PDF

Process PDFs by identifying the operation, selecting the right tool, implementing, and verifying output.

When NOT to use

• The task is pure image processing (no PDF involved) — use image-specific tools instead.
• The PDF is a scanned image where no text extraction is needed — the skill handles OCR internally, but if you only need image manipulation of extracted pages, route differently.

1. Identify the operation

Determine what the user wants to do:

• Read/extract: text, tables, metadata, images
• Transform: merge, split, rotate, watermark, encrypt/decrypt
• Create: new PDFs from scratch or from other formats
• OCR: scanned PDFs that need text recognition

2. Select library or tool

Choose based on the operation:

Check that the library is installed; install if missing. For the forms workflow (detect fillable fields, extract field info, fill, annotate), use pypdf field APIs directly (AcroForm / update_page_form_field_values). If a bundled forms.md reference is added to this skill in the future, load it for scripts/ helpers. Advanced / second-tier libraries (pypdfium2, etc.) are documented in references/ when present.

Common code snippet (use directly for routine work)

Read text with pdfplumber:

import pdfplumber
with pdfplumber.open("input.pdf") as pdf:
    for page in pdf.pages:
        print(page.extract_text())

All other snippets (merge, split, rotate, watermark, encrypt, create, OCR, CLI tools) live in references/code-snippets.md.

3. Implement the operation

Write a script that:

• Reads input PDF(s) with the selected library
• Performs the operation
• Writes output to the specified location
• Handles errors (missing files, corrupt PDFs, permission issues)

4. Verify output

• Confirm the output file exists and is non-empty
• Spot-check content (e.g., page count, extracted text sample, table row count)
• Report results to the user

Completion criteria

• Operation type and input/output files identified
• Tool selected and available in the environment
• Script runs without errors
• Output file exists and is non-empty
• Content spot-checked (page count, text sample, table row count)
• User informed of results

Related skills

• xlsx — sibling file-doc skill (share zero-error bar).
• data-access — read .pdf table via DuckDB fallback when needed.