7toCR/paper2patent

Paper2Patent – AI‑assisted tool that turns a research paper into a Chinese invention‑patent draft

What it does | The repository provides a complete workflow for converting a scientific paper (especially in AI/ML fields) into the five mandatory sections of a Chinese invention‑patent application. It supplies:

• Prompt templates for large‑language‑model (LLM) chats that output the five textual sections (abstract, claims, specification, etc.).
• A “Skill” package that can be loaded into AI assistants such as Codex, Claude, Claude Code, Cursor or Windsurf, allowing the whole pipeline—reading the paper, assessing novelty, drafting claims, generating black‑and‑white drawings, running automatic checks, and exporting Word/PDF—to be run with a single natural‑language command.
• Python scripts that perform automatic consistency checks, generate SVG/PNG line‑drawings for the patent figures, and assemble the final DOCX/PDF files.
• Reference documents that encode Chinese patent‑law requirements, claim‑drafting rules, drawing standards and a quality‑checklist.

Why it matters | Academic papers disclose inventions, but Chinese patent practice requires a very different structure and strict legal language. Researchers often lack the expertise to rewrite a paper into a compliant patent draft. Paper2Patent bundles the legal knowledge (via the reference markdown files) with LLM prompting and automation, letting scientists produce a draft that a professional patent attorney can review, dramatically lowering the barrier to technology transfer.

Key components

• skills/paper2patent/ – the Skill definition (rules, prompts, and workflow) for supported AI assistants.
• scripts/ – Python utilities:
  • check_patent_draft.py – validates formatting, forbidden words, claim‑reference consistency, abstract length, etc.
  • generate_patent_drawings.py – creates black‑and‑white SVG/PNG flow‑charts or block diagrams from the drafted text.
  • generate_patent_docx.py – builds the DOCX containing the five sections and a separate “writing‑instructions” document.
  • export_patent_pdf.py – converts the DOCX to PDF via LibreOffice.
• assets/example_patent_content.json – a fully‑structured example of the intermediate data model.
• evals/ – scripts for automated quality evaluation of generated drafts.

Getting started

1. Clone the repo.
2. Copy the Skill folder into the appropriate directory of your AI assistant (e.g., ~/.codex/skills/ for Codex, ~/.claude/skills/ for Claude, or upload the zipped folder in Claude’s UI).
3. Ensure Python 3.9+ is installed, along with Pillow and a CJK font for drawing generation.
4. In the AI assistant, issue a natural‑language command such as:
   • “使用 paper2patent skill，将这篇论文生成中国发明专利申请书 Word 和 PDF。”
   • “先帮我梳理发明点和权利要求布局，确认后再写。”
5. The assistant will output:
   • <title>_专利申请文件.docx (or .pdf) containing the five sections.
   • <title>_专利申请文件_撰写说明.docx with a checklist of missing information and risk notes.
   • SVG/PNG files for each required drawing.

Supported environments | The Skill works on macOS and Linux (direct copy) and Windows (PowerShell copy). PDF export requires LibreOffice; drawing generation needs Pillow and a Chinese font (system default or Noto CJK).

Legal & community | MIT‑licensed, with a custom CLAUSE.md that forbids reposting the project as one’s own. The authors ask that users cite the original repository when referencing the tool.

---

In a nutshell: Paper2Patent is an open‑source, AI‑driven pipeline that helps researchers automatically draft Chinese invention‑patent applications from academic papers, handling claim structuring, drawing creation, and compliance checking—all via simple natural‑language prompts in supported AI assistants.