Stop Getting Roasted in PR Review (CodeRabbit, Locally)

I honestly do like coding in our new agentic world. I think it's actually a way to be, you know, more productive and to kind of focus on higher leverage stuff. But I gotta say, it does lean into two areas of serious weakness for me. The first is around writing specs. I'm just not that good at kind of writing ahead of specs. In fact, I kind of write code as my spec, and always have. So I'm not real good at that. The other area of weakness for me is code reviews. I'm just not particularly good at looking at a Git diff and trying to figure out from the Git diff how that change is going to impact the overall system and if it's gonna cause any serious vulnerabilities. And if you let one little thing through, that's a serious problem, particularly when, well, at least I am working on TanStack libraries, which means that the changes that I push out will go to hundreds of thousands of developers. So potentially the risk is really high when it comes to changes to TanStack. So I was very happy when I found out that CodeRabbit was being run on every single PR that was going into all of our libraries. Check this out. Here's a PR that I just recently put up. Fixes a bunch of smaller issues, but in doing so, it actually, if I look down here, introduces a potential critical issue in a very critical area. This is our OpenAI adapter, and this is how we are going to package the messages going back from the server to the client. This is huge. It's done an analysis of what the issue is, and you can go through and see how it did that analysis. And then it gives you a description, sometimes even with sequence diagrams, shows you what's right and what's wrong. And then finally, and this is one part that I really love, you can just copy and paste an AI prompt that's very targeted and give that to your agent, and it'll do the fix. And I gotta say, every time that I've done that, the agents have found and agreed with CodeRabbit. So I really have come to trust and depend on these CodeRabbit reviews. But I'm the kinda guy, I don't actually like to get negative stuff in a PR in the first place. That's why I run all the tests, and I make sure that everything's working before I put it up. So the fact that I only got the CodeRabbit reviews in the PR cycle was a problem for me. I really wanted this, get those earlier in the development process. And I found out, turns out CodeRabbit has a CLI. I didn't know that, and it is easy to integrate with both Claude Code and Cursor, which are the tools that I use. And in fact the way that it's integrated with Cursor, and I'll show you that in a, just a bit, you can actually take lower-end models and because the code review is actually in the process of building out the feature itself, you can actually get lower-end models to do a better job by using CodeRabbit. So it becomes part of the development process, not just something that you do as a check before you do the PR. But of course, it's great that we can do the check before we do the PR. Now, that's why I'm really psyched that CodeRabbit is sponsoring this video, so that I can show you how to do these installs and that they are sponsoring TanStack. So thank you so much to CodeRabbit for doing that. Let's get right into it. It all really starts with bringing in the CLI, and honestly, it couldn't be easier. You could do the curl thing, or you can do the homebrew thing and install it. You have to source the zshrc. Obviously, it's gonna add it onto the path, and then you're gonna wanna log in. Now, I've already done the install, so I'm gonna run auth status, and I am logged in. Now I'm in this example to-do app. As you can see, I've made some changes here to some of these files, but I don't know yet if there's any issues with this. So I'm gonna bring up Claude and show you how to integrate with Claude Code. First you wanna make sure that your plugin marketplace is up to date, and then once you've done that, you can simply do plugin install and then install CodeRabbit. I'm gonna reload the plugins, and from here, I can do CodeRabbit and then request a CodeRabbit review. Now that plugin is running the CodeRabbit CLI to make sure that we're logged in, and now it's actually going to do the run to see if we have any issues in that code. Now, this is actually gonna take a while, and like other folks have said, good things take time. It's actually gonna do a really deep review of all these changes. The cool thing, though, is I could do Control + B at this point, put this into the background, and then do other things at the same time. All right, that took about two minutes or so, but it found some serious issues. And the really cool thing is that it already essentially knows all those fixes. So I could just say, sure, apply fixes and let it roll, or I could look at each one of the individual issues. I'm just gonna let it apply fixes. But in the meantime, I'm gonna show you how to do a similar install into Cursor. And of course, if you have a VS Code-based IDE, this is gonna work for you as well, at least the level of the extension. What you're gonna wanna do is go over into the Extensions panel and then type in CodeRabbit, and then install from there and log in. Once you've got that, you're gonna get the CodeRabbit panel over here. Now, from here, I can see that I've got a bunch of potential issues in this particular version of to-dos. So I'm gonna click on one of those. And this is giving me everything I need to know about this right in place. All I need to do is just kind of scroll down here. I can see the whole issue, and I can either choose to ignore it or fix it with AI right there. And I can just navigate around all these issues the same way. This is a fantastic interface and a fantastic way to use CodeRabbit. There's one more way to do this integration, and I'm gonna do it in Cursor, but there's probably ways to do this in other agentic IDEs as well. I'm gonna go over here to Settings, Cursor settings, and then navigate down to Rules. And in there, I've got a custom rule for the CodeRabbit CLI. So I got this from the CodeRabbit documentation. I didn't write this myself. And it tells the agent that CodeRabbit's already installed and how to use CodeRabbit on uncommitted changes And it makes it clear that it should run CodeRabbit on changes as they occur. So if I go over here to the to-dos and I bring up a new agent. I can just use my agent as I normally would, say for example, add a date field to each to-do. But where I normally would maybe choose something like Opus or Composer, I'm gonna stick with Auto in this case because I can use that Auto model, and then once the Auto model is done with those changes, it's gonna run CodeRabbit as it's working and then apply those changes from CodeRabbit. So I get better results from an Auto model, which I think is fantastic. Let's give it a try. Now we can see it's made a bunch of changes, and before it comes back to me for next steps, it's gonna run CodeRabbit to see, hey, does CodeRabbit like the changes that you just made? And then give it feedback on what it could do better for this particular set of changes. That is fantastic. All right, so there you have it, three different ways to bring CodeRabbit reviews into your pre-PR process. And thank you again so much to CodeRabbit for supporting this channel through this sponsorship and also for supporting TanStack. You know, I get a lot, and I mean a lot, of sponsorship requests for this channel, and I only ever do them if I actually use the product on a day-to-day basis, and CodeRabbit is definitely that product. So thank you so much to CodeRabbit for making a great code review product and really help saving my butt a couple of times. If you got any questions or comments, be sure to put that in the comment section right down below. And if you liked the video, hit that like button. If you really liked the video, hit the subscribe button and click on that bell, and you'll be notified the next time a new Blue Collar Coder comes out.