pengjunchi0/codex-visio-paper-figure-skill

What it solves

This project solves the problem of converting static images (screenshots, PNGs, JPGs, or AI-generated images) of scientific paper figures into fully editable Microsoft Visio (.vsdx) files. It prevents the common issue of simply embedding a reference image into a Visio page, instead ensuring that the final output consists of native, editable shapes, text, and connectors.

How it works

It functions as a "Codex Skill," providing a set of guidelines and PowerShell scripts that interface with the Visio COM Automation API on Windows. The system uses a rebuilding scaffold to map reference image coordinates to Visio page coordinates. For complex multi-panel figures, it employs a panel-calibration process where local coordinates (0-1) are used for internal elements to prevent overlapping or misalignment across different sections of the figure.

Who it’s for

Researchers and academics who need to recreate, modify, or standardize the visual style (fonts, colors, line weights) of model architectures, flowcharts, and scientific diagrams for their publications.

Highlights

• Native Editability: Forces the use of Visio native shapes, lines, and text rather than image embeds.
• Multi-Format Export: Automatically exports the master .vsdx file into PNG, SVG, PDF, and PPTX formats to ensure consistency.
• Panel Calibration: Includes helper functions (RectRel, TextRel, etc.) to handle complex multi-panel layouts and prevent element overlap.
• Automated Scaffolding: Provides PowerShell scripts to accelerate the 1:1 reconstruction of diagrams from reference images.