2026-06-14 · repo-slopscore: Detecting AI/LLM contributions in git repositories via commit history analysis

Transcript

Transcript. Paraphrased; sources in notes.md.

HostIt is June 14th, 2026, and you are listening to another daily dive into the weeds of the tech world. I am Daniel, and joining me today is our resident optimist, Samantha. Samantha, we have a packed slate today. We are looking at some aggressive repository scanners, a project that tries to bring a little bit of nineties-style social chaos back to the modern web, and a high-stakes manifesto on why AI needs to be open source.

GuestHi Daniel, great to be here. Honestly, I saw that repository scanner on Lobsters this morning and I am already buzzing. It is this funny, slightly hostile little tool, and it got me thinking about how we treat our git history as a sacred text. People are getting so worked up about whether a commit was written by a human or a prompt, and I just want to move past the finger-pointing and get to the shipping.

HostYou might be in the minority on that one. The project is called repo-slopscore, and it is a tool that scans your git commit history to calculate how much of it looks like it came from an LLM. It is essentially a purity test for your source code. Over on Lobsters, the community is not exactly thrilled with it. The consensus seems to be that it is less of a useful metric and more of an automated way to be unpleasant to maintainers.

GuestIt is the ultimate gatekeeping tool. Look, I use AI to help me boilerplate some of my unit tests or to clean up documentation, but I am still the one looking at the logic. To quote a Lobsters user called hungariantoast directly, they think the project feels like an automated way to express disdain for methods you disagree with. They point out that nixpkgs, a massive repository, gets a failing grade because of a tiny fraction of commits flagged by a heuristic. That is just performative math. It does not tell you if the code is actually good.

HostIt definitely feels like the tool is looking for a villain rather than a bug. It ignores the reality of how modern development actually happens. What do you make of the second story, then? This is something called Town Square, a plugin that lets you add real-time, transient avatars to your personal website so visitors can see each other. It is basically trying to turn a solitary webpage into a living room.

GuestI love this, Daniel. I am tired of the web feeling like a series of museums. The author of the project, Cauenapier, wants to bring back that early-internet charm where you felt like there were other people on the other side of the screen. I would actually use this on my personal site, just to see if someone else is reading my dev notes at the same time. But I get the anxiety. A Lobsters user named singpolyma hits the nail on the head, pointing out that even if you delete the chat history, you are still hosting an unmoderated space under your own name. That is a real risk.

HostIt is a classic trade-off. You gain that sense of community, but you inherit the responsibility of a moderator. In the Lobsters thread, a user named reezer mentions that this kind of experiment is something we have essentially outlawed through our fear of liability. It is a shame, because it is exactly the kind of playful friction that makes the web interesting.

GuestExactly. If we only ever build for total safety, we just get corporate, sanitized blobs. I would rather have a messy, transient chat room than another static landing page. Speaking of blobs, though, we have to talk about this AI infrastructure manifesto. The author argues that open-source AI needs to win, or we are going to be locked into a subscription economy for our own thoughts.

HostThe claim there is pretty existential. The author argues that if we let closed institutions control the models, we lose our operational freedom. He believes that intelligence infrastructure should be locally deployable and community-governed, regardless of what the major labs do. But the Lobsters crowd was pretty cynical about the definition of open source here.

GuestThey have to be. A user named Internet_Janitor makes a really stinging point, saying that running a pre-compiled model is not true freedom, it is just curated dependence on the megacorporations that have the compute power to train these things. To quote them, they say it is an act of cultivated dependence and an endorsement of technology that structurally cannot be controlled by individuals. It is a grim take, but they are not wrong about the power imbalance.

HostIt really highlights the gap between what we want to happen and the current reality of the hardware bottleneck. Some users, like creesch, pointed out that there are smaller, public-funded models that actually follow open-source principles, even if they aren't pushing the frontier of intelligence yet. It feels like we are in a transition phase where we have to choose between convenience and control.

GuestAnd I am always going to pick the side of the builder. I will take a slightly less powerful model that I can run on my own machine over a black-box API any day. I guess my weekend project is going to be messing with some local weights, just to prove to myself that I can still own the stack.

HostThat sounds like a perfectly productive way to spend a Saturday, Samantha. I think we have successfully navigated the ethics of slop, the risks of digital town squares, and the fragility of the open AI dream. Thanks for coming on, Samantha. And for our listeners, every story we talked about today was sourced from the community over on Lobsters. We will be back tomorrow. Catch you all then.