wtf is archforge?

archforge is a command line tool that checks PowerPoint files for defects that are invisible until someone actually opens the slide deck. It targets a specific problem: when an AI agent or script builds a .pptx file, PowerPoint will open it without any warning even when something is badly broken, because the flaws live in font resolution, text sizing, and shape positions that only show themselves when the file is rendered. archforge reads the raw .pptx file directly, including its XML, font chain, geometry, and image data, so it does not need PowerPoint installed and can run inside a script or continuous integration pipeline. The tool catches things like text that silently falls back to the wrong font because the intended font is missing the needed characters, a problem especially common with Korean, Chinese, and Japanese text on Latin only fonts. It also flags letter spacing that damages CJK characters, text shrunk below a readable size by autofit, text frames that overlap each other, and text or images that run off the edge of the slide. A separate set of checks looks for signs that a deck was written by an AI model rather than a human, such as dash punctuation used the way language models tend to use it, or repeated buzzwords. Getting started takes one command: installing the package with pip and running a built in demo that builds one broken and one fixed presentation and lints both. From there it can be pointed at a real deck, run across a folder of decks, or asked to output results as JSON for another program to read. It also has a fix mode that automatically repairs some of the simpler problems, and can generate an HTML report or output formats meant for continuous integration systems. archforge ships as a GitHub Action and a pre-commit hook, and it separates its checks into profiles: a core profile with only the objective, mechanical defects, and a full profile that adds the AI-tell and style checks, which is meant for pipelines that lint output from AI writing tools. The font resolution logic behind its main check was built by measuring how real PowerPoint actually resolves fonts, rather than relying only on the file format specification. The project is written in Python and distributed under the MIT license.