xiao24bei/xiaobei-skill-image-to-vba

XiaoBei Skill: Image to VBA is an add-on skill for AI coding agents, specifically designed for the OpenAI Codex workflow. Its purpose is to convert a screenshot or photograph of an academic figure, research diagram, or slide into editable Office objects in PowerPoint, Excel, or Word, rather than embedding the original image as a static picture. The approach prioritizes rebuilding visible structure as editable shapes. Text labels, arrows, boxes, tables, chart legends, and coordinate axes are reconstructed as Office Shapes and VBA macro code that recreates them programmatically. Regions that do not translate well into discrete shapes, such as microscope images, complex 3D renders, or logos, are cropped from the source and kept as small raster inserts, with a note in the output identifying them as such. The skill follows a manifest-driven process: it first lists all detected elements and builds a coordinate model before generating any code. This reduces the chance of missing elements or misaligning arrows. A render-verify step is available when Office is installed locally: the skill runs the generated VBA, exports a screenshot, and compares it to the original to measure reconstruction accuracy. Python utility scripts are included for detecting the local Office environment, converting pixel coordinates to the Office shape coordinate system, checking VBA code for common errors before running it, and comparing source and rendered images side by side. Separate scripts handle automating PowerPoint on macOS via AppleScript and on Windows via COM. The README notes that WPS, a common alternative to Microsoft Office, may not run macros automatically, and manual steps are provided for that case. The repository is released under the Apache 2.0 license. It is not affiliated with Microsoft, WPS, or OpenAI.