Patent Mining and Technical Disclosure Generation

Problem

Patent disclosure drafting often stalls when engineers cannot decide which technical points merit filing, when similar prior art makes a point seem unwritable, or when the draft lacks implementation details such as data flows, parameters, edge cases, and module mappings. This skill turns that process into gated steps instead of producing a high-level concept document.

How It Works

• Patent point mining: reads code, design documents, PDF, Word, and PowerPoint inputs, then creates a patent-point inventory that separates filing candidates, weak points, and trade-secret candidates.
• Selection and light screening: screens each candidate against similar prior art, records the closest references, similarity level, and distinguishable features; the user confirms what to draft before the process continues.
• Deep prior-art analysis: evaluates coverage rather than rejecting a point outright, then narrows, rewrites, or drops the point based on overlap.
• Specification-style drafting: requires method steps, system modules, beneficial effects, end-to-end data flow, key data structures, processing rules, input/output examples, exception paths, and alternative implementations, with mermaid diagrams for flowcharts and system structure.
• Versioned delivery: stores Markdown, Word, PDF, images, and patent-family notes in version folders, preserving iteration logs and avoiding overwrite of earlier drafts.

Boundaries

It is suited to engineering teams, IP engineers, and patent agents preparing technical disclosures. Inventor details must not be fabricated, trade-secret points must not be expanded into public descriptions, and automatic image generation should stop if the required image-service configuration is missing.