Mole developer Tw93 says the cleanup app’s recent changes came less from adding features quickly than from repeatedly installing software, checking what it left behind and revising what Mole should—or should not—remove. In a November 13, 2026 account published by 少数派, Tw93 describes work on Mole versions 1.13 through 1.19 and the Windows desktop release on October 14. The author disclosed a direct interest in the product.
For the Mac app, Tw93 says they tested 853 applications by installing and uninstalling them, with an AI agent helping inspect files and automate parts of the process. An earlier round covering more than 300 apps led to 13 reusable cleanup rule categories, more than 90 specific paths and over 100 unit tests. It also identified more than 30 places Mole had been set to clean but should leave alone—including data that might belong to users or be shared by other apps.
The developer also tested more than 30 AI tools, including 13 desktop and 13 command-line tools. Those tests changed Mole’s approach to AI-generated files: users can review older sessions by age, with a recovery copy made before deletion; worktrees appear for cleanup only when Git can establish that they contain no uncommitted changes. For opencode.db, Mole compacts the database rather than deleting conversations, Tw93 says.
System Data required similar caution. Tw93 notes that macOS’s System Data figure is a storage classification, not a list of files safe to delete. Mole therefore reports what it can measure and handle rather than claiming it can clear the entire figure. Large items such as iPhone backups and virtual machines are shown with management options but excluded from selectable cleanup.
On Windows, Tw93 says the desktop app was built with C# and WinUI 3, drawing on an existing Windows command-line version rather than reusing the Mac app’s Swift code. The developer reports testing 482 Windows applications, mostly in a Parallels virtual machine and also on a ThinkPad X1. These are the developer’s accounts of their own testing, not independent performance or safety benchmarks.