Gdrive Bulk Download

gdrive-bulk-download

Bulk-download Google Docs, Google Slides decks, and existing PDF/DOCX/PPTX files stored in Google Drive, using gstack's headless browser ($B) so it works with restricted/managed accounts that can't install Claude's Drive connector or browser extensions.

Inputs

• A list of URLs, either pasted directly in the prompt (one per line, in

any order, surrounding text ignored) or referenced as a file path (one URL per line; blank lines and #-comments ignored).

• Optional output directory argument. Defaults to ~/Downloads.

Procedure

1. Collect URLs

Parse every https://docs.google.com/... or https://drive.google.com/... URL out of the prompt text, or read them from the referenced file (one per line, skip blank lines and lines starting with #).

2. Verify authentication

B=~/.claude/skills/gstack/browse/dist/browse
$B goto https://drive.google.com
$B text

Check the page text/title for signs of a logged-in session (account name, "My Drive", etc.) rather than a Google sign-in form. If not authenticated:

• Run $B connect then $B goto https://drive.google.com and

$B handoff "log into the restricted Google account" so the user can complete login/MFA manually, then $B resume.

• /setup-browser-cookies is generally unreliable for restricted/managed

accounts — observed failure mode: Google auth cookies (SID, HSID, SAPISID, etc.) get scoped to a country-TLD domain (e.g. .google.ca) that doesn't carry over to drive.google.com/docs.google.com. Prefer /connect-chrome + manual login for these accounts; only suggest cookie import for accounts known to work with it.

Do not proceed to downloads until an authenticated session is confirmed.

3. Classify and resolve each URL

For each input URL, extract the file ID (the segment after /d/) and resolve a download URL:

| Pattern | Resolved type | Download URL | |---|---|---| | docs.google.com/document/d/{ID} | Google Doc | https://docs.google.com/document/d/{ID}/export?format=pdf | | docs.google.com/presentation/d/{ID} | Google Slides | https://docs.google.com/presentation/d/{ID}/export/pdf | | docs.google.com/spreadsheets/d/{ID} | Google Sheet | unsupported — skip, record in manifest, do not download | | drive.google.com/file/d/{ID} (or open?id={ID}) | Drive file (PDF/DOCX/PPTX) | https://drive.google.com/uc?export=download&id={ID} | | Any other direct .pdf/.docx/.pptx URL | Direct file | use as-is |

Anything that doesn't match one of these is also unsupported.

4. Determine a filename

For Docs/Slides/Drive files, navigate to the original (view) URL and read the title:

$B goto ""
$B js "document.title"

Strip trailing suffixes like - Google Docs, - Google Slides, - Google Drive. Sanitize the remaining title for filesystem safety (replace / \ : * ? " | and control characters with -, collapse whitespace, trim). Append the correct extension:

• Docs/Slides → .pdf
• Drive file → keep its real extension (.pdf/.docx/.pptx, inferred

from the original URL/filename or Content-Disposition)

If a title can't be determined, fall back to the file ID as the filename.

5. Create the output directory

mkdir -p "/"

` defaults to ~/Downloads`. Use one timestamp for the whole run — don't create a new folder per file.

6. Download each resolved file

$B download "" "/" --navigate

--navigate handles Drive's redirect/confirmation chains for larger files. If a download fails, record the error and continue with the rest — one failure must not abort the batch.

7. Write the manifest

Write /manifest.json as a JSON array, one entry per input URL:

{
  "source_url": "...",
  "resolved_type": "google_doc | google_slides | drive_file | unsupported",
  "output_filename": "..." ,
  "format": "pdf | docx | pptx | null",
  "status": "ok | unsupported | error",
  "error": "..." 
}

8. Report to the user

Summarize: total processed, counts of ok / unsupported / error, the batch directory path, and the filenames downloaded. For any auth-shaped errors (redirected to a Google login page, 403s), suggest re-running /connect-chrome (or /setup-browser-cookies, for accounts known to work with it) and retrying just the failed URLs.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: MattModeCode
• Source: MattModeCode/claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.