handsomestWei/patent-disclosure-skill

What it solves

This project provides an AI-powered toolkit for the entire patent lifecycle, specifically targeting the friction between technical development and legal documentation. It solves the problem of engineers who have the technical solution but struggle to write formal patent disclosure documents, as well as the difficulty of understanding complex, jargon-heavy published patents.

How it works

The system operates as a set of AI skills (AgentSkills) that can be triggered via natural language or specific commands. It leverages LLMs to process technical materials (code, documents, CAD files) and transform them into structured patent outputs.

Key workflows include:

• Disclosure Writing: Scans project files, identifies patentable points, performs prior art searches via CNIPA, and generates disclosure documents in Markdown or Word format, including Mermaid diagrams for inventions.
• Patent Interpretation: Extracts content from PDFs, creates "plain language" summaries, and builds a knowledge graph in Obsidian to link claims, terms, and figures.
• Policy Monitoring: Uses web search to track recent CNIPA policy changes and updates the agent's internal logic to align with current examination guidelines.
• Office Action (OA) Response: Creates a RAG-based knowledge base of historical responses and legal clauses to help draft responses to patent examination notices.

Who it’s for

• Technical Researchers and Engineers: Those who need to turn their code and design documents into formal patent disclosures without being patent experts.
• Patent Attorneys and Agents: Professionals who want to automate the drafting of initial disclosures or the analysis of existing patents.
• Intellectual Property (IP) Managers: Those who need to maintain a private, searchable patent knowledge base.

Highlights

• Multi-type Support: Handles Invention, Utility Model, and Design patents with specific templates for each.
• Visual Integration: Generates line art from product images and extracts projections from CAD models to support structural patents.
• Obsidian Integration: Deeply integrates with Obsidian to create a visual, linked knowledge base of patents using Canvas and relationship graphs.
• RAG-powered OA Responses: Uses a local vector database (optional) to retrieve similar historical cases for drafting legal responses.
• CNIPA Integration: Prioritizes official Chinese patent publication data for prior art searches.