patent-disclosure-skill
这是一款面向专利技术交底书撰写的专业 Skill，可通过多轮自适应对话引导发明人从零梳理技术方案，自动生成完整交底书，并将口头描述、文档或代码转化为专业专利语言。同时支持黑白规范附图自动生成和 A4 排版 Word 文档一键导出，显著提升专利交底材料的撰写效率与规范性。
A plug-in skill for AI coding assistants like Claude Code and Cursor that guides inventors through a multi-round conversation to produce a complete, formally structured patent disclosure document with SVG diagrams.
This is a skill (a plug-in set of instructions) for AI coding tools such as Claude Code, Cursor, Windsurf, and similar assistants. Its purpose is to help inventors write patent disclosure documents, which are the detailed technical write-ups that patent attorneys and agents need before they can file a patent application. The skill guides the user through this process using a multi-round conversation rather than asking them to fill out a form.
When you activate it, the skill starts by asking a few questions to understand who you are (inventor or patent agent), what you have (a rough verbal description, an existing document, or a code repository), and which technical domain your invention belongs to. From there it asks one focused question at a time to extract the necessary information. The whole conversation typically takes between five and twenty rounds depending on how much material you already have ready.
Once it has gathered enough detail, the skill generates a complete patent disclosure document structured the way the China National Intellectual Property Administration expects: technical field, background, description of the invention, technical problem, technical solution, beneficial effects, description of figures, and specific implementation methods. It writes in formal patent language rather than casual AI-generated phrasing, filtering out generic filler phrases automatically.
The skill also produces supporting diagrams in SVG format, following the strict rules that patent offices apply to figures: pure black-and-white lines, three-digit reference numbers, no overlapping lead lines. Supported diagram types include system architecture diagrams, flowcharts, and data-flow diagrams. Once the document and diagrams are ready, you can export the whole package to a properly formatted Word file with a single command.
The project includes reference files for three technical domains (computer software, electronics and communications, and mechanical manufacturing), each containing a terminology guide and standard sentence patterns used in that field. You can add new domains by creating additional reference files in the same format. The repository is written in Python and is licensed under MIT.
Where it fits
- Walk through a guided conversation in your AI coding tool to generate a complete patent disclosure document ready for your patent attorney.
- Automatically generate black-and-white SVG patent figures such as system diagrams and flowcharts that meet patent office formatting rules.
- Export your completed patent disclosure and diagrams to a formatted Word file with a single command.