name	pdf-downloader
description	Download articles, papers, docs, or webpages as local PDF files. Use when the user asks to download a PDF, save an article/page/link as PDF, fetch a paper by title or URL, or preserve an online page offline. Prefer direct PDF download when a URL resolves to a PDF; otherwise generate an offline PDF from the webpage content and include the source URL.

PDF Downloader

Workflow

Identify the best source URL.
- If the user gives a URL, use it directly.
- If the user gives only a title, search the web and prefer official or primary sources.
- If there is a direct .pdf link or Content-Type: application/pdf, download it directly.
Save final files under output/pdf/ unless the user gives another destination.
Run the bundled script:

python3 ~/.codex/skills/pdf-downloader/scripts/download_pdf.py "$URL" --output-dir output/pdf

Validate the result:
- Check file size and page count with pypdf.
- Render a thumbnail with qlmanage on macOS when available, or open/inspect another quick preview if available.
- Confirm the title/source URL appears in generated offline PDFs.
- For generated webpage PDFs, scan the first page visually when possible. Latin-heavy pages should use normal Latin word wrapping, not CJK character wrapping; CJK-heavy pages should still render with CJK-capable fonts.
Reply with absolute clickable file paths and note whether the file was directly downloaded or generated from the webpage.

Blocked Pages

Some sites block direct Python requests but are still readable through browser/search tooling. If the script fails with 403, challenge pages, empty content, or missing article text:

Use browsing/search to retrieve the article text from the official page.
Save a temporary Markdown file with title, source URL, date/author if known, headings, paragraphs, bullets, code blocks, tables, and image captions.
Convert it with:

python3 ~/.codex/skills/pdf-downloader/scripts/download_pdf.py --markdown tmp/article.md --source-url "$URL" --output-dir output/pdf --filename article-name.pdf

This fallback is an offline webpage capture, not an official PDF. Say that clearly.

Script Notes

scripts/download_pdf.py writes a .pdf; for generated webpages it also writes a sibling .md.
Use --filename name.pdf for stable names.
Use --title "Title" when the source page lacks a clean title.
The script chooses reader-friendly fonts automatically: English/WinAnsi pages prefer built-in Helvetica with normal LTR wrapping; mixed Unicode pages use local TTF fonts when needed; CJK-heavy pages prefer CJK-capable fonts and only then use CJK wrapping. If no local TTF is available, it falls back to built-in PDF fonts or STSong-Light for CJK.
If a generated PDF looks cramped, regenerate it with the current script before trying manual fixes; older outputs may have used CJK wrapping for English pages.
If the source page contains dynamic images or video, preserve captions/source links rather than trying to embed media unless the user asks for a visual archive.

PDF Downloader

Workflow

Identify the best source URL.

If the user gives a URL, use it directly.
If the user gives only a title, search the web and prefer official or primary sources.
If there is a direct .pdf link or Content-Type: application/pdf, download it directly.

Save final files under output/pdf/ unless the user gives another destination.

Run the bundled script:

python3 ~/.codex/skills/pdf-downloader/scripts/download_pdf.py "$URL" --output-dir output/pdf

Validate the result:

Check file size and page count with pypdf.
Render a thumbnail with qlmanage on macOS when available, or open/inspect another quick preview if available.
Confirm the title/source URL appears in generated offline PDFs.
For generated webpage PDFs, scan the first page visually when possible. Latin-heavy pages should use normal Latin word wrapping, not CJK character wrapping; CJK-heavy pages should still render with CJK-capable fonts.

Reply with absolute clickable file paths and note whether the file was directly downloaded or generated from the webpage.

Blocked Pages

Some sites block direct Python requests but are still readable through browser/search tooling. If the script fails with 403, challenge pages, empty content, or missing article text:

Use browsing/search to retrieve the article text from the official page.

Save a temporary Markdown file with title, source URL, date/author if known, headings, paragraphs, bullets, code blocks, tables, and image captions.

Convert it with:

python3 ~/.codex/skills/pdf-downloader/scripts/download_pdf.py --markdown tmp/article.md --source-url "$URL" --output-dir output/pdf --filename article-name.pdf

This fallback is an offline webpage capture, not an official PDF. Say that clearly.

Script Notes

scripts/download_pdf.py writes a .pdf; for generated webpages it also writes a sibling .md.

Use --filename name.pdf for stable names.

Use --title "Title" when the source page lacks a clean title.

The script chooses reader-friendly fonts automatically: English/WinAnsi pages prefer built-in Helvetica with normal LTR wrapping; mixed Unicode pages use local TTF fonts when needed; CJK-heavy pages prefer CJK-capable fonts and only then use CJK wrapping. If no local TTF is available, it falls back to built-in PDF fonts or STSong-Light for CJK.

If a generated PDF looks cramped, regenerate it with the current script before trying manual fixes; older outputs may have used CJK wrapping for English pages.

If the source page contains dynamic images or video, preserve captions/source links rather than trying to embed media unless the user asks for a visual archive.