AI BriefingPaper Instruments Releases Paper Office Python Suite to Boost AI Agent Document Editing Accuracy
Key point
The new suite achieves a 92.5% task success rate across five LLMs, significantly outperforming standard libraries and Anthropic's office skills.
Details
Paper Instruments has released Paper Office, a Python package suite designed to enable AI agents to accurately manipulate Word, PowerPoint, and Excel files. The suite is a fork and patch of existing open-source libraries (python-docx, python-pptx, OpenPyxl) with a focus on agent-first use cases, addressing the lack of correctness contracts and feature breadth in legacy tools.
Performance Benchmarks
In evaluations across 61 tasks and 5 models (Opus 5, GLM 5.3 Flash, GPT 5.6 Sol, Grok 4.6, DeepSeek v4.1 Flash), Paper Office demonstrated superior reliability:
- Paper + Guidance: 92.5% task success rate.
- Upstream Packages (No Skills): 80.7%.
- Upstream + Anthropic Skills: 69.5%.
The suite reduced the need for agents to write raw code for internal file bookkeeping. The rate of agents writing direct ZIP manipulation code dropped to 1.6% with Paper, compared to 78.7% for upstream packages and 50.5% for Anthropic skills.
Key Technical Features
The suite maintains backward compatibility with standard imports while exposing hidden OOXML structures as typed, machine-readable data. Key capabilities include:
- paper-docx: Cross-run text search, native Word redlines (tracked changes), and package-preserving saves that restore original bytes for unchanged parts.
- paper-pptx: Relationship-safe slide cloning, effective formatting inspection through master/theme chains, and policy-based slide imports.
- paper-xlsx: Reference-aware structural edits that update dependent formulas and ranges, formula cache invalidation, and LibreOffice-based recalculation certification.
Impact on Agent Workflows
The primary advantage is preventing silent regressions. Legacy libraries often force agents to manually manage complex relationships (e.g., comment anchors, chart workbooks, field dependencies), leading to broken documents even when the primary text edit is correct. Paper Office handles these linked changes via API, ensuring document integrity during automated editing.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Jevper Released: A Jev Classification Wrapper for OpenAI-Compatible Models with Automatic Logprobs Detection