Scanned PDF Text Repair

Use this skill for image-only PDFs where text extraction returns empty or does not contain the target text. The goal is visual repair: rebuild the page image with a small localized patch, not semantic PDF text editing.

Workflow

1. Inspect the PDF before editing.

   • Count pages with PyMuPDF.
   • Extract text with page.get_text().
   • If the target text is absent and the page has one or more images, treat it as scanned/image-only.
   • Render the page at 2x or 3x scale and inspect the target area.
   • Prefer scripts/inspect_pdf.py for this step.

2. Locate the exact patch.

   • Use rendered pixel coordinates, not guessed PDF point coordinates.
   • Crop the target line and identify dark-pixel runs for each character.
   • Patch only the old character/phrase plus a tiny margin.
   • Do not repaint surrounding text unless the user explicitly wants phrase-level replacement.
   • Use scripts/find_dark_runs.py on the rendered page or crop when exact character bounds are unclear.