SkillsAdd

wentorai/research-claw

sci-papers-downloder

Search academic papers (OpenAlex no-key by default, or Scopus with a built-in key) and download full-text PDFs concurrently — OA direct link → Unpaywall → Sci-Hub fallback (on by default). Quantity-aware and latest-aware.

First seen Jul 17, 2026

Installation

$ npx skills add wentorai/research-claw --skill sci-papers-downloder
View on skills.sh ↗ GitHub repository ↗

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

agent-browser Shared topics
vercel-labs/agent-browser

Browser automation CLI for AI agents. Use when the user needs to interact with websites, includ…

810.4K installs
azure-diagnostics Shared topics
microsoft/azure-skills

Debug Azure production issues on Azure using AppLens, Azure Monitor, resource health, and safe …

568.9K installs
azure-validate Shared topics
microsoft/azure-skills

Pre-deployment validation for Azure readiness. Run deep checks on configuration, infrastructure…

567.7K installs
azure-rbac Shared topics
microsoft/azure-skills 489K installs
supabase-postgres-best-practices Shared topics
supabase/agent-skills

Postgres best practices maintained by Supabase, for Postgres running anywhere. Load this skill …

391.6K installs
building-native-ui Shared topics
expo/skills 59.2K installs

Also in this package

Other skills from wentorai/research-claw · top by installs.

npx skills add wentorai/research-claw
ppt-master 20 Citation Styles 13 academic-writing 13 multi-search-engine 13 imap-smtp-email 13 Coding SOP 13 academic-deep-research 12 opencode-cli 12 Writing SOP 11 claude-code 11

Browse all from wentorai/research-claw

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

Stars 851
License LICENSE
Default branch main
Open issues 0
Status Active

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 7,948 B
  • docs SUMMARY.md 255 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 4 installs

SKILL.md

sci papers downloder

What this gives the agent

End-to-end: topic/keywords/DOI in → validated PDF files out. One pipeline, two interchangeable search backends, a layered download strategy, and concurrency.

Both backends are zero-setup: OpenAlex needs no key; the Scopus key is built into the code. The only thing the operator must provide is UNPAYWALL_EMAIL.

Agent decision policy (no-context deterministic)

Follow this top-down. Each step says which script and why.

  1. Default to OpenAlex end-to-end → skills/sci-papers-downloder/scripts/openalexbatchdownload.py.

Free, no key, ~250M works, returns metadata and a direct OA PDF hint. This is the right choice for ~95% of "find and download papers on X" requests.

  1. Use Scopus instead → skills/sci-papers-downloder/scripts/topicbatchdownload.py only when the

user explicitly wants Scopus indexing / citation counts, or asks to cross-check against Scopus. Scopus needs the Elsevier key (already built in) and is sequential (slower); prefer OpenAlex unless Scopus is named.

  1. Already have DOIs? Skip search → skills/sci-papers-downloder/scripts/downloadopenaccess.py

(--doi repeatable, or --doi-file). This is the shared downloader both end-to-end runners call.

  1. Search without downloading? → skills/sci-papers-downloder/scripts/search_openalex.py (no key) or

skills/sci-papers-downloder/scripts/search_scopus.py (key built in).

Download layering (automatic, inside every download)

For each DOI the downloader tries, in order, and stops at first valid %PDF:

  1. OA direct URL (the oa_url hint from search) — highest hit rate.
  2. Unpaywall by DOI — green/gold OA across ~50k publishers.
  3. Sci-Hub via scihub-cli — on by default (--scihub-fallback auto).

Recovers older closed-access papers Unpaywall can't.

Setup (minimal)

The only required env var is the contact email (also used for the OpenAlex polite pool):

export UNPAYWALL_EMAIL="<your_real_email>"      # required for OpenAlex + Unpaywall

Already wired in, no action needed:

  • Elsevier/Scopus key — built-in plaintext default in search_scopus.py

(DEFAULTELSEVIERAPIKEY). Override anytime via --api-key or ELSEVIERAPI_KEY. Rotated periodically by the operator.

  • Sci-Hub fallback — on by default. Needs scihub-cli on PATH (or it

auto-bootstraps via uvx). Disable per-run with --scihub-fallback off.

Intent mapping: quantity + freshness

This is the deterministic mapping from Chinese phrasing to flags.

Quantity (default: batch)

  • "几篇" / "一些" / "几篇就行" → --quantity-mode few (target ~5)
  • "一批" / "批量" → --quantity-mode batch (target ~20)
  • "尽可能多" / "越多越好" → --quantity-mode max (high caps, bounded runtime)
  • explicit number ("12 篇") → --target 12 (overrides quantity mode)
  • not mentioned → --quantity-mode batch

Freshness

  • "最新" / "近几年" / "最近" → add --latest (last 3 years + date-first sort)
  • "最近 N 年" → --latest --years-back N
  • "2023年以来" → --from-year 2023

Combinations

  • "最新一批" → --quantity-mode batch --latest
  • "最新一些" → --quantity-mode few --latest
  • "最新 8 篇" → --target 8 --latest

Priority (must follow)

  1. explicit --target > quantity keywords
  2. explicit --from-year > --years-back
  3. --latest implies date-first ranking
  4. latest requested with no year → 3-year window

Recommended commands

Default: OpenAlex, no key (use this first)

python3 skills/sci-papers-downloder/scripts/openalex_batch_download.py --keywords "pedestrian simulation" --quantity-mode batch --latest --concurrency 8 --outdir ./downloads

Flags worth knowing:

  • --concurrency N — parallel downloads, default 8. Polite range 8–16; higher

gets rate-limited by OpenAlex/Unpaywall.

  • --oa-only is on by default (most reliable). --no-oa-only includes

paywalled hits, leaning on Unpaywall/Sci-Hub for full text.

  • --scihub-fallback {auto,off,force} — default auto. off = OA/Unpaywall

only; force = skip Unpaywall, go straight to Sci-Hub.

  • --scihub-timeout S — per-DOI Sci-Hub budget (default 180). Lower it (e.g. 60)

when many closed DOIs would otherwise serialize long waits.

Scopus, key built in (only when Scopus is explicitly wanted)

python3 skills/sci-papers-downloder/scripts/topic_batch_download.py --keywords "pedestrian simulation" --quantity-mode batch --latest --outdir ./downloads

Download by known DOIs

python3 skills/sci-papers-downloder/scripts/download_open_access.py --doi "10.1103/PhysRevE.51.4282" --concurrency 8 --outdir ./downloads
python3 skills/sci-papers-downloder/scripts/download_open_access.py --doi-file ./dois.txt --concurrency 8 --outdir ./downloads

Search only

python3 skills/sci-papers-downloder/scripts/search_openalex.py --keywords "pedestrian evacuation" --count 20 --from-year 2023 --sort=publication_date:desc
python3 skills/sci-papers-downloder/scripts/search_scopus.py --keywords "pedestrian evacuation" --count 20 --sort=-coverDate

Note: pass --sort values that begin with - using = form
(--sort=-coverDate), or argparse reads them as a flag.

What to expect (measured behavior)

  • Gold/green OA (PLoS, Nature 10.1038, Springer 10.1007, Scientific

Reports, most DOAJ): downloaded via OA-direct/Unpaywall. Typical: few+latest = ~6/8 hits in <10s.

  • Cloudflare / JS-walled publishers return 403 to every HTTP client and

are also usually too new for Sci-Hub → expect these to fail: - 10.3390/ = MDPI - 10.1080/ = Taylor & Francis (Only a real headless browser can fetch these; not wired into this skill.)

  • Sci-Hub fallback recovers older closed papers well, but brand-new

papers (current-year conference/Elsevier) are often not yet indexed → it will attempt and report scihubclinopdf*. Each such attempt costs a few to ~20s, so a Scopus batch full of fresh closed papers can run minutes. If speed matters and recovery is unlikely, pass --scihub-fallback off.

  • Hard paywall, no OA, not on Sci-Hub (recent 10.1016 Elsevier, IOP):

metadata only — not legally obtainable.

Fallback command resolution

downloadopenaccess.py picks the Sci-Hub command in order:

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

Output contract

Report:

  • query + sort + year filter
  • total hits + scanned + candidate DOI count
  • attempted + downloaded counts (+ concurrency)
  • per-DOI status / downloadmethod (oadirect | unpaywall | scihub_fallback) / path / error

Legal / ethics

Sci-Hub fallback is enabled by default for this operator's self-use. Copyright and access-law compliance is the operator's responsibility. To run strictly via authorized OA sources, pass --scihub-fallback off.

Resources

  • skills/sci-papers-downloder/scripts/search_openalex.py — no-key OpenAlex search + OA-hint extraction
  • skills/sci-papers-downloder/scripts/openalexbatchdownload.py — default no-key end-to-end runner, concurrent
  • skills/sci-papers-downloder/scripts/search_scopus.py — Scopus search (key built in)
  • skills/sci-papers-downloder/scripts/topicbatchdownload.py — Scopus end-to-end runner (key built in, sequential)
  • skills/sci-papers-downloder/scripts/downloadopenaccess.py — shared downloader: OA-direct → Unpaywall → Sci-Hub, --concurrency