102 lines
13 KiB
Markdown
102 lines
13 KiB
Markdown
# The toolkit from Y Combinator CEO that Will Makes Claude Code Amazing
|
|
|
|
**URL:** https://youtu.be/kLq5p43huYQ
|
|
**Data salvare:** 2026-03-18
|
|
**Durată:** 8:29
|
|
**Tags:** @work @project @scout
|
|
|
|
## TL;DR
|
|
|
|
Gary Tan (CEO Y Combinator) a creat **GStack** - toolkit pentru Claude Code cu 9 workflow-uri specializate. Include: plan CEO review, plan engineering review, headless browsing (Playwright), integrare Grepile, QA diff-aware. Demo impresionant: feature screenshot tweet cu dark/light mode, aspect ratios, custom backgrounds - totul generat autonom. Discuție controversată: "Markdown is the new code" - modelele noi obey Markdown instructions mai bine, dar GStack are și TypeScript/implementare.
|
|
|
|
## Puncte cheie
|
|
|
|
1. **GStack capabilities:**
|
|
- 9 specialized workflows (plan CEO review, plan engineering review, ship, QA, review)
|
|
- Headless browsing cu Playwright
|
|
- Integrare Grepile (code review automat)
|
|
- Diff-aware QA (testează doar ce s-a schimbat)
|
|
- Auto PR creation
|
|
|
|
2. **Demo workflow complet:**
|
|
- Plan mode → plan CEO review → scope expansion
|
|
- Architectural decisions (mega plan cu diagrame)
|
|
- Implementation → review/command (bug detection)
|
|
- Ship/command (sync, tests, PR automat)
|
|
- QA/command (test local cu screenshots)
|
|
|
|
3. **Feature screenshot tweet:**
|
|
- Light/dark mode
|
|
- Multiple aspect ratios (9x6, 16x9, 1x1)
|
|
- Custom backgrounds (LinkedIn, Twitter, Blog, gradients)
|
|
- Gradient angle customization
|
|
- Hide/show images, actions
|
|
|
|
4. **"Markdown is the new code" clarificare:**
|
|
- NU înseamnă că codul e irelevant
|
|
- Modelele noi (Opus) obey Markdown instructions 90-95%
|
|
- Detailed + well-structured markdown → good software
|
|
- GStack = Markdown skills + TypeScript implementation + templates
|
|
|
|
5. **Arhitectură GStack:**
|
|
- Fiecare skill = director propriu
|
|
- Template files + skill files
|
|
- Scripts/gen skill = TypeScript read templates, replace placeholders
|
|
- Source directories cu implementări (ex: browse skill = browser management)
|
|
- Test directories pentru fiecare skill
|
|
|
|
6. **Advanced features:**
|
|
- End-to-end observability
|
|
- Incremental eval saves
|
|
- Reviews în format to-do
|
|
- Screenshot element & region clipping
|
|
- Built cu Conductor în minte
|
|
|
|
7. **Decizie autor:**
|
|
- Va încerca GStack 30 zile
|
|
- Va șterge superpowers plugin
|
|
- GStack = main tool pentru features & bug fixes
|
|
|
|
## Quote-uri notabile
|
|
|
|
> "His secret to crushing almost a 100 PRs in 7 days"
|
|
|
|
> "This mode rethinks the problem from the perspective of a founder/CEO and tries to think of the best possible version of what we're trying to build"
|
|
|
|
> "I can kind of see where he's coming from. I don't think he's saying someone with a computer science degree has wasted their time purely because you can write Markdown"
|
|
|
|
> "If you have a detailed enough and well structured markdown file, the model can create a good piece of software based on those instructions"
|
|
|
|
> "This is an insanely detailed plan similar to something I'd get from superpowers"
|
|
|
|
## Idei pentru Marius
|
|
|
|
**✅ RECOMAND să explorăm GStack:**
|
|
- **Timp:** 1-2h instalare + testare
|
|
- **Beneficiu:** Workflow-uri structurate pentru PRD → implementare (similar cu Ralph dar cu mai mult control)
|
|
- **Integrare:** Se potrivește cu fluxul curent (Opus planning → Sonnet implementare)
|
|
- **Impact:** Ar putea înlocui/completa Ralph pentru proiecte mai complexe
|
|
|
|
**De ce recomand:**
|
|
- Plan CEO review = PRD automat cu scope expansion
|
|
- QA diff-aware = testează automat doar ce s-a schimbat
|
|
- Grepile integration = code review automat (catch bugs înainte de production)
|
|
- Se integrează în Claude Code (deja îl folosești)
|
|
- Gary Tan proven track record (100 PRs în 7 zile)
|
|
|
|
**Pași sugerați:**
|
|
1. Instalăm GStack (bun + skills)
|
|
2. Testăm cu un feature mic din roa2web
|
|
3. Comparăm cu Ralph workflow
|
|
4. Decidem dacă îl păstrăm sau combinăm
|
|
|
|
**Trade-off:**
|
|
- Pro: Workflow mai structured, QA automat, Grepile integration
|
|
- Con: Learning curve, dependență de tooling, posibil overhead pentru taskuri simple
|
|
|
|
---
|
|
|
|
## Transcript complet
|
|
|
|
CEO of Y Combinator has built his own toolkit for Claude's code called GStack. His secret to crushing almost a 100 PRs in 7 days which includes nine specialized workflows, a headless browsing mode using Playright, Grapile integration, a diffaware QA, and much much more. But Gary's recent tweet about the future of code has got a lot of developers really annoyed. So what does that mean for the future of GStack? Hit subscribe and let's find out. Gary Tan has been the CEO of Y Combinator since 2023 and before that he co-founded a venture capitalist firm in 2011. So he has loads of experience when it comes to going through pitches and finding out what makes a new piece of tech unique. And he's put all of that knowledge into his own Claude Code toolkit, which you can tell by looking at the names he's given to a lot of his workflows. In fact, let's give GStack a spin. So, for GStack to work, you'll need to have Claude code installed as well as bun. But once you've properly installed it on Claude Code by prompting with this exact text or from just downloading the skills, you should have this information added to your Claude MD file. Mine was empty. That's why this is the only thing here. But if you have some text, then this will be added to it. It also puts all the relevant skills into the skills directory if you want to share it with your teammates and then installs Playright with the appropriate browser. Now, I'm going to use GStack to add a feature to this React Beat application to give the user the ability to download an image of a tweet with a specific URL. Now, you may have seen me add this feature in a previous video. I'll have a link to it in the description if you want to see what the results were, but we'll see if GStack can do better than that. So, first I'll need to start in plan mode, then use the plan CEO review skill and give GStack some information about the feature. So I'm going to say add a feature that takes a screenshot of a tweet from URL provided by the user. I also want the user to customize and downloads the image and I want Claude to honor the existing layout and styles. So after I hit enter, GStack first checks if there are any updates to that skill and then checks the git log before proceeding. Now, this mode rethinks the problem from the perspective of a founder/CEO and tries to think of the best possible version of what we're trying to build and challenges assumptions about the scope and value. So, once it's done that, it allows us to choose how much we want to challenge the original scope. And here, I'm going to go with the scope expansion because it has the most amount of features. Then, it lets us choose a critical architectural decision. I'm going to go with the recommended since it's the easiest. And then, it asks a few more questions, which again, I'm going to go with the recommended approach. And now that it's finished, it's come up with a mega plan showing the scope mode selected and everything it's going to do that's in that scope. And it's also written some things that are not in scope for this feature. And then down here we have the implementation plan which has an architecture diagram, key decisions and different steps. This is an insanely detailed plan similar to something I'd get from superpowers if I went through the same route. Note, there's also a plan engineering review skill in GStack which turns Claude into an engineering manager or tech lead to come up with architectural diagrams, lock in the tech stack, define edge cases, and so on. But it looks like the plan CEO review skill has gone ahead and done some of that already. So, we're jumping to the implementation. And now that it's done, we can run the review/comand to review missing edge cases, find bugs that would have passed CI, and basically catch any issues before they hit production. So again, that checks for new updates inside the script, checks the diff, and now it's checking the completeness of the task before giving us a summary saying that no issues have been found. And now we can run the ship/comand, which syncs with the main branch, runs tests, and resolves any gile reviews if they exist. And here we can see it's gone ahead and created a pull request without me even telling it to. And then at this stage, we can run the QA/ command, which will test only the changes we've made based on the diff. And here we can see it started the server locally and it's going through the website to test the features that have just been implemented using screenshots and much more. It's found some 500 errors from screenshots and has found a bug with JSON pass which it looks like it's fixed. Here we go. It's verified and pushed the fix and now it's written a final report with the issues that it solved. This is very cool. Okay, so now it's done. Let's go ahead and try the feature. So now we have a screenshot page. Let's grab a tweet from Talent. So this one. And I'll paste that in here. It's not the most exciting tweet, but it's just to test if this works. And wow. Okay, this is super impressive. We have the tweet here. We can pick between light and it's capturing again. Oh wow. Okay, so we've got light and dark mode. We'll see if it's catch that. And it has. Very cool. I can hide the actions. And here we go. So I can show and hide the images. And I can change the background. This is very cool. So, we've got LinkedIn, we've got Twitter, blog, gradient purple, and we can even customize it or change the angle of the gradient. Wow, this is super fully fledged. And we can change the aspect ratio. So, we've got 9x6, 16x 9, 1x 1, and so on. Let's now actually download the image. And here we go. If I now click on this, you've seen all my tabs. We have the image here. I'm going to open in preview. And this is it. This is the image I just took with the feature I just built with GStack, which is insanely impressive. But there's more that we can do because if we go back to the PR, we can see Grapile has a summary. So, it's found some resource exhaustion from the server race condition, no cache expiry, and so on. And instead of me asking Claude to look at the issues and solve them, we're just going to run the review slash command. It's found all the comments. It's given me some options down here on how to fix them, which I'll go through. And now it's fixed all the issues. Well, apart from one false positive and has pushed the code. Greel seems to be happy. As someone who regularly uses superpowers, I can already see the benefit of GStack, even though some aspects of it are quite complex. But what about Gary's comment on Twitter saying that Markdown is the new code? Well, I can kind of see where he's coming from. I don't think he's saying someone with a computer science degree has wasted their time purely because you can write Markdown and it will write the code. I think it's more to do with the instructions because newer models are getting better at obeying markdown instructions. Before there was a time when I would need to have a clawed code hook just to make sure it uses bun to install instead of using npm. But now I can put it in the clawed MD and with a good model like Opus, it tends to obey 90 to 95% of the time. So I think what he's trying to say is that if you have a detailed enough and well ststructured markdown file, the model can create a good piece of software based on those instructions. But this isn't to say that GStack is just a bunch of markdown instructions. Each skill has its own directory, even the ability to upgrade GStack. And if we focus on the browse skill, we can see there's a template file and the actual skill file. And this isn't anything to do with Go templates regardless of what the GitHub page says. The way this works is if we go to scripts and then we go to gen skill. The TypeScript file will read the temple files and replace any placeholders inside them with actual markdown. But I'm not going to focus on each skill individually because they're quite detailed. But what I will focus on is that the browse skill has more than just a skill and D file because we have a test directory here and we also have the source directory which contains the actual implementation for browser management and so on. So we can already see that the commands here are fairly involved. But if we take a look at the change log, this shows some really interesting features like end toend observability, incremental eval saves and so on which is used for developing the app. It shares reviews in a to-do format. It supports screenshot element and region clipping, not to mention all of the integrations it has with Greile and the fact that it was built with conductor in mind. So, the million-doll question is, will I personally use GStack? And I would say actually, I'm going to try it out for 30 days. So, I'm going to delete the superpowers plug-in and make GStack my main claw code tool for preparing features and fixing bugs and see how it goes. Who knows, I might just clone the next Versel open- source tool and start some more beef on Twitter.
|