If you want to apply for a technology patent but find it troublesome to draw system block diagrams and flowcharts for the disclosure document, and have to edit Word documents, here’s a solution. I found this “中国专利.skill” (Chinese Patent.skill) on GitHub that automates the entire pipeline from project documents to a finalized patent disclosure. It automatically scans project documents and code to mine patent points, and can even connect to the CNIPA publication site to perform novelty searches, avoiding conflicts with existing patents. GitHub: http://github.com/handsomestWei/patent-disclosure-skill… The generated disclosure includes system block diagrams and flowcharts. After desensitization, it directly produces a Word document, making it easy to hand off to an agent for revision. If you need to supplement materials or correct errors, you don’t have to rewrite everything—you can iterate on the original draft. Ideal for developers who have technical solutions but don’t want to spend excessive time writing disclosure documents.

handsomestWei/patent-disclosure-skill

Source: https://github.com/handsomestWei/patent-disclosure-skill

中国专利.skill

From project documents to deliverable technical disclosures: patent point mining, novelty search prioritized on CNIPA publication site, desensitized drafting, and self-check loop.

License: MIT

Python 3.9+ (https://www.python.org/)
Node.js (https://nodejs.org/)
AgentSkills (https://agentskills.io)

Have design documents and code but haven’t sorted out patent points yet?
Need system block diagrams, flowcharts, and a Word document that agents can directly edit?
After finalization, still need multiple rounds of supplementary materials and corrections, and want revision traceability?
When searching the CNIPA publication site, want reliable crawling success and precise search?

This Skill follows AgentSkills conventions to orchestrate the entire workflow; SKILL.md + prompts/ are step-by-step, readable, and iterable.

Features · Installation · Usage · Project Structure · Examples · Screenshots · References · Detailed Installation Guide · Entry Point

Features

Capability Description

• Project Scan: Reads documents/code by priority; converts .docx/.pptx to Markdown first, then scans (see prompts/project_scan.md)
• Patent Points: Discussion and fusion of candidate patent points (patent_points_analyzer.md)
• Novelty Search prioritized on CNIPA · Chinese Patent Publication Website (tools/cnipa_epub_search.py); falls back to WebSearch (Google Scholar / Patents) on error or no results. Bibliographic data and external links are written into Chapter 1 (prior_art_search.md)
• Disclosure Drafting: Desensitization template + Mermaid system block diagrams and flowcharts; mermaid_render.py → PNG, default output to .docx. Delivery files named as: {case_name}_{YYYYMMDDHHmmss}.md with a .docx counterpart (disclosure_builder.md §7.3)
• Self-check: Logic closure, formula and parameter consistency check (disclosure_self_check.md, not written into the main body)
• Iteration Merge / Correction: Saves new files; “Disclosure Revision Dialogue Log.md” appended incrementally (iteration_context.md, iteration_dialog_log.py)
• Office Extraction: .docx / .pptx are first converted to Markdown using docx_to_md.py / pptx_to_md.py from this repo before scanning (see SKILL.md).

Python Dependencies (Separate Files):

• Base (Office / disclosure conversion): Root directory requirements.txt — pip install -r requirements.txt
• Novelty Search (CNIPA publication site, optional): tools/requirements-cnipa.txt — pip install -r tools/requirements-cnipa.txt, then run python -m playwright install chromium.
  If not installed: Step 5 will only use WebSearch as a fallback according to prior_art_search.md. See INSTALL.md, tools/README.md.

Installation

Claude Code

Place this directory at the root of your git repository or in a global skills path, so that SKILL.md is at the root level of the skill folder (consistent with INSTALL.md).

# Example: Install into the current project's skills directory
mkdir -p .claude/skills
git clone <this repo URL> .claude/skills/patent-disclosure-skill

Cursor

Copy the entire contents of this repository to the skills path designated by Cursor (see INSTALL.md for the table). Restart and confirm in Settings → Rules that the skill has been detected.

Dependencies

# Base (Office conversion, disclosure-related Python packages)
pip install -r requirements.txt

# Optional: CNIPA novelty search (epub.cnipa.gov.cn)
pip install -r tools/requirements-cnipa.txt
python -m playwright install chromium

Diagram rendering also requires Node.js; run npm install in tools/ or use npx mmdc (see tools/README.md for details).

Usage

Simply describe your needs in natural language to the Agent. For example:

• Patent mining, patent points, technical disclosure, novelty search, prior art comparison
• Slash commands (depending on the host configuration): e.g., /patent-disclosure-skill, /交底书

It is recommended to also specify the project path or technical subject (consistent with argument-hint in SKILL.md).

Novelty Search (Step 5) will first attempt to retrieve Chinese patent publication information from the China Patent Publication Website (http://epub.cnipa.gov.cn/), then supplement with other sources as needed; see prompts/prior_art_search.md for the workflow.

When supplementing materials or correcting errors on an existing disclosure file, you don’t need to explicitly say “iterate”—the skill will handle it via merger.md / correction_handler.md; see SKILL.md for details.

Project Structure

This repository follows AgentSkills (https://agentskills.io); the root directory constitutes one skill:

patent-disclosure-skill/
├── SKILL.md                   # Entry point: trigger conditions, tool table, steps, and prompt references
├── prompts/                   # Step-by-step templates (read by Agent, then followed)
│   ├── intake.md
│   ├── project_scan.md
│   ├── patent_points_analyzer.md
│   ├── prior_art_search.md
│   ├── disclosure_preview.md
│   ├── disclosure_builder.md
│   ├── disclosure_self_check.md
│   ├── iteration_context.md
│   ├── merger.md
│   ├── correction_handler.md
│   └── template_reference.md
├── tools/                     # mermaid_render, md_to_docx, docx_to_md, pptx_to_md; cnipa_epub_* (novelty search); iteration_dialog_log, etc.
├── docs/                      # PRD, repository structure description, screenshots (example-effect-*.jpg)
├── examples/                  # Sample raw materials (e.g., example_batch_job_scheduler/knowledge/)
├── outputs/                   # User outputs; entire directory is .gitignore'd
├── requirements.txt
├── LICENSE
├── INSTALL.md
└── .gitignore

Examples

Fictional scanning raw materials are available in examples/README.md (e.g., examples/example_batch_job_scheduler/knowledge/).

Complete artifacts (patent points, novelty search notes, disclosure, etc.) are generated by the workflow and saved locally in outputs/{case_identifier}/.

Screenshots

Initial generation (first delivery)

Initial generation: timestamped disclosure files, Mermaid diagram directories, etc. in the outputs folder.

Iterative update (merge/correct and re-deliver, multiple versions coexist + dialogue log)

Iterative update: new timestamped files and disclosure revision dialogue log.

References

• Entry Point & Agent Workflow (trigger conditions, prompts/ mapping, tool table)
• Detailed Installation Guide (Claude Code / Cursor paths)
• Diagram & Conversion Scripts (mermaid / mmdc, Word export, CNIPA epub search tool)
• Example Case & Raw Material Description
• Product Workflow & Directory Conventions
• Engineering Structure Description
• Disclosure Template Details

Support the Author

If this Skill saves you time writing disclosures, feel free to buy me a coffee ☕ — any support is appreciated. Thank you! 🙏🙏

MIT License © handsomestWei (https://github.com/handsomestWei/)