---

name: pdf description: Read, create, combine, annotate, or revise PDF documents using local Python libraries and inspect rendered pages. Use for PDF reports, forms, page operations, and extracted text; route source Word or Excel edits to their corresponding skills. license: MIT metadata: owner: svd-ai-lab/sim-skills source: https://github.com/svd-ai-lab/sim-skills/tree/main/pdf

PDF documents

Resolve input and output paths explicitly. Keep the original file and save edits to a new path unless the user asked for replacement. Prefer an installed local tool; otherwise use a project environment, for example uv run --with pypdf --with reportlab --with pypdfium2 --with pillow script.py. No hosted service or application account is required.

Read and inspect

Use pypdf.PdfReader(path) to inspect metadata, page count, outlines, forms and page.extract_text(). Keep page numbers with extracted evidence. Text extraction does not establish layout or reading order. Render tables, plots, equations and columns before interpreting them. For image-only scans use an available OCR engine, retain the original page, and flag uncertain characters. Do not treat an empty extracted string as an empty page. Encrypted documents require the user's authorized password; never print it.

Run the bundled renderer using its path relative to this skill directory:

uv run --with pypdfium2 --with pillow scripts/render_pdf.py input.pdf output-pages

It writes numbered PNGs without modifying the PDF. Open the relevant PNGs with the available image viewer. PDFium is a renderer, not an OCR engine.

Create and edit

For a report, use ReportLab SimpleDocTemplate, Paragraph, Table, and PageBreak with explicit page size and margins. Use paragraph styles for headings, wrapping, spacing and captions. Register an appropriate licensed font for the actual text, especially non-Latin characters. Escape user text when feeding ReportLab's paragraph markup.

Use PdfWriter to append, reorder, rotate or crop existing pages and to update supported fields. For a visible annotation or watermark, create a correctly sized overlay and merge it onto the target page. Ordinary overlays do not erase underlying text. For removal of confidential content, use a tool with actual redaction support and verify the removed content is absent from extraction and embedded objects. For substantial prose edits, modify the source document when available and export again instead of painting over the old text.

Verify delivery

Reopen the saved PDF, check page count and required text, then render every changed page. Inspect clipping, glyphs, margins, table splits and captions at readable resolution. Verify forms after saving. Report any rendering or OCR limitations honestly and link the final PDF using an absolute path.

API references: pypdf, ReportLab, pypdfium2.