SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsacademic-paper-download

academic-paper-download

Writing & research

View repository

Updated 10-16 05:42

About this skill

Academic Paper Download

Produce a structurally verified PDF and adjacent provenance manifest. Require an explicit final output directory and keep automatic resolution in this order: Unpaywall, Semantic Scholar OA, arXiv, then browser handoff.

Workflow

  1. For a DOI or exact title, select the caller's final output directory. Never guess a research directory. Add author/year when they help disambiguate a title.
  2. For a publisher URL, first resolve or confirm its DOI. Pass only that DOI to fetch.py; the CLI does not accept publisher URLs as inputs.
  3. Run the downloader and accept success only when the result contains a verified file, SHA-256, size, and adjacent manifest.
  4. If automatic OA sources are exhausted, or a publisher page requires login, institution access, or interaction, explicitly choose a browser backend and read references/browser-handoff.md. Prefer the current Chrome session when it already has authorized institutional access. Never silently switch backends after login failure.
  5. When the user explicitly selects CloakBrowser, also read references/cloakbrowser-handoff.md. Keep it outside PaperTransport; do not use its stealth or humanize features to solve CAPTCHA, Turnstile, paywalls, security warnings, or authentication.
  6. Read references/integration.md when embedding the library or injecting a provenance-recording transport. Read references/env.md for runtime configuration.

CLI

Resolve the installed skill directory to an absolute path. Use the path exposed by the skill loader or npx skills list --json; do not assume the current working directory is the skill directory:

SKILL_DIR='/absolute/path/to/academic-paper-download'

After installation, explicitly create the hash-locked CLI runtime and prove it with the network-free smoke test. bootstrap is the only core command that installs packages; normal commands never install or update dependencies:

python3 "$SKILL_DIR/scripts/runtime.py" bootstrap --locked --json
python3 "$SKILL_DIR/scripts/runtime.py" smoke --offline --json

The runtime lives outside the installed skill directory and works for copy, symlink, and read-only installations. Install requirements-cloakbrowser.txt only in a separate isolated environment when that optional backend is explicitly selected. Its browser binary is a separate, preflight-verified installation; the handoff script never downloads it.

Fetch by DOI:

python3 "$SKILL_DIR/scripts/runtime.py" fetch \
  '10.48550/arXiv.1706.03762' \
  --out ./papers --format json --pretty

Fetch by exact title:

python3 "$SKILL_DIR/scripts/runtime.py" fetch \
  --title 'A precise paper title' \
  --author 'First Author' --year 2024 \
  --out ./papers --format json --pretty

Use fetch schema to inspect the unchanged machine contract version. A verified existing artifact may return skipped: true. Read references/env.md when embedding the library or diagnosing Python/runtime compatibility.

Result Rules

  • Treat exit code 0 as complete, 1 as unresolved, 3 as invalid input, and 4 as retryable transport failure.
  • Require pypdf parsing, at least one page, a final %%EOF, matching size, and SHA-256 before committing the PDF and manifest.
  • Never infer redistribution permission from successful access. Preserve access_basis, license_status, and source-declared license fields.
  • Never select the newest file in Downloads or accept HTML, truncated PDFs, symbolic links, partial downloads, credentials, cookies, passwords, or session tokens.

Provenance

scripts/paper_fetch/ selectively adapts MIT-licensed ideas and code from Agents365-ai/paper-fetch at commit c3baaa3d5df9a7eecb16fc2b4c8d10416f59bcb7. See LICENSE.paper-fetch.txt for the retained license notice.