Paper Office is a new suite of Python packages designed to improve AI agent performance when editing Word, PowerPoint, and Excel files by ensuring document integrity and correctness.
Key Points
- Paper Office extends legacy libraries like python-docx, python-pptx, and openpyxl to support complex agent-first workflows.
- The suite achieved a 92.5% success rate across 61 tasks, significantly outperforming standard libraries and Anthropic’s existing office skills.
- New features include cross-run text search, relationship-safe slide cloning, formula-aware spreadsheet edits, and native Word redlining.
- The tools automatically handle internal OOXML bookkeeping, reducing the need for agents to write risky, low-level code to manipulate file internals.
- Paper Office includes validation mechanisms that refuse unsafe operations, preventing silent regressions in document formatting, comments, and formulas.