Seven fresh GitHub tools for ChatGPT and Claude users, from a chat-controlled flight tracker to a codebase mapper.
Posted
yesterday
Duration
Format
Listicle
hype
Views
32.8K
479 likes
57 · 43
Big Idea
The argument in one line.
Seven small, single-purpose GitHub tools, not one all-in-one platform, are what actually close the gaps AI workflows still have: unsafe skills, vague copy, one-model blind spots, and undocumented codebases.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
Someone building or shipping projects with ChatGPT and Claude who wants to close specific gaps, like deployment, safety scanning, or copy quality, without switching platforms entirely.
A solo builder or small team evaluating which open-source AI tooling is worth an afternoon to try this month.
Anyone who installs AI agent skills or connects new tools to Claude/ChatGPT and wants a concrete way to check them before running them.
SKIP IF…
You want a single unified platform recommendation rather than seven separate niche tools to evaluate individually.
You're not using ChatGPT or Claude day-to-day; several of these tools assume that starting point.
TL;DR
The full version, fast.
This video runs through seven GitHub tools for ChatGPT and Claude users, each fixing one specific workflow gap. God's Eye View aggregates flights, ships, and satellites into one explorable globe controllable by chat. NVIDIA's SkillSpector scans agent skills for hidden instructions and unsafe access before you install them. Floot turns a chat-built design into a hosted app with authentication and a database in one click. SlopMonster rewrites vague AI copy into specific, checkable claims using named clarity and positioning principles. Claudex Loop has Claude and Codex cross-review a build plan before coding starts. DeepSeek Harness swaps in a cheap model inside a fully customizable tool environment. Archify turns a codebase or plain-English process into an interactive system map.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Introduces the premise: seven best GitHub repos to pair with ChatGPT Astra, teased as solving AI slop, spyware-grade safety, and general workflow upgrades.
00:22 – 02:26
02 · Level 1: God's Eye View
A chat-controllable 3D globe that aggregates live flights, ships, satellites, and public cameras into one explorable, voice-directed view.
02:26 – 03:38
03 · Level 2: NVIDIA SkillSpector
An NVIDIA-built scanner that checks installed AI agent skills for hidden instructions, unsafe access, and data leaks, and gives each a risk score.
03:38 – 06:24
04 · Level 3: Floot
Publishes a chat-built app design with one click, handling authentication, database, and hosting so the backend doesn't become the bottleneck.
06:24 – 09:20
05 · Level 4: SlopMonster
Rewrites vague AI-generated copy into specific, checkable claims by scoring it against named clarity and positioning principles from Krug, Priestley, and Hormozi; also plugs the presenter's own paid course.
09:20 – 10:50
06 · Level 5: Claudex Loop
Has Claude and Codex draft and cross-review a build plan before coding starts, catching blind spots a single model would miss.
10:50 – 12:42
07 · Level 6: DeepSeek Harness
A customizable agent environment (files, terminal, browser) built around the cheap DeepSeek V4.1 model, launched with a single voice command via Glaido.
12:42 – 14:57
08 · Level 7: Archify
Turns a codebase or a plain-English process description into an interactive, exportable system map for onboarding, handoffs, and tracing requests.
14:57 – 15:11
09 · Sign-off
Closes by framing the seven repos as a starting point, not the endpoint, for building an agentic operating system.
Atomic Insights
Lines worth screenshotting.
A skill-scanning tool can give an installed AI agent skill a numeric risk score instead of leaving trust in third-party code as a guess.
The backend, not the interface, is the part of an AI-built app that actually stalls solo builders between prototype and live product.
Generic AI copy fails because it makes claims instead of showing checkable specifics, like naming six nails per shingle instead of saying 'reliable roofing.'
Scoring copy against named principles from Steve Krug, Daniel Priestley, and Alex Hormozi turns 'does this sound AI-written' into a repeatable test instead of a gut call.
Every AI model has blind spots, so having a second model review a build plan before coding starts catches mistakes the first model missed.
Cross-model review is worth the extra tokens and time for maybe 10% of decisions: the ones where being wrong is expensive.
The 'harness' around a model, meaning its file, terminal, and browser access, is a separate layer from the model itself, and either can be swapped independently.
A cheap model paired with a fully customizable harness can outperform a closed one-app-per-model workflow once you hit token or rate limits elsewhere.
Mapping a codebase or process visually makes onboarding, handoffs, and tracing a request through a system faster than a written explanation.
An interactive globe that aggregates flights, ships, satellites, and public cameras into one view removes the need to check five separate trackers.
Takeaway
Seven repos, seven specific workflow gaps
TOOLS TO WATCH
Each tool fixes one narrow failure point in an AI workflow, from unsafe skills to vague copy to blind trust in a single model, rather than trying to be a do-everything platform.
02Level 1: God's Eye View
A single interactive globe can aggregate scattered public data feeds (flights, ships, satellites, cameras) into one place instead of forcing you to check five separate trackers.
Natural-language control of a data tool works because the app exposes its state to a chat interface, not because the underlying data changed.
03Level 2: NVIDIA SkillSpector
Before installing any third-party AI agent skill, scan it for hidden instructions, unsafe access requests, and data leaks rather than trusting the description.
A risk score on an installed skill turns an invisible trust decision into a visible, checkable number.
04Level 3: Floot
The unglamorous part of shipping an AI-built app is the backend, not the interface, which is why a one-click publish layer that handles auth, database, and hosting saves the most real time.
Chaining a chat-built design straight into a hosting and backend service removes the step where most solo builders stall between 'I have a prototype' and 'this is live.'
05Level 4: SlopMonster
Generic AI copy reads as generic because it makes claims (trusted, reliable, seamless) instead of showing verifiable specifics (six nails per shingle, a fixed price before the ladder goes up).
Scoring writing against named principles (Krug's clarity, Priestley's positioning, Hormozi's proof) gives a repeatable check instead of a vague 'does this sound AI-written' gut call.
06Level 5: Claudex Loop
Every model has blind spots, so having a second model draft-review a plan before you build catches mistakes the first model didn't see, at the cost of extra review time and tokens.
Cross-model review is worth reserving for the decisions that are expensive to get wrong, not every routine task.
07Level 6: DeepSeek Harness
The 'harness,' meaning the environment of tools, file access, terminal, and browser around a model, is a separate layer from the model itself, and you can swap either independently.
A cheap, capable model paired with a fully customizable harness is a legitimate alternative to a closed one-app-per-model workflow, especially once you hit rate or token limits elsewhere.
08Level 7: Archify
Turning a plain-English process or an existing codebase into a visual map makes it easier to onboard someone, hand off work, or trace a request through a system than a written explanation would.
Mapping a process first and standardizing it second, not the other way around, avoids locking in a workflow before you know what actually works.
Glossary
Terms worth knowing.
Agent skill
A packaged set of instructions or tools an AI agent like Claude or ChatGPT can load to perform a specialized task, similar to installing a plugin.
Harness
The surrounding environment around an AI model, its file access, terminal, browser, and tool integrations, as distinct from the model itself.
Cross-model review
Having a second AI model check a plan or output produced by a first model before committing to it, to catch blind spots either model alone would miss.
“I invite you to join me in the war against AI slop. I am fighting hard. I am one man.”
quick, funny mission-statement line with a clean cut point→ TikTok hook↗ Tweet quote
08:11
“It's not just a roof, but peace of mind becomes a written scope and a fixed number before anybody climbs a ladder.”
self-contained before/after copy example, no setup needed→ newsletter pull-quote↗ Tweet quote
09:58
“There are two best models in the world right now. Claude Fable 5.1 and ChatGPT Astra.”
definitive, quotable claim about model landscape→ IG reel cold open↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphor
GitHub is the best place to get to the future of AI technologies. And in the last month, we've had so many incredible drops. In this video, I'm going to show you seven of the best GitHub repos that you need to use with chat dbt Astra.
It will do things like destroy AI slop, give you incredible spyware, and even help you level up all parts of your work. So grab that coffee. Let's check it out.
Beginning with what we call God's Eye View. This one is an incredible... interactive system that essentially gives you almost like hero level vision over the planet this one blew up and it's pretty exceptional so let's start with what it is you can see all the details for every flight.
You can see all the satellites, every single ship, and you can effectively go ahead and explore the world. This is what it looks like in action, and we're going to go through this together. I'm going to show you how you actually go ahead and use this.
Now, when we load this up, you've got all the intelligence on the left, life flights, military flights, earthquakes, LPR cameras, everything you can imagine. Probably the coolest thing you can do is chat with it. So check this out.
Hey there, take me to the Pentagon. Okay, let me take you there now. All set.
I've flown you into the Pentagon. Cool. Next thing I want you to do is...
Basically, beam me into the closest flight happening overhead. All right. Let me check where the camera is and pick the closest overhead traffic.
All set. Awesome. Now take me into cockpit mode.
Cockpit mode is entered for the selected aircraft. I also now want to hear the radio. Can you put a local radio station on for me, please?
Sure thing. Let me get a local station going for you now. I've already...
Listeners of WBUR Boston and from Mathworks. So as you can see, guys, you can see these plane views. You can zoom out.
There's like a million different things you can do with this. It's super cool. I can see every flight that's going on there.
And you basically can go ahead and connect this to many different API keys. It gives you loads of different views. You can do TD maps.
You can do anything. I mean, even now, I can go ahead and see essentially what's going on all over the globe. It's such a cool project built with a lot of love and attention.
Now, what this one enables you to do basically is to see the whole picture. So aircraft, ships, satellites together. you can explore any place to see what's happening in real time.
It's really good for visual storytelling. So live demos, exploring places, basically certified tracking or kind of like guaranteed coverage as well. And this essentially aggregates loads of independent public data into one view of the world where you can go anywhere you want to.
And it's got some really cool features. Now you can go ahead and install this by clicking on the GitHub link. It's super duper easy.
Just head over to this website, give it over to ChatGPT Astra and it'll go ahead and do everything. This has gotten a lot of stars and it's blowing up pretty quickly. exception now there's nothing though compared to the other levels we've got unless you're over and have a look at skill specter this is by nvidia this is a must -have one and it's just a really good safety skill that just sits over everything skill specter itself i'll show you right now we'll open this bad boy up made by nvidia 17 000 stars as of at the moment But it's really powerful.
So what does it actually do? Well, effectively, it scans your agent skills for any hidden instructions, any unsafe access, and any data leaks. So it's to kind of keep you protected with everything you're doing.
And it'll give you a risk, a risk factor, basically. It can catch anything that's hidden. So it can spot suspicious instructions and data flows, check the access that a skill actually asks for.
And essentially, whenever you install anything, just use Skill Spectre to go ahead and physically do that for you. And to do that, all you're going to do is... come over literally come over grab the url like so click on copy and then you can say hey there i want to go through every skull i have and run it through this skill specter and once you've done that just come down and just enter in that repo and then it will use this nvidia skill to go ahead and check that everything is okay and as you can see even right now it's using the school specter to look at my inventory of install skills and run that across it to make sure that we're safe and we're not compromised.
Now, the third one I want to talk about isn't actually a repo, but it's something I've been having a lot of fun with that I had to include it. It's called Flute. And what's really interesting about this is when you're using ChatGPT or Claude, it essentially takes anything you're building and turns it into an app.
any complexity. And I wasn't aware of this concept until recently, but I think it's a really interesting one. So effectively what it does, and I'll put the website and show you this.
So you're building in cursor, chat, GPT, Claude. Effectively what you can do with one click, you can take the designs that you published and created inside chat, GPT and Claude. and publish them.
And they look after everything from the authentication and the forms and the databases, saving you lots of effort. So you publish with one click, it manages all the backend, the data, and the logins, and it can even sort out emails, notifications, and it's built to be found. And obviously you can integrate all the tools, which is cool.
And what this effectively means is that all you do is build in ChatGPT, you build in Cloud, you just connect Flute, and then effectively you can just manage all the publishing and the displays, which is really cool. Key thing being we go from an idea to an app. So we build it in ChatGPT or Claude.
And it isn't just a going live because ChatGPT sites does that. It's mainly the database, the authentication, the logins, all that sort of thing. And the backend is notoriously the bit that we don't like.
So this really helps us do it. And you can get started for free, which is freaking cool. So let me show you an example of this to best understand it.
For example, I went over to ChatGPT and I built a homepage. And I was like, look, I want you to go ahead and build me an example form, okay? Like let's say we built something we like.
Like what's the company called? What things could be better? What would you love to automate?
Loads of different questions. questions what investment feels right maybe we've got a really cool business idea and we want to get this live but you know i don't want to have to sign up to different forums i just want to get one place it's really cool i can do all of this right it's got loads of interesting questions here and once that's done all i'm going to do is hey i want you to connect to flute and i want you to publish this as an app.
And then once you've done that, it will then go through the authentication process. It's quite like using ChatGPT for this. We just go ahead and click on approve, come down and click on finish connecting.
And right now it's finishing the connection and then go ahead and create that for us. Then essentially ChatGPT will just work with Flute to build everything over there. As you can see now, it's moved over the entire design.
It's doing some authentication, making sure it creates some accounts. And that is the whole thing that you need to do. And then guys, you can come back over to Flute.
You can see all the plans and usage here. For example, on this computer, I'm on one of the free accounts. accounts and you can see exactly what happens and the usage but if i come into projects for example i can click on this one here you can build directly and when i open this up you should be able to see the exact same thing that i built in claude but now it's something that can go ahead and be used anywhere we want to here we go and here's all of the questions that are ready to rock and roll and then literally guys this will go ahead and publish and you have everything connected from basically all the forms or the databases and you can share this and start growing whatever projects you're working on now as you can see there we have the full app with the form and everything connected in the background but this takes is very nice onto the next level, which is going to be the fourth repo.
And this one is called the slop monster. Now I invite you to join me in the war against AI slop. I am fighting hard.
I am one man. We can do this together. Now, the whole idea of Slop Monster, essentially, is it takes any text and then turns it into something that doesn't sound like it's written by AI.
And I actually built this one myself because it was probably the most frequent thing that I was asking AI to do in many of the websites that I built. And essentially, I saw that there were so many different anti -AI slop... kind of repos out there.
And I thought, what if I actually got all of them and took the best bits from every one of them and then kind of fix the thing that I think a lot of them miss. So essentially what this will do is go ahead and actually score the work that you've got. It tests that across a few different factors.
It looks at AI vocabulary, looks at AI constructions, punctuation cadence, rule of three rhythm. And then the bit that a lot of them miss is actually sales and marketing. So if you're building copy for your website, for example, for example, instead of saying the area's most trusted roofing experts, roofing and...
only roofing since 2001. It looks at what the trade actually does, what the customer fears, what the competition won't say, and just basically runs us through in a million different directions. And it's based on work by Steve Krug.
We wrote a summer book called Don't Make Me Think. The idea is you shouldn't have to think about what you're reading. We've got Daniel Priestley.
We've got Alex Ramos's offers all in here. All you literally need to do this, guys, is come over, grab the link, and just tell ChatGPT, hey, run this text or this website through the slop monster. And the results have literally blown my mind.
I'm a big fan of it. this. And it's just going to rewrite any vague AI copy into clear, specific writing.
For example, it takes stuff like this, trusted, reliable, and built to last. Like AI loves this rule of three. And instead you get something like six nails per shingle.
Every shingle. It's specific. And each one of these has a different thing underneath it, a different principle as to why it goes ahead and uses it.
And you can see what loads here. Not just a roof, but peace of mind becomes a written scope and a fixed number before anybody climbs a ladder. We leverage industry -leading materials to deliver unparalleled protection.
That becomes resource materials from manufacturers who test for wind, hail, and sun. You can just see the difference here, right? It's really great.
After the demo, take the build apart, change it until it solves a problem of your own. It's really great like that. It makes every line count and essentially just takes you from generic.
It works on emails, works on websites. You can throw it in anywhere. And of course, it is completely free to use.
And if this all sounds like I'm speaking Spanish, I'm going to put a link down below for my full course that I'm dropping this week, essentially on mastering ChatGPT Asterisk. It covers everything from AI foundations, productivity in life, business and sales, marketing, automation, agents. You also get full access to all of my memory systems I've been building for months.
Content I never share. on YouTube. You get full access to things like the business operating system, as well as my design operating systems, all down below.
I'll put a link so you can grab that and take everything to the next level. Now, speaking of the next level, we need to go ahead and talk about level six. Now, level six is going to come after level five, and level five is the Claudex loop.
This is a handy skill. One of the things that you always want to think about when you're building with Claude and ChatGPT is the fact that they are awesome. But actually, when you use both of them together at the same time, they become even more powerful.
So I always, for example, when I'm even building on my memory systems, right, or I'm connecting anything, I will always have Claude or ChatGPT cross -verify with the other best models. There are two best models in the world right now. Claude Fable 5 .1 and ChatGPT Astra.
Sometimes they catch things that the other individual misses at the same time, which is why this is so handy. Now, effectively what works is Cloud and Codex can review a plan before you build it. One model drafts a plan and the other one looks for weak spots.
So they basically cross verify one another. And they just work beautifully together in that sense. Now, they will catch mistakes that the other model missed.
So why not use both at the same time? I would say that I would use this loop. probably in about 10 % of the time I'm using AI.
It's for big stuff. It's where the decision matters and you're looking for diversity because every model has its blind spots. It doesn't matter how good it is.
Use it for complex builds. And again, all you're doing is just tagging it in and it just connects with what we call the command line interface. Nerd speak for thing happening on your computer.
It just reaches over and uses that. A very powerful system to go ahead and use Claude. And again, I'll put all the repos down below.
So you go ahead and grab it. Now let's have a chat about. level six, which is the deep seek harness.
Now the deep seek harness is interesting. Let me show you what I mean. Hey there, could you go ahead and open the deep seek harness and just ask it, what is it?
Explain itself to me. Now I'm just using Gladiator for this. It's something that I set up basically to, you know, one prompt, open up deep seek whenever I want to.
Now the deep seek harness itself is, if you imagine that you have the race car driver and you have the Formula One car. Now the Formula One car is like, codex the chat gpt app it's like the claude app and deep seek themselves also now have an app and an environment a harness it's all the stuff around the model it's the car basically deep seek now have their own version that's called the deep seek harness now you have the harness and then you have the model.
So we can use any model in any harness that we want to. It's very, very cool like that. What it makes this particularly unique though, is effectively this deep seek harness can access files, terminal tools, browser, and it is fully customizable.
So you don't have to just do what ChatGPT says and what Claude says. You can use this for anything. Me personally, I like having it because essentially the deep seek V4 .1 model is so cheap and I can just launch it up with literally one command to do that.
I go ahead and I use Glido. It's the speech tech startup. that I found that it's really, really cool.
But effectively, you have these commands where you can open things up. Of course, you can use any speech to text to chat, but we built this because we thought it was really freaking cool. I just find that very handy and helpful to do.
Let's go ahead and build those things. Now, the thing about it is you can run coding tasks using your model files and tools. It's really simple.
You can shape your own workflow. And remember, performance is a combination of the driver and the car. And the idea with this is it just kind of lets you customize the car.
as well as choose what driver you want in it. So that's some of the really interesting things that you can do with it. And it's really good to have a little play around with.
You can just fire off little questions. And if you find yourself hitting token limits, it can be a great way to play around as well to take you to another level. And then that takes us nicely onto level seven, which is Archify, which is...
A really interesting one actually, and effectively what it does is it takes anything from plain English into architecture in seconds. So essentially it will map out any kind of complicated process for you. So let me show you an example of what that would look like.
So I've just asked it here, could you use Archify to map out the optimum sales system according to Alex Ramozy, please? And then just like that, we've got an example of this. Okay, so let's expand this a wee bit.
I say we, but that's what they talk like in Scotland. I'm hearing on good authority. Well, it moves your sales system, mark an offer, generate demand, earn interest.
So essentially you can take any idea that you like and see it and interact with it graphically, which I think is cool. So this may not win the world's best award for like most like visually perfect, but I see visually perfect. I mean like.
designy, edgy, cool thing. But it does really do the job. And honestly, it's a light year ahead of anything you're going to find in any corporate office.
We have different themes you can play. Dark mode, I think, looks better to me. And again, look at this.
Look at the level of things you can do from here. Like author reach downstream, upstream, what happens when you're in interest. You know, you imagine just playing around with this and thinking about, okay, look, what is our sales cycle?
What actually is the way that we want to do business here? Having this really makes an idea. Look, fit and intent.
What is this? And you can present to them. You have classic mode, signal flow.
You can play around so many different things. It's just such a cool way to get ideas out and play around with them. Of course, we can export it, share it with people.
I mean, the amount of times I've spent, guys, when I was in corporate, but if you're on here, I'm Jack. I built and sold my last tech startup with a gazillion customers. But before that, I was working in corporate in a strategy job.
And the amount of times I'd have to map out like a process, thinking about like the future of the business, the employees and stuff was crazy. Time that now, honestly, this can just save for straight away. And obviously you can build.
out and add many different elements to this and some of these cases that you're going to find very helpful are you know seeing components and their connections tracing a request through the system we once did something that had like It must be like 500 different steps. It was crazy.
A lot of coffee required. When a code base is new, so onboarding, handoff, and reviews, imagine having a full meeting explaining process. Then you just use Archify to kind of build a bit of a process map that everybody can follow.
Remember, you want to find excellence first, and then you can proceduralize after. And again, it is literally free to use. Understanding what's trending on GitHub is one thing, but if you really want to get ahead, you need to understand how to use an agentic operating system that is going to add value in ways that you probably aren't even familiar with.
about right now and we're going to do that together in this video right here
The Hook
The bait, then the rug-pull.
Seven fresh GitHub repos, gamified here as 'levels,' promise to plug the gaps ChatGPT and Claude still leave open: installing agent skills blind, publishing an app's backend, writing copy that reads as AI, and trusting a single model's judgment.
Frameworks
Named ideas worth stealing.
07:36list
SlopMonster's named working principles
Steve Krug (Don't Make Me Think) — open on the problem and the insight, never the product
Daniel Priestley — named pain, checkable specificity
Alex Hormozi — proof you actually own
SlopMonster scores AI-written copy against three named marketing principles instead of a generic 'does this sound robotic' check, so the rewrite has a defensible standard behind it.
Steal forany AI-copy QA pass before publishing website or landing page text
CTA Breakdown
How they asked for the click.
VERBAL ASK
08:41product
“I'm going to put a link down below for my full course that I'm dropping this week, essentially on mastering ChatGPT Astra.”
Dropped mid-video, right after the SlopMonster demo, framed as 'if this all sounds like I'm speaking Spanish' — an easy off-ramp for viewers who got lost, into a paid course covering AI foundations through business/sales/marketing/automation.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
OpenAI's voice mode became a persistent agent that keeps working while it talks and follows you across every window — a level-by-level walkthrough of what it can actually do.
A walkthrough of a free GitHub tool that reroutes Claude Code through free-tier API keys from NVIDIA and other providers once the paid usage limit runs out.
A fourteen-minute walkthrough of color grading, animated B-roll and film-style transitions, built entirely inside Claude and a personal design dashboard, no editing software touched.
A reference-site-to-Claude workflow covering site mapping, AI image generation, mobile optimization, de-AI'd copy, and SEO: the seven levels between a pretty AI site and one that actually converts.
00:08
00:16
00:33
00:46
01:01
01:12
01:24
01:35
01:46
01:52
02:09
02:17
02:40
02:56
03:08
03:14
03:32
04:00
04:09
04:27
04:35
04:48
04:57
05:13
05:26
05:39
05:52
06:04
06:17
06:46
06:55
07:08
07:20
07:33
07:45
07:57
08:10
08:23
08:50
09:01
09:13
09:36
09:45
09:59
10:10
10:21
10:33
10:44
11:05
11:18
11:34
11:40
11:51
12:04
12:15
12:25
12:36
12:58
13:10
13:21
13:32
13:38
13:55
14:06
14:17
14:26
14:40
14:51
15:04
15:08
13:19
11:13
11:39
14:00
22:56
20:10