Application Document Workflow in the Patent Disclosure Skill: 6 Steps Explained

The application document workflow follows six sequential stages—guardrails and intake, optional iteration, material validation, core content generation, design patent handling, and final document emission—to transform a raw invention disclosure into a complete Chinese patent filing package (四件套).

The handsomestWei/patent-disclosure-skill repository automates patent preparation through a deterministic orchestration defined in skills/patent-application/SKILL.md. This application document workflow validates inputs, chains specialized prompts, and executes Python tools to generate claims, specifications, figures, and a formatted Word manuscript.

Step-by-Step Breakdown of the Application Document Workflow

Step 1: Guardrails and Intake

The workflow begins by loading safety constraints and collecting parameters. According to SKILL.md lines 14–15, the system first reads prompts/guardrails.md to establish safety boundaries, then executes prompts/intake.md to capture the disclosure directory path and case metadata. This ensures all required context is present before processing begins.

Step 2: Iteration Path for Existing Applications

If the user provides an existing application requiring updates (rather than a full rewrite), the workflow enters a short-circuit branch. The system reads prompts/iteration_context.md followed by prompts/iteration.md, creates a new timestamped output folder under outputs/patent-application/, regenerates the Word document, and produces a fresh issue list (问题清单.md). The workflow terminates here; steps 3 through 5 are skipped entirely when this path is taken.

Step 3: Material Gate Validation

Before content generation, the workflow executes tools/material_gate.py to verify that all source assets—such as technical drawings, CAD exports, and schema files—are present in the disclosure directory. If validation fails, the script exits with code 2, aborting the entire workflow immediately. A zero exit code permits progression to content generation.

Step 4: Core Content Generation for Invention and Utility Models

For standard invention patents (发明专利) and utility models, the workflow chains six sequential prompts to build the technical document:

1. claim_strategy.md – Establishes claim breadth and dependency strategy
2. claims_builder.md – Drafts independent and dependent claims
3. figures.md – Generates numbered technical drawings
4. specification_builder.md – Composes the detailed description
5. numeral_register.md – Registers reference numerals for figures
6. consistency.md – Validates internal consistency between claims and specification

This sequence ensures that claims, figures, and specification remain synchronized before final assembly.

Step 5: Design Patent Branch

When processing design patents (外观设计), the workflow diverges after material validation. Instead of the six-step core chain, the system reads references/design_view_cnipa.md to load CNIPA figure requirements, then executes prompts/design_application.md. Notably, this branch does not invoke the figure composition scripts used for invention patents.

Step 6: Emission of Word Document and Issue List

The final stage executes tools/emit_application_docx.py to render the completed Markdown components into a formatted Word manuscript. Subsequently, the workflow reads prompts/issues.md to compile ambiguities or missing disclosures into 问题清单.md. The script appends the issue list path and a brief summary to the final response, completing the four-piece deliverable stored under outputs/patent-application/{case_id}_{timestamp}/.

Implementation Commands

Below are the command-line invocations that implement each stage of the application document workflow. These commands reference the Python tools located in skills/patent-application/tools/:

# Step 1: UTF-8 I/O enforcement (handled automatically during intake)

python -m skills.patent-application.tools.stdin_utf8

# Step 2: Optional iteration for existing cases

python skills/patent-application/tools/iteration_dialog_log.py \
    --case-dir outputs/patent-application/案_20231101091523

# Step 3: Material gate validation

python skills/patent-application/tools/material_gate.py \
    --case-dir outputs/patent-application/案_20231101091523

# Exit codes: 0 (continue) or 2 (abort)

# Step 4: Core content generation sequence

python skills/patent-application/tools/claims_builder.py
python skills/patent-application/tools/figures.py
python skills/patent-application/tools/specification_builder.py
python skills/patent-application/tools/numeral_register.py
python skills/patent-application/tools/consistency.py

# Step 5: Design patent processing (if applicable)

python skills/patent-application/tools/design_application.py

# Step 6: Final document emission and issue tracking

python skills/patent-application/tools/emit_application_docx.py \
    --dir outputs/patent-application/案_20231101091523

Summary

• The workflow is defined in skills/patent-application/SKILL.md and produces a four-piece deliverable (Word document, claims, specification, and figures) under timestamped directories.
• Iteration mode provides an early exit path for updating existing applications without regenerating core content.
• The material gate (tools/material_gate.py) enforces asset completeness with exit code 2 as a hard failure signal.
• Content generation follows a strict prompt chain: claim strategy → claims → figures → specification → numeral register → consistency check.
• Design patents bypass the core chain in favor of a specialized two-step prompt sequence.
• Final output is assembled by tools/emit_application_docx.py, which also generates the mandatory issue list (问题清单.md).

Frequently Asked Questions

What triggers the early exit in the application document workflow?

The iteration path triggers an early exit when updating an existing application. When the workflow detects a revision request (rather than new drafting), it executes iteration_context.md and iteration.md, regenerates the Word document in a new timestamped folder, and stops before reaching the material gate or content generation stages.

How does the material gate enforce requirements?

The material gate runs tools/material_gate.py to inspect the disclosure directory for required assets. If any mandatory files—such as technical drawings or schema diagrams—are missing, the script exits with code 2, immediately halting the workflow. A zero exit code confirms asset presence and permits progression to Step 4.

What distinguishes the design patent branch from standard invention patents?

Design patents follow a simplified two-step sequence using references/design_view_cnipa.md and design_application.md, whereas invention patents execute a six-step prompt chain including claim strategy, claims building, figure generation, and specification drafting. The design branch specifically omits the figure composition scripts invoked for utility models.

Where are the final deliverables stored?

The workflow deposits all outputs into outputs/patent-application/{case_id}_{timestamp}/, including the Word manuscript generated by emit_application_docx.py, the Markdown source files, and the 问题清单.md issue list. This path structure ensures version control through timestamped subdirectories.

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