Main Capabilities of the Patent-Disclosure-Skill: A Complete Agent-Skills Workflow for Chinese Patents

The patent-disclosure-skill is a modular Agent-Skills package that automates end-to-end Chinese patent workflows through five isolated sub-skills: disclosure drafting, CNIPA bibliographic search, plain-language reading, office-action response, and policy briefing.

The handsomestWei/patent-disclosure-skill repository provides a structured framework for intellectual property professionals working with Chinese patents. Built as a composable Agent-Skills system, it bundles specialized tools and prompts under a unified root router that maps natural-language intents to specific patent tasks. This architecture ensures that each capability—from prior-art mining to examination response generation—operates within strict isolation while sharing common output conventions.

Five Core Patent Capabilities

The root SKILL.md (source lines 14-20) defines five distinct sub-skills, each residing in its own directory under skills/ with dedicated prompts, tools, and documentation.

Patent Disclosure Drafting (交底)

The disclosure capability automates the creation of invention disclosure documents for invention patents, utility models, and design patents. Located in skills/patent-disclosure/, this workflow scans project directories, mines patent-point candidates, performs lightweight prior-art searches, and generates both Markdown and Word outputs.

Key tools include cnipa_search.py for prior-art checks and md_to_docx.py for document conversion. The process handles line-art generation, figure planning, auto-naming, self-check routines, and iterative revision. Outputs land in outputs/<case>/disclosure.md and corresponding .docx files, providing a complete technical disclosure package ready for patent attorney review.

CNIPA Bibliographic Search (检索)

The record search capability executes advanced queries against the China National Intellectual Property Administration (CNIPA) "epub" publication portal. Defined in skills/patent-search/SKILL.md, this sub-skill returns bibliographic listings filtered by inventor, applicant, classification symbols, or keywords.

The core script skills/patent-search/tools/cnipa_search.py handles pagination and result formatting, saving markdown reports to outputs/patent-search/. Users trigger this via the /patent-search natural-language command or direct CLI invocation to monitor competitor filings or validate inventor portfolios.

Plain-Language Patent Reading (解读)

The patent reader transforms complex patent PDFs into accessible knowledge graphs. Found in skills/patent-reader/, this capability extracts technical "key-points," builds terminology tables, and generates Obsidian-compatible canvas files for visual knowledge mapping.

Using skills/patent-reader/tools/pdf_text.py, the system parses patent documents from PDF inputs or public numbers, creating structured outputs in outputs/patent_reader/. This enables rapid technical due diligence and competitive analysis without manual claim charting.

Office-Action Response Generation (审查答复)

The OA response skill streamlines the creation of examination response documents. Located in skills/patent-oa/, it ingests examiner office-action notices and produces structured draft responses moving from outline to full text to final Word documents.

The tool skills/patent-oa/tools/emit_opinion_docx.py parses rejection rationales and suggests argumentative frameworks, optionally pulling precedent cases from a local "vault" repository. This reduces attorney drafting time while ensuring consistent response formatting for CNIPA examination proceedings.

Examination Policy Briefing (政策简报)

The policy brief capability monitors CNIPA examination guidelines and summarizes their impact on disclosure drafting practices. Implemented in skills/patent-exam-policy/, this sub-skill retrieves the latest examination policies—such as updates regarding AI-related inventions—and generates impact reports in outputs/exam-policy/.

Unlike other skills, this capability operates in read-only mode relative to the workflow state, ensuring it never mutates active disclosure documents unless explicitly instructed. The system produces markdown briefs (POLICY-*.md) that keep inventors and agents current on procedural changes.

Routing and Sub-Skill Isolation

The architecture enforces strict capability isolation through the root SKILL.md router (lines 24-28). This design prevents cross-contamination between workflows—a disclosure flow cannot accidentally invoke policy-brief tools, and search operations remain sandboxed from OA response generation.

Each sub-skill follows a consistent directory convention:

• SKILL.md defining entry points and constraints
• prompts/ containing LLM templates
• tools/ housing Python automation scripts

User-facing commands use intuitive natural-language triggers: /交底书 (disclosure), /patent-search, /读专利 (read patent), /oa, and /政策简报 (policy brief). All outputs write to unified outputs/<skill-name>/ directories, enabling straightforward downstream automation and artifact collection.

CLI Usage and Toolchain Examples

While the Agent framework handles natural-language routing, each capability exposes direct CLI access for automation pipelines:

# Draft an invention disclosure from project source code

python -m skills.patent-disclosure.tools.disclosure_builder \
    --type invention \
    --project /path/to/project

This command scans the specified project, identifies patentable technical points, executes a lightweight prior-art check via cnipa_search.py, and writes both disclosure.md and disclosure.docx to outputs/<case>/.

# Search CNIPA for inventor "张三"

python -m skills.patent-search.tools.cnipa_search \
    --inventor "张三"

Returns paginated bibliographic records from the CNIPA EPUB database, saving the report to outputs/patent-search/.

# Parse a patent PDF into a knowledge graph

python -m skills.patent-reader.tools.pdf_text \
    /path/to/patent.pdf

Extracts technical content and generates an Obsidian canvas in outputs/patent_reader/.

# Generate an office-action response draft

python -m skills.patent-oa.tools.emit_opinion_docx \
    --notice /path/to/oa.pdf

Produces a Word document response draft based on the examiner's rejection notice.

# Create a policy brief on AI examination standards

python -m skills.patent-exam-policy.tools.policy_brief \
    --topic "AI"

Fetches the latest CNIPA guidelines regarding AI inventions and writes a summary to outputs/exam-policy/.

Summary

• The patent-disclosure-skill provides five specialized sub-skills for Chinese patent workflows: disclosure drafting, CNIPA search, plain-language reading, office-action response, and policy briefing.
• Strict isolation is enforced by the root SKILL.md router, ensuring that tools from one workflow cannot inadvertently affect another.
• Unified output conventions place all artifacts under outputs/<skill-name>/, facilitating automation and audit trails.
• Bilingual support defaults to Simplified Chinese UI but supports multilingual outputs while maintaining stable machine-readable fields.
• Extensible tooling includes scripts like md_to_docx.py, cnipa_search.py, and emit_opinion_docx.py that function both offline and against live CNIPA databases.

Frequently Asked Questions

What is the patent-disclosure-skill used for?

The patent-disclosure-skill is an Agent-Skills framework designed to automate Chinese patent prosecution tasks. It enables inventors and patent professionals to generate technical disclosure documents, search CNIPA bibliographic records, simplify patent reading through knowledge graphs, draft examination responses, and track policy updates—all through a modular, command-line accessible system.

How does the root SKILL.md coordinate the five capabilities?

The root SKILL.md acts as a capability router that maps user intents (expressed as natural-language commands like /交底书 or /patent-search) to specific sub-skill entry points. According to the source code (lines 14-20), it declares the five high-level capabilities and enforces isolation rules that prevent cross-package tool calls, ensuring each workflow operates within its defined scope.

Where are the output files generated by each sub-skill?

All sub-skills write to dedicated directories under the outputs/ folder. The disclosure skill writes to outputs/<case>/, the search skill to outputs/patent-search/, the reader to outputs/patent_reader/, the OA response skill to its respective subdirectory, and the policy briefs to outputs/exam-policy/. This unified structure simplifies downstream processing and artifact management.

Can the patent-disclosure-skill operate offline?

Yes, several tools function without internet connectivity. For example, md_to_docx.py handles document conversion locally, and the disclosure builder can generate line art and structural diagrams offline. However, capabilities like cnipa_search.py require online access to query the CNIPA EPUB publication portal for prior-art checks and bibliographic searches.

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