Codex CLI + New Goal = Full Apps INSTANTLY (FREE*)
Chapters27
The speaker notes that the new goal has been running for 15 hours 30 minutes, highlighting its vagueness and the difficulty of verifying such long backend work.
Codex CLI’s new {slash} goal feature can spin up ambitious app work for a fraction of the cost, raising questions about verification and long-running tasks but delivering surprising, production-ready progress.
Summary
Income stream surfers’ video dives into Codex CLI’s latest {slash} goal capability, showing how a vague prompt can drive hours of backend work for as little as $20 a month. The creator tests the feature by enabling goals in the CLI and watching Codex push toward an “agency ready” product. He compares the experience to Claude’s orchestration and notes the long-running nature of the task (about 15 hours) on a vague directive. Key findings include that the feature is currently CLI-only, not yet exposed in the app, and requires a few config steps (config.toml and enabling features). Throughout the session he validates what Codex delivered by inspecting agency HQ, keywords, content goals, personas, and a writer integration, all while logging deployments and verifying changes in real time. He highlights how this can automatically seed discovery, target keywords, and writing prompts for client projects, effectively turning a vague goal into concrete project scaffolding. The video emphasizes cost efficiency, with the caveat that higher-tier plans (ultra/Pro) would likely extend uptime or guarantee, and notes Harbor’s evolving roadmap around agency automation and Build. In the closing remarks, he positions Codex as a disruptive, budget-friendly accelerator for developers and SEO agencies, encouraging viewers to explore Harbor and Codex together. Finally, he teases pricing stability for Harbor users and hints at future features to come, inviting viewers to watch for more updates.
Key Takeaways
- Codex CLI’s {slash} goal can run a 15+ hour backend effort from a vague prompt, delivering tangible app components and dashboards.
- Enabling the feature requires editing config.toml (features = true, goals) inside Codex CLI, then deploying to see live results.
- The workflow includes tracking keywords, setting content goals, defining personas, and shaping the writer’s outputs to target specific terms like Kitton.
- The user interface shows live status like agency HQ, site health audits, and the ability to audit a client’s SEO stack via Claude/agent tooling.
- The approach can automate a day’s work for a client and push new features into Harbor with Codex-driven autopilot and discovery improvements.
- 5000+ lines of code reportedly generated by Codex in a single sleep cycle underscores the scale of automation possible at this price point.
- The creator frames Harbor and Codex as a cost-effective, scalable path for agencies to deliver ready-to-launch products without breaking the bank.
Who Is This For?
Essential viewing for developers and SEO agencies exploring budget-friendly automation. If you want to accelerate feature delivery and client reports with minimal cost, this video shows what’s possible with Codex CLI {slash} goal and Harbor integration.
Notable Quotes
"Look at this guys, the new {slash} goal has been running for 15 hours and 30 minutes."
—The core claim of the video: a vague goal can drive extended, productive backend work.
"This is basically OpenAI's version of Ralph Wiggum loops inside Codex, which is super interesting."
—The creator likens the feature to a whimsical but powerful loop mechanism and sets expectations for its novelty.
"Codex with GPT 5.5… running for 15 hours on a vague task for my app is definitely very, very interesting to me."
—Highlighting perceived value and cost-performance tradeoffs of different model generations.
"This is basically only available inside the CLI. It is not available inside the app, unfortunately, just yet."
—Important limitation clarified for viewers considering future adoption.
"Codex basically just ran for 15 hours for me. Like, you for, for 20 bucks a month, guys."
—Emphasizes the price-to-output narrative the creator is selling.
Questions This Video Answers
- How does the Codex CLI {slash} goal feature actually work under the hood?
- Can Codex CLI run long-running tasks reliably on a budget plan?
- What are the steps to enable goals in Codex CLI on config.toml?
- What can Harbor’s agency automation and Codex integration deliver for a client project?
- Is the Codex CLI feature already exposed in the app or only via CLI?
Full Transcript
Look at this guys, the new {slash} goal has been running for 15 hours and 30 minutes. Now, a couple of things to know about this. Number one, the goal is extremely vague, right? Number two, I don't know if I'll stick with any of this code because it's very, very hard for me to verify over 15 hours of complicated back-end coding. Number three, I'm only paying 20 bucks a month for this, which is just absolutely absurd. You wouldn't get anywhere close to the amount of use that you would here. So, like I literally just hit my work my usage limit just here.
So, what I could do here is I could log out and change to API usage, but I mean, that is just absolutely insane. 15 hours of running for you know, really, really complicated back-end changes, which honestly, I'm going to look through with Claude and I'm going to see what it's done and see if we can, you know, manage to keep a lot of this. So, yeah, if people don't know about this just yet, this is a pretty big thing that is going fairly viral on the internet. Now, apparently, it's like basically OpenAI's version of Ralph Wiggum loops inside Codex, which is super interesting.
Um people are saying this is really, really good, right? So, yeah, just so everyone knows, this is basically only available inside the CLI. It is not available inside the app, unfortunately, just yet. Now, the way that I got this is I just went on Codex and you just go up here to settings, right? And then um configuration, open config.toml. And then all you need to do is just add this line here. Um let me just find it. So, yeah, so it's under features, goals equals true, and then just save it. That's literally all you have to do, right?
And then you have access to this incredibly powerful new feature. Now, obviously, I have run out of usage as I showed you at the beginning of this video. So, it's going to be very, very hard for me to use it here, but once what basically what you do is you write Codex. And what I did was I just gave it an incredibly vague prompt. I think it was something like, you know, help me make agency like ready to release to the public, right? So, {slash} goal, that's all you have to do, and then just give a vague goal that you want for your app, and then literally just leave it, right?
Now, obviously, I would probably recommend having either ultra or whatever, I don't know Pro, I think it's called. You know, the $100 plan, you'd probably have be able to run this 24/7 knowing OpenAI. I only have the $20 plan, and I have to say, if I was on a budget, I would probably be using Codex with GPT 5.5. I don't rate GPT 5.5 as much as I do Opus 4.7, but seeing it run for 15 hours on a vague task for my app is definitely very, very interesting to me. Like that was not possible before.
I think the longest I've ever seen an app run is um probably something like an hour and a half, right? So, yeah, obviously, this is not the best test, I have to say, right? I I didn't give it any tools to check anything. I gave it a very, very vague prompt. I actually forgot that I would I had even, you know, used the command, right? But, its ability to run for that long is is rare in AI, so I have to say. So, what I'm going to do is I'm going to do a little check here with Claude code and just go through everything that was done, and I'm going to start to kind of understand, you know, what the hell it even did for that long.
So, it was working towards a goal, which was to get uh agency, which is I'm basically working on agency inside Harbor, to a point where I can give it to people, I can push it to people, and people can start using it. So, I actually have a dev branch, which I can test everything on, right? I actually don't know when the the last deployment was. It's been a while, I think. Oh, no, it was 1 minute ago. So, this was probably Claude pushing, right? So, I'm going to actually see here, once this finishes building, what's been built and whether or not Codex basically just one-shot 15 hours of work.
Okay, so we have agency trial here. That's really, really good. Um nice. Okay, really nice. That's really, really good to see. So, we have agency HQ here on the side. What I'm going to do is I'm just going to ask Claude to tell me what to check. This is really nice. This is new. This is super, super new. This is much better than what we had before. Wow. So, it was working towards a goal. This is actually quite exciting, guys. This is like, yeah, so it's like Ralph Wiggum loops, which I was not a fan of because it was done by, you know, an amateur, done by a, you know, trillion-dollar company, given to us at 20 bucks a month, starts to sound interesting, guys, I'm not going to lie.
So, this is how I kind of do all my checks, all my sanity checks, I think they're called by Claude. Basically, I'm just going through everything, and I'm just seeing, you know, what is going on here, and um you know, whether all the changes are actually uh good or bad, right? Okay, so it says if I add a keyword here, it should Okay, there's no data. That's probably because search console isn't connected. Okay, so we have this uh dashboard here. This is a um proper account that is should have everything connected. So, let's go to a client, go to keywords, and let's just add Wait, let's remove this first.
So, these are tracked keywords. So, if a client asks you to uh work on a keyword, right? You can start to track how well or how badly the keyword is doing, basically. So, that's now working. Beautiful, look at that. So, Chez Adeline 10.9 is the position. So, this is basically just search console data, obviously, um but still really, really interesting. Okay, so we have new sections here. So, content goals, um this is also been implemented properly. This should go into and also target keywords here. It's very, very nice. So, like if they're trying to rank for Kitton, then you can add the keyword Kitton.
That should be saved. Let's have a look. Yeah, it did save. So, that will be fed to the keyword tools, right? So, if you if if a person's client is trying to specifically rank for a specific keyword, this isn't a Harbor ad, by the way, guys. I'm just verifying Codex has actually worked on these things independently of me. Um then it will basically add those um target keywords into the discovery. Same with content goals, if they're trying to help me rank for Kitton, which is literally all this client cares about, right? Then it starts to, you know, shape content around Kitton.
Same with these, these should all be fed into the writer, which is really, really nice. We've got personas now, style guide, which is really, really important for clients. Yeah, so if we run a discovery now for this client, so this should work. It should be very, very Kitton specific. Okay, so you can see here, a lot of these are Kitton related. So, Kitton, Kitton, Kitton. So, that that's now working. Super, super nice. Um very, very useful as a feature cuz obviously, clients will come to you, they'll say, "I want to rank for this keyword." You don't want to just rank for random keywords.
So, once you've got that information from the clients, you can now put that into Harbor. Codex did that completely independently again with a very, very vague instruction of just helping me get agency ready for launch. I'm not going to go through the writer right now, but I'm going to guess that that works. Uh really, really nice. So, billing uh overview, if I go back on agency HQ. Okay, so I'm not actually on a trial here. That's because I'm an or I'm already in an agency account, so it's kind of hard to do that, but I'll I'll test that another time.
And then we're getting more into the Codex changes here. This is 9,308 lines of code that were done by um Codex while I was sleeping. So, agency autopilot, look for a button from triggering autopilot runs. Autopilot settings, okay. Very, very interesting. I'll need to test all of this in detail, but uh yeah, this should, if I'm not mistaken, be uh automation. Okay, so this looks like it works with Copilot in order to help you um automate, you know, a day's work for a client. So, let's see if this is working properly. So, show me my agency clients.
Let's see what happens here. This should render a card inside the chat. Okay, nice. What's my work day today? Like, what should I work on, basically? Let's see what this comes out with. So, this uses the Claude agent SDK, in case people are curious. Really, really interesting. Look at this. So, Codex basically just ran for 15 hours for me. Like, you for for 20 bucks a month, guys. Do you know how insane that is? Do you know how absolutely insane that is? That's just crazy. Let's do audit Harbor SEO .ai. Get active site, list sites, found this.
Okay. Site health audit's now running the site health audit. Wow, really, really cool. Really, really cool. Okay, so it's still running, so we'll leave that for now. What did we deliver to client X? Okay. Create client, delete client, agency. Okay. Portal sanity. I can't really test the portal right now. Um Wow, I mean, seriously, just guys, if you've got if you're on a budget, forget Claude code, all right? Go and get Codex. Codex CLI with {slash} goal working on existing projects to bring new features. I wouldn't be surprised if it could one shot any new feature.
Now, if you want a new feature, be specific, but if you want to work towards a goal, which I think is more interesting, like, you know, help me get this ready for production or whatever, in my opinion, that's where this is really, really going to shine. Now, I did use extra high for this or X high and I just let it run and I'm super, super impressed with the result. Now, if you do want to go check out Harbor, guys, I'm pretty much ready to release agency and then the next big thing that I have coming is Harbor Build, which I, you know, I'm still working on, but basically it will build websites, which will all be tied together, where you can build websites for a client, put them inside Harbor and then run the entire SEO program from here.
Likely, the price of Harbor will be going up eventually, as well. So, uh just to help me pay for API costs, basically. So, go and check out Harbor at its current pricing. If you sign up, guys, I will not change the pricing on you. So, go and check it out now. There is a free trial that you can get. And yeah, like I said, once you sign up for this price, I won't be putting it up. So, there's a link in the description and in the pinned comment. Go and check out harborseo.ai, codex/goal is completely absurd.
That's probably going to be the title for this video, guys. If you are watching until the end of the video, you're an absolute legend. Thank you so much for watching and I'll see you very, very soon with some more content. Peace out.
More from Income stream surfers
Get daily recaps from
Income stream surfers
AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.









