I didn't write a single line of Desloppify's 56K LOC codebase and, while I spent a lot of time designing the overall approach, I don't meaningfully understand a lot of its actual implementation.
Despite this, I'd invite any senior engineer to look at the codebase and tell me where the warts are — I think they'll have trouble finding many.
This is because I used Desloppify on itself: it combines subjective agent reviews (modern agents can genuinely understand high-level issues like abstraction quality, naming consistency, error handling patterns) with mechanical issue detection, then gives your agent a prioritised list to work through to get everything as clean as possible.
Combined with anti-gaming measures, this allows it to be run autonomously for extended sessions, chipping away at real issues across scans.
Deeply supports TypeScript, Python, C#, Dart, and GDScript with shallow support for 23 more languages.