Summary
This is a solid PDF workhorse that covers the full lifecycle: extract text and tables with pdfplumber, merge and split with pypdf, create new documents with reportlab, and handle forms. It includes both Python APIs and command-line tools like qpdf and pdftotext. The table extraction is more robust than basic pypdf, the form filling guide covers both fillable and non-fillable PDFs, and there's OCR support for scanned documents. If you're doing any serious PDF automation beyond just reading them, this gives you the right tool for each job instead of forcing everything through one library.
Install to Claude Code
npx -y skills add appautomaton/document-skills --skill pdf --agent claude-codeInstalls into .claude/skills of the current project.
Keeps your Mac awake while Claude Code, Codex or Cursor works. Lets it sleep when they are done.
Try free for 7 days →
Block distracting apps from your iPhone permanently without a 3rd party app. Free and open source.
Block now (100% free) →
Notes, actions and memory. Without a meeting bot. First month 100% off.
Download for free →
belt cli automatically finds the best tools and skills for your agent. image, video, music, tts...
one prompt install →
Free an open-source skills that make AI coding agents easier to understand, verify, and control.
Download for free →
Connect your Claude agent to live crypto prices and trading routes via 1inch
Get the MCP →Files
SKILL.mdView on GitHub
PDF files
You already know pypdf, pdfplumber and reportlab. This skill adds what they don't give you: form-filling scripts that check their own input, a way to see pages as a reader will, and the PDF behaviours that silently put content in the wrong place.
Running scripts
<skill> is the directory containing this file. Run bundled scripts by absolute path from the user's working directory, so relative paths to their files keep working:
uv run <skill>/scripts/convert_pdf_to_images.py report.pdf pages
Run Python you write with its dependencies declared, either uv run --with pypdf script.py or a PEP 723 header. Rendering pages needs poppler (pdftoppm); OCR also needs Tesseract.
Choosing a tool
| Task | Use |
|---|---|
| Fill a form, fillable or flat | forms.md, always. It has scripts for every step. |
| Text | pdfplumber, or pdftotext -layout from the command line. Text missing from pages that visibly have it means a scan: see ocr.md. |
| Tables | pdfplumber: see tables.md |
| New PDFs | reportlab: Platypus for flowing documents, canvas for exact placement |
| Merge, split, rotate, crop, stamp, encrypt | pypdf: see reference.md for the calls that keep bookmarks, links and orientation intact |
| See what a page looks like | scripts/convert_pdf_to_images.py |
Checking the output
Whenever you create or change a PDF, render the pages you touched and look at them before delivering. Check that the text is present and readable, nothing is clipped or overlapping, and every stamp or value is where it belongs:
uv run <skill>/scripts/convert_pdf_to_images.py out.pdf check --pages 1-3
Pitfalls that fail silently
- Two coordinate systems. PDF coordinates start at the bottom-left of the page box, in points. pdfplumber and page images start at the top-left. The page box need not start at (0, 0), so read
page.cropbox.leftand.bottom. - Rotated pages.
/Rotateturns a page only for display; content and form fields keep unrotated coordinates. Anything placed with a plain scale and y-flip lands in the wrong spot. Convert boxes withscripts/pdf_coords.py, and callpage.transfer_rotation_to_content()before stamping. - Non-Latin text in new PDFs. reportlab's built-in fonts draw Chinese, Cyrillic and similar scripts as black boxes. Register and embed a TrueType font; see reference.md.
- Merging.
PdfWriter.append()keeps bookmarks and links, while copying pages withadd_page()drops them. Merged forms with matching field names become one field, so filling one fills both. - AES encryption needs
uv run --with 'pypdf[crypto]'. Without it, pypdf raisesDependencyErroron open or on encrypt.
Scripts
| Script | Does |
|---|---|
check_fillable_fields.py <pdf> | Says which form path in forms.md applies |
extract_form_field_info.py <pdf> <out.json> | Lists fillable fields, their pages, boxes and accepted values |
fill_fillable_fields.py <pdf> <values.json> <out.pdf> | Fills fields after checking every id, page and value |
convert_pdf_to_images.py <pdf> <dir> [--pages] [--fields] | Renders pages one at a time to PNG, optionally outlining each fillable field |
check_bounding_boxes.py <fields.json> [pdf] | Finds overlapping or too-short boxes before a flat form is filled |
create_validation_image.py <page> <fields.json> <png> <out.png> | Draws those boxes on a page image for checking by eye |
fill_pdf_form_with_annotations.py <pdf> <fields.json> <out.pdf> | Fills a flat form with text that renders identically everywhere |
flatten_form.py <pdf> <out.pdf> | Draws filled values into the page so they can't be edited |
pdf_coords.py <pdf> <page> <WxH> --to-pdf/--to-image <box> | Converts boxes between page-image pixels and PDF points |
Featured
Keeps your Mac awake while Claude Code, Codex or Cursor works. Lets it sleep when they are done.
Try free for 7 days →Block distracting apps from your iPhone permanently without a 3rd party app. Free and open source.
Block now (100% free) →Notes, actions and memory. Without a meeting bot. First month 100% off.
Download for free →belt cli automatically finds the best tools and skills for your agent. image, video, music, tts...
one prompt install →Free an open-source skills that make AI coding agents easier to understand, verify, and control.
Download for free →Connect your Claude agent to live crypto prices and trading routes via 1inch
Get the MCP →
