handsomestWei/patent-disclosure-skill

What it does

This repository packages a set of Agent Skills for Chinese patent work. Instead of being a single CLI, it splits into nine sub-skills, each triggered by a phrase in an agent conversation: disclosure drafting for invention, utility model and design patents; turning a finished disclosure into claims, specification, abstract and drawings; a combined "docket" workflow that role-plays a disclosure engineer and a patent attorney across multiple rounds; plain-language reading of published patents (Chinese or foreign) into notes and graphs; patent map building; office-action reply assistance; claim charts exported to Excel; bibliographic and image/claim-based search; and a policy briefing that checks recent CNIPA announcements against the skills' drafting advice README lines 103–191.

The intended audience is engineers who have working code or design documents but no patent-writing experience, and anyone who wants to turn dense patent PDFs into an Obsidian knowledge base. Outputs are file-based: disclosures are written into timestamped files under an outputs directory with Mermaid diagrams, and repeated revisions keep earlier versions plus a revision log README lines 42–58.

Quick start

There is no PyPI package. Installation is done by adding the skills to an agent environment, following the installation guide. After installation, read the skill entry point to see how the agent routes a request to each sub-skill. A typical first run:

1. Have Python 3.9+ and Playwright available, since the skills use Playwright for Mermaid rendering and CNIPA site interaction.
2. Install the skills into your agent per the installation guide.
3. Start a conversation and trigger a sub-skill with its keyword, for example "交底书" for disclosure drafting or "读专利" for patent reading README lines 103–191.
4. Provide the project materials, design document or patent publication number.
5. Collect the generated files from outputs/; the disclosure case produces a timestamped document plus a Mermaid diagram directory.

For patent reading, an Obsidian vault is optional but needed if you want linked notes and maps; a Windows setup guide for Obsidian and its community plugins is provided Obsidian setup guide.

Constraints

• The project targets Chinese patent practice and CNIPA sources, so filing-style conventions and policy briefings apply to CN filings.
• Python 3.9+ and Playwright are required for the parts that render diagrams or access the CNIPA site.
• Patent maps and reading workflows depend on Obsidian; linking many patents together is an Obsidian feature, not something the skill does on its own.
• The claim chart output is an Excel file and explicitly is not legal advice; patent drafting and office-action replies still need review by a qualified practitioner.
• Published under the MIT license.

HotRepo's take

On the Python daily list for 2026-11-04, the repository reached position 13 with 72 stars added that day and totals of 12,884 stars and 1,391 forks. Earlier daily records show the growth was steadier than a one-day spike: 571 stars on 2026-08-31, about 500 per day through 2026-09-02, and a much smaller 38 the day before. That pattern is consistent with a sustained run of attention rather than a single burst, though the gaps in the record mean the full trajectory is unknown. The contribution itself is domain packaging: patent practice is a field where agents often give generic answers, and the sub-skill split plus the emphasis on traceable revisions addresses a concrete gap.