patent-disclosure-skill
ai-tools
Best For
R&D teams filing patents in China who need to produce technical disclosure documents (jiaodishu) quickly. Particularly useful when inventors know what they built but struggle to describe it in patent language. If your patent workflow involves CNIPA filings and you use Claude Code or Cursor, this skill automates the most tedious part of the process.
How I Actually Use It
The skill runs a five-stage pipeline. First, patent point mining: it extracts patentable innovation points from your technical description. Inventors usually describe what they built functionally, and this step translates that into the structured format patent agents need.
Second, CNIPA novelty search: it automatically queries the Chinese National Intellectual Property Administration database for prior art, so you can see what already exists before writing. This step alone saves hours of manual database browsing.
Third, disclosure document generation with sanitization. The output strips sensitive business information, producing a document safe to share with external patent agents. This matters because inventors routinely include proprietary details that shouldn't leave the organization.
Fourth, self-check and multi-round revision. The skill reviews its own output against patent writing standards and iterates.
Finally, it outputs in dual format: .md for the technical team and .docx for the patent agent. This matches how disclosure documents actually flow between teams.
Where It Is Strong
- Full pipeline from invention to document. Most tools handle one step; this chains mining, searching, generating, sanitizing, and revising into a single workflow. The value is in the integration
- Automated CNIPA prior art search. Novelty searching is one of the most time-consuming steps in patent preparation. Having it built into the pipeline removes a manual bottleneck
- Sanitization is built in. The generated document automatically removes commercially sensitive information before it reaches external parties. In enterprise settings, this is a real compliance concern
- Dual format output (.md + .docx) fits the actual handoff between engineering and legal teams
- AgentSkills standard compatibility. Installs as a Claude Code or Cursor skill without additional frameworks
Where It Fails
- China-specific. The entire pipeline is built around CNIPA. Using it for USPTO, EPO, TIPO, or other patent offices requires significant adaptation of the search integration and document structure
- Chinese-language output. Documents are generated in Chinese. International filings or teams working in other languages need additional translation
- Niche audience. The intersection of people who understand patent disclosure workflows, use Claude Code or Cursor, and file patents in China is small
- Novelty search accuracy is unverified. How much relevant prior art the automated search actually catches compared to a professional patent search is unknown
- Not a replacement for patent agents. The output is a disclosure document (input for the agent), not a patent application. Legal review and claim drafting still require a professional
Pricing, Difficulty, and Risk
Free, MIT license. Requires Claude Code or Cursor with the AgentSkills system. Medium difficulty: you need to understand the patent disclosure workflow to use the output effectively, but the skill handles the technical writing. The main risk is over-trusting the novelty search results. Automated prior art searches can miss relevant references that a professional searcher would find through keyword variations, classification codes, or non-patent literature.
Verdict
A well-designed automation for a genuinely painful process. Writing patent disclosure documents is tedious manual work, and this skill compresses the timeline from days to hours. The limitation is clear: it's built for the Chinese patent system. If you file patents through CNIPA and want to accelerate disclosure document preparation, this is worth trying. If your patents go through other offices, the concept is sound but you'll need to wait for broader jurisdiction support or adapt it yourself.