My 24/7 Claude Code AI Agent's Biggest Win... That Won't Happen Again

[00:03] agent uh review. So this uh Mac Mini AI agent here has been running like almost 24/7 for the last 12 days, like I said. And yesterday I mentioned I wanted to do like an own video on what I think has been the biggest accomplishment of the

[00:18] agent. So today I just want to try to show you how that happened and what the outcome was. So I'm going to try to just recreate the instructions I gave it. uh it would not be the same of course but uh and then we will go into what

[00:31] head over to the Mac Mini and I'm going to try to show you what I think has been the biggest accomplishment so far. Okay, so uh I think we should just start by going to our skills here because uh of course this um also relate to the skills

[00:49] here, you can see we have something called a GitHub skill. And I guess I could show you kind of quickly what this does. Let's say I just said open up uh does. Let's say I just said open up uh GitHub. Uh okay. So uh we have already

[01:03] kind of trained our AI agent to navigate GitHub. You can see it goes into my profile. I'm already logged into GitHub, right? Okay. So that is kind of the start. This is what I started with. I wanted to see how well the agent could

[01:16] navigate the browser version of GitHub. And this means that as long as we are logged in, we could navigate and look at projects and everything on GitHub, right? Uh we also can use the API key. So we can do both uh if we wanted to

[01:30] because of committing I kind of like to use the I think it's more efficient to use the CLI for that uh doing version control, right? So now I kind of want to show you how we ended up in the situation where I think is kind of the

[01:44] biggest accomplishment of the agent so far. Uh but before I show you that I want to tell you quickly about today's sponsor and that is codebuff. For today's video we have partnered up with codebuff. So I want to show you how I

[01:57] have been using their CLI tool. To get started on codebuff is super simple. Codebuff.com. You just do like an mpm install g codebuff. I just copy that. that in. You know I have already

[02:10] now we just going to do codebuff right. And we get into this terminal CLI tool is that it kind of can use your skill that you already have set up in your now, you can see, okay, we have all these skills I've been using. I can also

[02:26] do something like list my skills, right? And we just run this. And this should give us a list now of the skills uh we already have. And you can see here we get this nice li list of all my skills, right? Here are 14 skills in your cloud

[02:39] skills. And I got to say, I really like kind of the layout here. So, let's just try it now. So, let's open X. So, let's see if we can use my skill to open up the X page. You can see no hands here. And I want to see if this is

[02:52] transferable over to um yeah, you can see it opened up the X page. Perfect. See, they kind of claim on their page that uh higher quality output and 100 plus seconds faster and using just claw code, but it is kind of running Claude

[03:06] Opus on top of codebuff that is more of like a uh orchestrator and it pulls out like a uh orchestrator and it pulls out faster models in between or like below one more thing about codebuff is that they don't store your codebase anywhere.

[03:20] And of course, they support agents MD, skill MD like we saw. So, that is pretty interesting, too, if you kind of have your workflow already set up. So, to try in the description. And a big thanks to ColdBuff for sponsoring this part of the

[03:33] project. So, since I didn't record this, I'm just going to try to recreate exactly what uh instructions I gave the agent because I can't remember 100%. But it went along something like this. So, I said, I want you to go to a trending uh

[03:47] to trending GitHub repos. This is the page, right? Look for a newest project. I want you to I want to challenge you to be a contributor to an open source be a contributor to an open source project. Uh if you uh you if you find a

[03:59] project that looks interesting, look into what we can contribute with. Check if there's uh any contributions MD. Look at issues etc. If you need to fork the repo, follow the um the yeah the skills we have here or the rules uh within the

[04:13] there was something like that. I gave the instruction to the agent. So, let's just see what happens now. And hopefully it's going to do this in the browser now so we can kind of follow along. So, you can see it goes to the github.com

[04:27] trending and it kind of looks at different projects here. Okay. So, I'm here. But basically, you can see it's working here looking up the context from the page and stuff. And that is kind of what I did. So, I want to see if it kind

[04:39] of settles on a project now. It's going to look more into. Okay. So, you can see it found some different uh posts now. And I'm just going to scroll further down to do the second screenshot. And that should mean that we have every

[04:51] that should mean that we have every single uh GitHub trending repos uh yeah uh covered. If you do the couple of screenshots we need here. Okay. So you kind of project that is called Claude Skills. It's just a public project. Uh I

[05:06] don't know if this was the best selection, but basically this is what I did when I Yeah, you're going to see soon what happened. So now you can see soon what happened. So now you can see it navigated into this um project here

[05:19] uh all by itself, right? And now let's see if it's going to follow our contribution uh you can see contribution to full stack dev skills plugin. You can see we have a contributing MD file here and you

[05:32] can see we have some instructions. We can fork and clone suggesting new skills reporting issues. This is basically what we can contribute it. So what I like here is that it's actually reading uh the contributing MD file. So we kind of

[05:47] get some uh initial uh instructions on how you can contribute to this open source project. Now you can see it's reading all the issues, right? And in theory it could pick some issues. It could clone fork the repo, try to solve

[06:01] some of them. So now it's basically just navigating the the repo here. is looking at kind of what skills are in the in the repo here. So, it's basically just browsing the repo. So, you can see it kind of wants

[06:15] bigger picture of everything. But, uh I'm going to stop it here because I don't want to do that. This was not the part of what I wanted to show you today. I guess it was because I wanted just to showcase how well this agent can kind of

[06:28] navigate GitHub. Uh, of course it can do a lot of this with uh just using the CLI tools too, but I found it a bit interesting just to using the browser interesting just to using the browser for tasks like this. So, uh, I asked it

[06:41] what would it try to if it needed to do some contributions here. It would look into kind of adding some more skills here. This could be like a chronuler, a task automation skill, MCP developer skill. Okay, so that's fine. So what I

[06:56] story of kind of the biggest accomplishment. It did kind of start the same. It kind of went to GitHub and it went to kind of the what do you call it? Uh the trending repos and this if we found this repo called uh nano claw.

[07:14] Okay. So this is a very popular repo. It has So this is a very popular repo. It has like 8,000 stars now. And it's a pretty neat um like a very light version of open

[07:27] claw. So it's called nano claw, right? You get the point. And it's super cool. I've checked it out and I tried it and it's very good. So basically the the workflow you saw earlier was kind of the same. It found this on uh the trending

[07:40] page and it went ahead and kind of did the same thing. It went to the issues, right? It looked at all the issues. It went to if you go to here and we kind of went to if you go to here and we kind of check out the contributing MD file and

[07:55] it kind of looked at this and you can see it had something accepted bug fixes simplifications reducing code and not accepted features. Yeah, you get the point. So it's a clear indication of what you can do to contribute. and we

[08:10] what you can do to contribute. and we read this uh we did fork the repo and it started looking up all these issues. Okay. And it figured out early that a lot of them was already in the works. Okay. So it didn't really find anything

[08:26] to do on the issues part. So it kind of went back again to the contributing part and it kind of looked at uh simplifications if I remember correctly. So or reducing code. I guess that's basically most the same. Uh, okay. So,

[08:42] what happened actually is that uh I saw it started working. It kind of went into do a pull request and it did this uh from the browser actually. Uh I think I can't really remember. So now I want to kind of show you where it landed and I

[08:57] want to show you kind of the small things we did here. So just give me a second. And here you can see it right. So uh if we go into this you can see we have this uh dduplicate logger into shared module. So this was the type of

[09:11] shared module. So this was the type of change we uh put into the PR and you can see we added this like simplification and we described the three files here. Uh each created their own identical pen logger with the same configuration. This

[09:25] extracts into a single logo.ts module replaces the import. So it's already clear. It did a really good way of explaining what it did. So you can see in this repo now we kind of reduced the lines of code and this was one of the

[09:39] accepted code changes or contrib contributions you could do. So you can see chv here kind of reviewed and tested this on our fork. Looks good to me. Textbook drive refactor three identical pen configs uh consolidated into our

[09:52] logger ts net negative lines zero behavior change clean compression no regression. Okay, recommended merging this into before PR. Okay, and you can this into before PR. Okay, and you can see uh Gabby here push this into then

[10:06] and we got this nice comment. EJ dev less code. Love it. More of this please. So yeah, that was pretty cool and I think this is the biggest accomplishment of the agent so far actually being a contributor to a quite popular project

[10:21] here. And you can see we are now kind of named as one of the 10 contributors. So yeah, that was basically the story of how we got this nice achievement here. We appear extraordininaire and we contributed to an open source project.

[10:35] Uh but I like I said I don't have any more plans of actually using the agent for this type of work. Now uh like I said if uh I know that a project is run by an agent. So let's say EJ dev here this is kind of my agent. She has a

[10:50] project called MCP cost calculator. And if someone wants to make a pull request here using an AI agent, I don't really care because I know that uh this agent is going to handle those PRs anyway. But uh I wouldn't do this to like a super

[11:05] cool open source project that is someone is spending their time managing. So yeah, like I said, I turned that off on my agent. Uh, and I'm only going to do this on uh agent dribbon repos in the future because I know it's just going to

[11:21] be a flood of those uh request to popular repos. Uh I can't even imagine. Let's say we go to open. What happened here? Let's say we go to open claw now. here? Let's say we go to open claw now. Okay, let's find this project. 190,000

[11:37] Okay, let's find this project. 190,000 stars and yeah, you can see 2,700 pull stars and yeah, you can see 2,700 pull request. So basically this is not this is impossible to handle, right? For I think it's one guy that owns this

[11:49] project. I guess he has a lot of people helping him, but there's just no way you can uh handle all of these pull requests. So this project is kind of

[12:01] broken on the open source part at the moment, right? 3,000 issues. Yeah, you're not going to handle that. So yeah, just something I wanted to yeah, just something I wanted to mention. Uh, I know this uh kind of it

[12:14] contradicts what I said, but I think in my case, this was just a test and like I said, I'm not going to try that again. So, yeah, that was basically the story of what I think has been the biggest achievement of our AI agent so far. And

[12:29] if you like this, uh, just give this video a like and subscribe. And I'm going to keep you updating throughout the next weeks while we do the experiment how it goes. Uh I saw on X that we had another some few successful

[12:42] postings. So our follower count and our YouTube views are still rising. So if you want to see where we kind of end up at the end of the experiment, just uh follow me and you will get the updates. So yeah, don't forget to check out the

[12:55] yeah, thank you for tuning in. Have a great day.