/openai.yaml - agent UI metadata
• scripts/search_scopus.py - Scopus search utility
• scripts/download_open_access.py - DOI downloader (Unpaywall + fallback)
• scripts/topic_batch_download.py - end-to-end topic search + quantity/latest-aware download

🚀 Quick Start

1) Get API access

Elsevier / Scopus API key

1. Create an account on Elsevier Developer Portal: https://dev.elsevier.com/
2. Create an API key in your profile.
3. Make sure your key/account has access to Scopus Search API (depends on institutional entitlement).

Unpaywall email

Unpaywall API requires an email parameter. A real or virtual email can be used.

2) Configure environment variables

export ELSEVIER_API_KEY="your_elsevier_api_key"
export UNPAYWALL_EMAIL="your_email_or_virtual_email@example.com"

3) Run via script or AI conversation

Option A: Direct script

# Download a latest batch in a topic
python3 scripts/topic_batch_download.py \
  --keywords "pedestrian simulation" \
  --quantity-mode batch \
  --latest \
  --outdir ./downloads

# Download exactly 5 latest papers
python3 scripts/topic_batch_download.py \
  --keywords "pedestrian simulation" \
  --latest \
  --target 5 \
  --outdir ./downloads_latest_5

Option B: Ask your AI agent (example prompts)

• "Download a batch of pedestrian simulation papers."
• "Download 5 latest papers in building evacuation simulation."
• "Download as many latest crowd simulation papers as possible."

The skill maps these words into concrete CLI behavior (few/batch/max + latest filters).

🤖 Automated Agent Workflow

Use JSON output for pipelines:

python3 scripts/topic_batch_download.py \
  --keywords "building pedestrian evacuation simulation" \
  --latest \
  --quantity-mode batch \
  --json --out ./summary.json \
  --outdir ./downloads

Then parse summary.json for downloaded paths, statuses, and DOI lists.

🧷 Optional: Install Sci-Hub CLI fallback

uv tool install git+https://github.com/Oxidane-bot/scihub-cli.git

download_open_access.py fallback resolution order:

1. custom --scihub-cmd
2. local scihub-cli in PATH
3. uvx --from git+https://github.com/Oxidane-bot/scihub-cli.git scihub-cli

🔌 Use as a Codex Skill

Install to your skill directory:

git clone https://github.com/wdc63/sci-papers-downloder.git ~/.codex/skills/sci-papers-downloder

Restart your AI CLI/session so skill discovery refreshes.

⚖️ Legal & Ethics

• Unpaywall is used as the primary legal OA source.
• If you enable fallback sources, ensure your usage complies with local laws, institutional policy, and publisher terms.
• This repository is provided for research workflow automation only.

📄 License

MIT - see LICENSE.