Core Capabilities of the patent-disclosure-skill Project: Automating Chinese Patent Workflows

The patent-disclosure-skill repository delivers seven modular AI skills that automate the entire Chinese patent lifecycle, from mining technical codebases for patentable inventions to generating disclosure documents, handling examination responses, and monitoring policy changes.

The handsomestWei/patent-disclosure-skill project is an open-source collection of AgentSkills modules designed to streamline intellectual property workflows under the Chinese patent system. At its foundation, the patent-disclosure skill converts raw technical materials—such as source code, project documentation, and R&D logs—into structured invention disclosures ready for legal review and filing.

Core Patent Disclosure Capabilities

The central functionality resides in the patent-disclosure skill, defined in skills/patent-disclosure/README.md. According to the source documentation, this skill helps users mine technical materials to identify writable patent points, generate disclosure documents (交底书) for inventions, utility models, or designs, and provides novelty search, desensitization, and iteration capabilities.

Specifically, the implementation supports five distinct operations:

• Automatic Mining (自动挖掘): Parses input materials including code repositories, design documents, and research logs to automatically extract potentially patentable technical features.
• Disclosure Generation: Produces structured invention disclosure documents formatted for Chinese patent applications across invention, utility model, and design categories.
• Novelty Search (查新): Conducts prior-art searches to assess the patentability of identified technical solutions before drafting.
• Data Desensitization (脱敏): Automatically redacts sensitive business information or technical secrets from generated documents to protect proprietary data.
• Iterative Refinement (迭代): Supports multi-round refinement of generated disclosures based on user feedback or additional technical inputs.

Extended Workflow Automation Skills

Beyond the core disclosure generation, the repository provides six additional skills that form a complete patent management pipeline, as enumerated in the root README.md (lines 121–163). These modules integrate via the AgentSkills framework to handle post-disclosure workflows:

Patent Application Drafting (patent-application)

Transforms approved disclosures into complete filing documents, including claims, specifications, abstracts, and drawing descriptions. This skill manages the conversion from technical descriptions to legally formatted application drafts.

Interactive Docket Management (patent-docket)

Implements a role-play workflow simulating collaboration between disclosure engineers and patent attorneys. The skill automatically plans work procedures, generates draft responses, and interactively requests missing technical facts to complete application files.

Patent Document Analysis (patent-reader)

Processes public patent documents (PDFs or publication numbers) to generate plain-language summaries and visual knowledge graphs. Outputs are structured for integration with Obsidian vaults, enabling linked reference networks for prior-art research.

Office Action Response (patent-oa)

Assists practitioners in drafting responses to examination opinions (审查意见) from the China National Intellectual Property Administration (CNIPA). The skill decomposes examiner rejections, retrieves relevant prior art via RAG (Retrieval-Augmented Generation), and generates structured rebuttal drafts.

Bibliographic Search (patent-search)

Executes advanced searches across CNIPA publication databases using applicant names, companies, classifications, or by uploading product images and claim excerpts. The underlying implementation in skills/patent-search/tools/cnipa_search.py handles query construction and result parsing.

Policy Monitoring (patent-exam-policy)

Tracks the latest examination guidelines and policy changes issued by CNIPA, automatically generating concise briefs to keep patent practitioners informed of regulatory shifts affecting filing strategies.

Key Source Files and Architecture

The repository organizes functionality into skill-specific directories under skills/, with shared utilities in tools/. The following files constitute the core architecture:

• README.md – Root documentation providing the capability matrix and navigation index (lines 121–163).
• skills/patent-disclosure/README.md – Defines the disclosure generation workflow and automatic mining parameters.
• skills/patent-application/README.md – Documents the application document compilation logic.
• skills/patent-docket/README.md – Describes the interactive drafting protocol.
• skills/patent-reader/README.md – Specifies plain-language translation and knowledge graph generation.
• skills/patent-oa/README.md – Details the examination response assistance features.
• skills/patent-search/README.md – Outlines bibliographic search interfaces.
• skills/patent-exam-policy/README.md – Contains policy briefing generation rules.
• skills/patent-search/tools/cnipa_search.py – Core implementation for CNIPA database queries.
• skills/patent-oa/tools/redact.py – Handles sensitive information detection and removal.
• tools/stdout_utf8.py – Utility module ensuring UTF-8 encoding for Chinese character output streams.

Programmatic Usage Examples

While designed for the AgentSkills framework, you can invoke these capabilities programmatically using the standard skill execution API. Below are runnable patterns for the three primary workflows:

# Example 1: Generate a disclosure draft from project materials

from agentskills import run_skill

disclosure_result = run_skill(
    name="patent-disclosure",
    inputs={"material_path": "/path/to/project"},
    trigger="交底书"
)
print(disclosure_result["draft"])

# Example 2: Convert a patent PDF to a plain-language analysis

reader_output = run_skill(
    name="patent-reader",
    inputs={"pdf_path": "/path/to/patent.pdf"},
    trigger="读专利"
)
print(reader_output["summary"])

# Example 3: Search CNIPA records using a product image

search_result = run_skill(
    name="patent-search",
    inputs={"image_path": "/path/to/product.jpg"},
    trigger="著录检索"
)
print(search_result["records"])

These calls rely on the internal tool modules—such as cnipa_search.py for web scraping and redact.py for data sanitization—to handle the underlying heavy lifting.

Summary

• The patent-disclosure-skill repository provides seven modular AI skills covering the complete Chinese patent lifecycle.
• The core patent-disclosure skill automatically mines technical materials, generates disclosure documents, conducts novelty searches, and manages data desensitization.
• Extended skills handle application drafting, interactive docket management, patent reading, office action responses, bibliographic searches, and policy monitoring.
• Implementation relies on specific source files including skills/patent-disclosure/README.md, skills/patent-search/tools/cnipa_search.py, and skills/patent-oa/tools/redact.py.
• All capabilities are accessible via the AgentSkills framework using standardized run_skill invocation patterns.

Frequently Asked Questions

What is the primary function of the patent-disclosure skill?

The patent-disclosure skill serves as an intelligent assistant that analyzes technical materials—such as code, documentation, and logs—to identify patentable inventions and generate structured disclosure documents (交底书) suitable for Chinese patent filings. It additionally provides integrated novelty searching and sensitive data redaction to prepare documents for legal review.

How does the patent-search skill handle image-based queries?

According to the source architecture in skills/patent-search/tools/cnipa_search.py, the patent-search skill accepts image uploads (such as product photographs) as input parameters. It processes these images to extract visual features or text via OCR, then constructs specialized queries for the CNIPA publication database to retrieve relevant prior art or similar外观设计 (design) patents.

Can these skills be used outside the AgentSkills framework?

While the repository is architected for the AgentSkills ecosystem—evidenced by the standardized run_skill interface and shared utilities like tools/stdout_utf8.py—the underlying Python modules in skills/*/tools/ can theoretically be imported directly. However, the intended usage pattern involves invocation through the AgentSkills runtime to ensure proper context management and UTF-8 handling for Chinese text.

How is sensitive information handled during the patent workflow?

Sensitive data management is implemented primarily in skills/patent-oa/tools/redact.py, which provides automated desensitization capabilities. During both the initial disclosure generation and examination response phases, this module scans content for business secrets, technical identifiers, or personal information, applying redaction rules before documents are stored or transmitted.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}