Back to browse
GitHub Repository

Trust, but check. A Claude skill that turns AI answers into visual proof — highlighted screenshots and confidence scores, straight from the source PDF.

74 starsPython

PDF-Proof – Make Claude show where it found each value in a PDF

by young_mete·Apr 12, 2026·4 points·0 comments

AI Analysis

●●●BangerSolve My ProblemBig BrainSlick

Verifies AI extractions by reading text back from highlighted regions — actual audit trails for PDFs.

Strengths
  • PyMuPDF coordinate extraction plus verification read-back catches false positives automatically.
  • Standalone HTML proof pages embed screenshots — shareable without the original PDF.
  • OCR support via Tesseract handles scanned documents, not just text-based PDFs.
Weaknesses
  • Requires Claude Desktop with code execution enabled — limits deployment scenarios.
  • No batch processing mode for verifying dozens of values across multiple documents.
Category
Target Audience

Accountants, legal professionals, data analysts working with documents

Similar To

JinaAI Reader · Firecrawl · Nanonets

Similar Projects