Show HN: PDF-Proof – Make Claude show where it found each value in a PDF (github.com)

A new skill called "pdf-proof" has been introduced for Claude that significantly enhances trust in AI outputs by providing visual verification of data taken from PDFs. This skill allows users to request that Claude not only finds specific values in documents—including tax returns, contracts, and invoices—but also highlights them within the original PDF, confirms their accuracy, and compiles all findings into a shareable HTML proof page. The process involves extracting text coordinates using PyMuPDF, cropping relevant sections, and employing back-read verification to ensure data integrity. The implementation of pdf-proof is particularly significant for the AI/ML community as it addresses a critical concern regarding the reliability of AI-generated results, especially in high-stakes areas like finance and legal documentation. By delivering a visual audit trail alongside extracted data, the skill fosters increased accountability and transparency. Additionally, its integration process is user-friendly, requiring minimal setup while automating dependencies and OCR support for scanned documents. This innovation demonstrates a valuable convergence of AI capabilities with practical needs, setting a new standard for verifiability in AI applications.