中国专利.skill: Patent Disclosure Assistant
/patent-disclosure-skill
patent-disclosure-skill is a Python-based AI workflow for Chinese patent drafting and interpretation, automating disclosure document generation with diagram support and handling the full lifecycle from project scan to Word output. It also performs prior art searching via CNIPA's official API and builds a searchable Obsidian knowledge base from patent PDFs. Designed for China's patent system, CAD/STEP parsing is disabled by default and opt-in per session, and vector similarity search requires explicit activation.
The Hurdles of Patent Disclosure and Interpretation
Chinese patent drafting forces engineers to split time between actual R&D and the bureaucratic overhead of disclosure writing. Developers identify the technical innovation, but the patent office expects a structured disclosure document with diagrams, claims, and prior art references in a specific format — skills that never appeared in any engineering curriculum. Meanwhile, reading competitor patents means wrestling with dense PDF text and manually mapping claims to technical context, a process that produces no reusable knowledge.
What is 中国专利.skill?
patent-disclosure-skill is an AI-powered assistant designed for Chinese intellectual property workflows, scanning project materials, extracting patentable points, running prior art searches, and drafting disclosure documents in Chinese for various patent types. It processes documents, code, and optional CAD/STEP files, conducts searches against CNIPA's official database, and drafts invention, utility model, or design patent disclosures, outputting Markdown and Word files. A companion mode converts public patent PDFs into Obsidian notes with claim trees, glossaries, and knowledge graphs.
Key Features for Patent Drafting and Analysis
- ✓Multi-type disclosure drafting: supports invention patents, utility models, and design patents with separate templates and diagram requirements
- ✓CNIPA prior art search: crawls the China National Intellectual Property Administration's official publication database with applicant filtering to reduce false positives from name collisions
- ✓Document-to-disclosure pipeline: scans project files including .docx/.pptx (converted to Markdown), code directories, and optionally STEP/CAD models to surface patentable points
- ✓Diagram generation: creates mermaid block diagrams for invention disclosures, SVG structural schematics with part numbers for utility models, and product line art for design patents
- ✓Dual output format: delivers finished disclosures as both Markdown (for Obsidian ingestion) and Word (.docx) with embedded figures
- ✓Patent interpretation mode: parses public patent PDFs into claim trees, terminology glossaries, and feature-to-specification cross-references, stored as Obsidian notes
- ✓Revision tracking: merges or corrects disclosures into new files with side-by-side change logs and revision conversation records
- ✓Optional policy intelligence: monitors CNIPA guidelines and review trends, outputting a reference list for skill maintenance decisions
Practical Scenarios for Patent Professionals
- •An R&D engineer finishes a module and runs the skill against the project directory to auto-generate an invention disclosure draft, complete with mermaid architecture diagrams and prior art references from CNIPA
- •A design team submits product sketches and the skill produces design patent documentation with auto-extracted product line art and multi-view drawings sorted by a figure plan
- •A patent analyst receives a batch of competitor patents as PDFs; the skill parses each into Obsidian notes with claim trees and an auto-generated canvas graph showing term relationships
- •A patent attorney uploads historical office action responses; the skill indexes them in a local RAG knowledge base with legal-code tags and vector similarity, then uses the index to draft responses to new office actions
Setting Up Your Patent Assistant
Installation requires a Python environment with pip, a compatible AI model interface (OpenAI-compatible API or local model), and optionally an Obsidian vault. Clone the repository, install dependencies listed in INSTALL.md, set required environment variables including API credentials, and configure an Obsidian vault path if using the knowledge base mode. The README points to docs/obsidian-setup-guide.md for Obsidian-specific setup. CAD/STEP parsing is not enabled by default; it requires per-session opt-in configuration.
Interacting with the Patent Skill
Users interact with 中国专利.skill through natural language prompts in Chinese or English. Example triggers: 「按发明写交底，项目路径 …」 for invention disclosure, 「读专利」 followed by a public number or PDF path for patent interpretation, or 「审查答复」 for office action response drafting. Slash commands like /交底书, /读专利, /patent-evolve, and /审查答复 provide direct entry points. The skill expects a project path or technical topic to anchor the context, and users specify patent type explicitly when needed.
Strengths
- ✓Covers the entire disclosure lifecycle: material scanning, prior art search, drafting, diagramming, and multi-version revision with change logs
- ✓Prior art search targets CNIPA's official publication database with applicant-level filtering, reducing false matches from shared names
- ✓Dual output (Markdown plus Word) accommodates both Obsidian-native workflows and traditional document submission requirements
- ✓Patent interpretation mode transforms PDFs into linked Obsidian notes, enabling compounding knowledge from multiple patents over time
- ✓Open source under MIT license, allowing inspection and customization of the disclosure generation logic
Current Scope and Considerations
- △Designed exclusively for China's patent system (CNIPA filings, Chinese-language output, Chinese policy awareness) — not applicable to USPTO, EPO, or other jurisdictions
- △CAD and STEP file parsing is disabled by default; the README states this must be explicitly configured per session, and parsing does not block disclosure drafting when STEP files are absent
- △Vector similarity search for office action responses is optional and requires manual activation; the skill functions with tag-based retrieval alone when vectors are unavailable
- △The Obsidian integration is recommended for full feature access; without a configured vault, the patent interpretation mode falls back to flat file output in outputs/patent_reader/
- △Generated disclosures are drafts requiring human review before submission; the README explicitly states outputs are not submission-ready without legal review
Exploring Other Patent Tools
Frequently Asked Questions
中国专利.skill supports three patent types: invention patents (发明), utility model patents (实用新型), and design patents (外观设计). Each type uses a separate template and diagram approach — mermaid block diagrams for inventions, SVG structural schematics with part numbers for utility models, and product line art for designs.
Yes. The tool is built for the Chinese patent system, targets CNIPA's official database for prior art searches, outputs disclosures in Simplified Chinese by default, and monitors Chinese Patent Office review guidelines. It does not support USPTO, EPO, or other patent office workflows.
Patent interpretation mode parses public patent PDFs and writes notes to an Obsidian vault, including claim trees, terminology glossaries, and Canvas-based knowledge graphs. Set the PATENT_READER_OBSIDIAN_VAULT environment variable to point to your vault. Without this configuration, the tool falls back to flat Markdown output in outputs/patent_reader/.
No. The README explicitly states that generated disclosures are drafts requiring human review before submission. The review-response mode similarly produces drafts that must be reviewed against legal requirements before filing with the patent office.
中国专利.skill is released under the MIT license, as confirmed in the repository metadata. This permits reuse, modification, and commercial use with minimal restrictions (attribution required).
The repository is written primarily in Python, according to the GitHub language statistics. The codebase includes Python scripts for crawling CNIPA data, patent parsing, and disclosure generation, along with markdown prompts and configuration files.
Who should try it — and who should skip
Chinese patent disclosure assistant targets engineers and technical staff at Chinese companies who identify patentable innovations but lack patent writing experience, as well as small legal or IP teams that need to handle disclosure drafting and competitor patent monitoring without dedicated patent attorneys. Researchers building personal knowledge bases from technical literature will benefit from the Obsidian integration. Teams outside China or those filing primarily with non-CIPA patent offices should look elsewhere — this tool assumes a CNIPA workflow throughout.
Related repositories
Still deciding about patent-disclosure-skill?
One click hands the question to an AI along with this page — see what it says about patent-disclosure-skill.