Rebuild scanned pages as vector PDFs
Web page · https://agentusecases.ai/use-case/rebuild-scanned-pages-as-vector-pdfs
Done with Claude Code, Codex, Claude. A Claude Code and Codex skill redraws scans and phone photos as new PDFs with real text, vector diagrams and original layout, benchmarked with and without the skill.
First seen by Oriel on .
AI summary
A Claude Code and Codex skill reconstructs scans and phone photos as layout-matched vector PDFs, achieving benchmark scores up to 0.998 while requiring minutes of agent time per page.
- The workflow measures scan pixels, drafts every page in one command, and lets the agent inspect uncertain words through enlarged per-page contact sheets.
- Each rebuild is overlaid on its source, with a report identifying omissions; individual word corrections require one JSON line.
- On unseen nine-page synthetic documents, Haiku and Fable improved from about 0.84 to 0.997, while Sonnet and Opus rose from roughly 0.994 to 0.998.
- Improvements were within noise for Sonnet and Opus on three-page packets, and Opus and Fable ran slower when using the skill.
- For searchable text without reconstruction, the page recommends OCRmyPDF instead.
#claude code #codex #pdf #document processing
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| Agent Use Cases | https://agentusecases.ai/use-case/rebuild-scanned-pages-as-vector-pdfs | https://agentusecases.ai/use-case/rebuild-scanned-pages-as-vector-pdfs Read discussion |