I built a Skill: ky-markdown-rebuilder

It is specifically designed to rebuild “visual documents” like PDFs, PPTs, white papers, proposal documents, course materials, and long screenshots into structurally reliable Markdown. Since AI struggles to read these source files, Markdown not only makes them more readable for AI, but also turns these source files into reusable and reconstructable assets.

Microsoft’s official MarkItDown exists, but anyone who has used it knows it’s not very good; it’s more suited for “format conversion” and “text extraction.”

For regular Word documents, web pages, and linear PDFs, MarkItDown is sufficient. But when it comes to complex reports like PPTs, sales decks, charts, etc., problems arise:

• Left and right columns get mixed up, timelines become disordered, architecture diagrams turn into scattered text, tables, screenshots, and annotations get jumbled together. When AI later reads it, it takes the messy text as fact.

ky-markdown-rebuilder first extracts text, then renders the original file into page-by-page screenshots, and reorganizes the Markdown against the real layout.

It can do the following:

1. Page-by-page restoration
   Each page forms its own section, without merging adjacent pages.

2. Organize by visual relationships
   Left/right columns, cards, flowcharts, timelines, comparison matrices, and screenshot annotations are reconstructed according to the original structure.

3. Suitable for complex documents
   Handles PPTs, PDF reports, white papers, course materials, product decks, sales proposals, and long screenshots.

4. Supports deep visual mode
   For pages with high information density, it specifically checks layout, charts, and module relationships, not just extracting a few lines of text.

5. Can also generate an outline
   If an outline is needed, it generates one based on the calibrated Markdown, rather than summarizing from disordered text directly.

I used a middle school chemistry textbook and a PPT sent by a partner today, and produced some example result images. You can see the effects intuitively in the images below.

A well-structured Markdown document can greatly improve efficiency during use.

I’ve put this Skill on Github. If you need it, feel free to try it out!
https://github.com/KyrieCheungYep/ky-markdown-rebuilder

KyrieCheungYep/ky-markdown-rebuilder

Source: https://github.com/KyrieCheungYep/ky-markdown-rebuilder

KY Markdown Rebuilder

A document reconstruction Skill for Codex / Claude Code.

Purpose: To rebuild visually complex PDFs, PPTs, long screenshots, reports, proposal decks, etc., into clear, reliable, page-aligned Markdown.

Preserves page structure, tables, cards, diagram relationships, and visual reading order as much as possible.

Suitable for

• PPT / PPTX / slide deck
• PDF reports, white papers, course materials
• Documents with mixed text and images
• Materials containing tables, flowcharts, architecture diagrams, timelines, matrices
• Long screenshots or visually dense pages

Not suitable for simple plain text files. For plain text files, a direct conversion is sufficient; this Skill is not needed.

Install to Codex

mkdir -p ~/.codex/skills
git clone https://github.com/KyrieCheungYep/ky-markdown-rebuilder.git ~/.codex/skills/ky-markdown-rebuilder

Install to Claude Code

mkdir -p ~/.claude/skills
git clone https://github.com/KyrieCheungYep/ky-markdown-rebuilder.git ~/.claude/skills/ky-markdown-rebuilder

Usage

In Codex or Claude Code, simply reference this Skill:

[$ky-markdown-rebuilder] Convert this PPTX into Markdown

You can also trigger it with natural language:

Rebuild this visually complex PDF into page-aligned Markdown

Default Output

Typically generates:

source.calibrated.md

If you request an outline or both files, it will also generate:

source.outline.md

Output Features

For complex documents, the Markdown is rebuilt as closely as possible to the original pages:

• Each page or slide forms its own section
• Preserves page titles and main content
• Tables are converted to Markdown tables where possible
• Cards, columns, flows, and matrices are organized by structure
• Necessary descriptions are added for images, screenshots, charts, and architecture relationships
• Avoids merging page content into a single block of messy text

English

KY Markdown Rebuilder is a document reconstruction skill for Codex / Claude Code.

It converts visually complex PDFs, PPT decks, long screenshots, reports, and proposal documents into clean, reliable, page-aligned Markdown.

It is designed for documents where plain text extraction is not enough, especially when layout, tables, diagrams, cards, timelines, screenshots, or visual hierarchy matter.

Install for Codex

mkdir -p ~/.codex/skills
git clone https://github.com/KyrieCheungYep/ky-markdown-rebuilder.git ~/.codex/skills/ky-markdown-rebuilder

Install for Claude Code

mkdir -p ~/.claude/skills
git clone https://github.com/KyrieCheungYep/ky-markdown-rebuilder.git ~/.claude/skills/ky-markdown-rebuilder

Usage

[$ky-markdown-rebuilder] Convert this PPTX into Markdown

Typical output:

source.calibrated.md

If an outline is requested:

source.outline.md