Patent-Disclosure-Skill Sub-Skills: A Complete Guide to the 7 Modular Capabilities

The patent-disclosure-skill repository provides seven autonomous sub-skills—Disclosure, Application Documents, Docket, Search, Reading, Office-Action Response, and Policy Brief—each encapsulated in a dedicated skills/*/ folder with isolated tools, prompts, and entry-point commands.

The patent-disclosure-skill open-source project implements a modular AI-driven workflow for Chinese patent prosecution. Its architecture splits complex IP tasks into seven distinct patent-disclosure-skill sub-skills, ensuring each capability remains isolated, testable, and independently invocable via slash commands defined in the root SKILL.md.

The Seven Patent-Disclosure-Skill Sub-Skills Explained

The repository organizes capabilities into self-contained units under the skills/ directory. Each sub-skill follows a strict contract: it must expose a SKILL.md entry point, maintain its own prompts/ and tools/ subdirectories, and avoid cross-module dependencies (enforced per line 27 of the root SKILL.md).

交底 (Disclosure) — Mines invention points, performs lightweight prior-art searches, and drafts disclosure documents for inventions, utility models, or designs. Invoke via the /patent-disclosure slash command; core logic resides in skills/patent-disclosure/SKILL.md with crawling tools located under tools/crawl/cnipa_epub_search.py.

申请文件 (Application Documents) — Transforms existing disclosures into the four-document filing set required by CNIPA: claims, description, abstract, and drawings. Trigger with /patent-apply or /patent-application; the primary conversion tool is skills/patent-application/tools/emit_application_docx.py.

案卷 (Docket) — Executes a one-shot workflow that synthesizes both a disclosure and its accompanying application documents directly from raw inventor materials. Use the /patent-docket entry point; the orchestration script is skills/patent-docket/tools/init_docket.py.

检索 (Bibliographic Search) — Performs advanced CNIPA queries against published patents using inventor names, applicants, classifications, titles, or abstracts. Access via /patent-search; the query engine is implemented in skills/patent-search/tools/cnipa_search.py.

解读 (Patent Reading) — Parses patent publication numbers or PDF files to generate plain-language summaries and visual knowledge graphs. Invoke with /patent-read or /读 Patent; the vault integration tool is skills/patent-reader/tools/vault/write_patent_obsidian_note.py.

审查答复 (Office-Action Response) — Assists with examiner office-action question-and-answer sessions, drafts formal response documents, and optionally ingests cases into a knowledge vault. Call via /oa or /patent-oa; the opinion generator is skills/patent-oa/tools/emit_opinion_docx.py.

政策简报 (Policy Brief) — Produces strategic briefings that align the latest CNIPA examination policy updates with specific disclosure or filing strategies. Trigger with /policy-brief or /patent-brief; the analysis playbook runs from skills/patent-exam-policy/tools/playbook.py.

Modular Architecture and Routing Layer

The root SKILL.md (lines 14‑22) functions as the central routing table, mapping each user-invokable slash command to its corresponding sub-skill directory. This entry file declares the command interface, while the heavy lifting occurs inside each sub-skill’s own SKILL.md and executable scripts.

Each patent-disclosure-skill sub-skill package contains:

• A local SKILL.md describing the capability and command-line interface
• A prompts/ directory holding instruction templates for LLM interactions
• A tools/ directory containing executable Python scripts that implement business logic
• Unit tests under tests/ to ensure reliability

The architecture enforces strict isolation: a sub-skill must not invoke tools from another sub-skill (see line 27 of root SKILL.md). This prevents hidden side-effects and guarantees that dependencies remain explicit and predictable.

Running the Patent-Disclosure-Skill Sub-Skills

Each sub-skill exposes command-line scripts in its tools/ directory. Below are minimal invocations from the repository root, demonstrating how to exercise each capability independently.

Generate a draft disclosure for an invention:

python skills/patent-disclosure/tools/crawl/cnipa_epub_search.py --type invention --title "智能数据处理"

Convert an existing disclosure into a four-document filing set:

python skills/patent-application/tools/emit_application_docx.py --disclosure-dir outputs/patent-disclosure/my_project

Execute the full one-shot docket workflow:

python skills/patent-docket/tools/init_docket.py --material-dir inputs/inventor_material

Query CNIPA bibliographic data:

python skills/patent-search/tools/cnipa_search.py --inventor "张三" --max-pages 2

Parse a patent PDF and create a knowledge-graph note:

python skills/patent-reader/tools/vault/write_patent_obsidian_note.py --pdf inputs/patent.pdf

Draft an office-action response document:

python skills/patent-oa/tools/emit_opinion_docx.py --oa-json inputs/oa.json

Generate the latest examination policy briefing:

python skills/patent-exam-policy/tools/playbook.py --policy-latest

Each script emits a machine-readable prefix (e.g., EPUB_SEARCH_MD:) and writes user-visible artifacts under the outputs/ hierarchy, complying with the repository’s execution conventions defined in the root SKILL.md.

Summary

• The patent-disclosure-skill repository decomposes patent prosecution into seven isolated sub-skills: Disclosure, Application Documents, Docket, Search, Reading, Office-Action Response, and Policy Brief.
• Each sub-skill resides in skills/<name>/ with its own SKILL.md, prompts/, tools/, and tests/ directories.
• The root SKILL.md (lines 14‑22) routes slash commands like /patent-disclosure and /patent-search to the appropriate module.
• Strict isolation rules (line 27) prevent cross-module tool dependencies, ensuring predictable behavior.
• Executable Python scripts in each tools/ directory implement the core functionality, accepting standardized CLI arguments and writing to the outputs/ folder.

Frequently Asked Questions

How are the patent-disclosure-skill sub-skills organized?

Each sub-skill occupies a dedicated folder under skills/, containing a local SKILL.md entry point, prompts/ for LLM instructions, and tools/ for executable scripts. This structure ensures that capabilities remain modular and independently testable.

Can sub-skills call tools from other modules?

No. According to line 27 of the root SKILL.md, sub-skills are explicitly prohibited from invoking tools belonging to other sub-skills. This isolation rule prevents hidden side-effects and keeps the dependency graph explicit.

What entry points are available for the Disclosure sub-skill?

The Disclosure sub-skill (交底) is accessed via the /patent-disclosure slash command. Its core implementation resides in skills/patent-disclosure/SKILL.md, with prior-art crawling handled by tools/crawl/cnipa_epub_search.py.

Where are the executable tools located for each sub-skill?

Executable scripts are stored in each sub-skill’s tools/ subdirectory. For example, the Search sub-skill uses skills/patent-search/tools/cnipa_search.py, while the Office-Action Response sub-skill uses skills/patent-oa/tools/emit_opinion_docx.py.

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