A simple tool to figure out just how much LLM generated code has been used in a project. Obviously not perfect, but it automates the task of going through commits for “co-authored by” statements and other signs. In the off chance you don’t want to use the CLI, there’s also a web app!
I mean, realistically speaking that’s what I would do manually. So it’s just making my life easier by automating it.
I don’t expect it to analyse individual files.
Exactly. It can produce false negatives, but by nature of Git’s blaming everyone for every change, it makes it (probably) won’t produce false positives. It’s a give and take in that instance. Do you do a witch hunt and try and track down every LLM generated repo and get some legit repos caught up in an imperfect process. Or do you use actionable evidence to make an informed decision?
In my opinion, witch hunts are really quite unhealthy for society. When possible, I choose projects with blanket no-AI statements, but when not, this expedites the process of choosing between alternatives. Rustic has less AI generated code than Restic, let’s do that. Kitty has hundreds of AI generated commits, let’s delete that immediately.
Yeah, and at least the site shows you the content of the files.
I think there can be a reasonable way for it to get false positives, and it’s when a repo has an
AGENTS.md/<insert-llm-here>.mdfile, or a.skills/,cursor/ortabby/folder, for example, but they’re adversarial (for example, theres awful.system’s AAA-NO-SLOP.md).