Top AI Product

We track trending AI tools across Product Hunt, Hacker News, GitHub, and more  — then write honest, opinionated takes on the ones that actually matter. No press releases, no sponsored content. Just real picks, published daily.  Subscribe to stay ahead without drowning in hype.


A CTO Called It “God Mode” — Inside GStack, Garry Tan’s 6-Role Claude Code Toolkit

“Your gstack is crazy. This is like god mode. Your eng review discovered a subtle cross-site scripting attack that I don’t even think my team is aware of.”

That’s the text message Y Combinator CEO Garry Tan received from a CTO friend shortly after open-sourcing GStack — a set of six specialized Claude Code skills that simulate an entire engineering leadership team. The friend went on to predict that “over 90% of new repos from today forward will use gstack.”

Bold claim. But when the person behind it runs Y Combinator, the world’s most influential startup accelerator, and the toolkit racked up 155 upvotes on Product Hunt within its first day, it’s worth paying attention.

What GStack Actually Does

GStack is not another AI coding assistant. It’s a skill pack — six slash commands you install into Claude Code, each designed to mimic a specific engineering role. The core idea: planning is not reviewing, reviewing is not shipping, and founder instinct is not the same as engineering rigor.

Here’s the breakdown:

/plan-ceo-review — Founder/CEO mode. Before you write a single line of code, this skill forces a product-level rethink. Garry Tan describes its purpose as “find the 10-star product inside a boring ticket.” It pushes Claude to challenge assumptions about what should be built, not just how.

/plan-eng-review — Engineering Manager mode. Architecture review, data flow analysis, edge case identification, complete with ASCII diagrams. This is the skill that caught the XSS vulnerability that an entire engineering team missed.

/review — Staff Engineer mode. Focused on production risk identification and bug detection. Think of it as a senior code reviewer who never gets tired and never rubber-stamps a PR.

/ship — Release Engineer mode. Syncs with main, runs the full test suite, pushes, and opens a PR. No manual git dance. One command, and your feature is packaged and ready.

/browse — QA Engineer mode. This one ships a compiled Playwright binary, so Claude can launch a browser, take screenshots, click through flows, and visually verify that what it built actually renders correctly. QA without leaving the terminal.

/retro — Engineering Manager mode (again). Runs weekly retrospectives by analyzing commit history. Useful for tracking what actually shipped versus what was planned.

The Philosophy: Role Separation Over One-Size-Fits-All

Most developers use Claude Code (or any AI coding tool) in a single, undifferentiated mode. You ask it to plan, review, write, test, and ship — all in the same conversation, with the same level of scrutiny for everything.

Garry Tan’s argument is that this is fundamentally wrong. Real engineering organizations don’t work that way. A CEO doesn’t do the same job as a QA engineer. A staff engineer reviewing production code applies different thinking than a release manager packaging a deploy.

GStack enforces this separation. Each skill has its own system prompt, its own priorities, and its own definition of “done.” The /plan-ceo-review skill doesn’t care about implementation details — it cares about whether you’re building the right thing. The /review skill doesn’t care about product strategy — it cares about whether your code will break in production.

This “role-based AI development” pattern mirrors how real engineering organizations already operate. And it’s not unique to GStack — other developers have independently arrived at similar approaches. But GStack is the first high-profile, plug-and-play implementation backed by someone with Garry Tan’s credibility.

How It Compares to Other Claude Code Extensions

The Claude Code skill ecosystem has exploded in 2026. Skills and slash commands now operate as a unified system, with thousands of community-contributed options available. So where does GStack fit?

vs. DIY custom skills: Most developers write their own .claude/skills/ files or slash commands. GStack’s advantage is opinionation — Garry Tan has already done the prompt engineering, tested the workflows, and packaged them for one-line installation. The downside: you’re adopting someone else’s workflow preferences.

vs. awesome-claude-code collections: Curated lists like awesome-claude-code aggregate individual skills from across the community. GStack is different because it’s a coherent system — the six skills are designed to work together as a pipeline, not as isolated utilities.

vs. Cursor/Copilot/Cline workflows: These are different tools entirely, but the comparison matters because developers choosing their AI coding setup need to decide where to invest. GStack is Claude Code-specific. If you’re already committed to Claude Code, GStack slots in neatly. If you’re using Cursor or Copilot, this doesn’t help you directly — though the underlying philosophy of role-separated AI assistance is portable to any tool.

vs. conductor.build and similar agent orchestrators: One Hacker News commenter described using GStack alongside conductor.build, noting “dramatically improved code quality and speed of development.” These tools can complement each other rather than compete.

Installation and Setup

One of GStack’s strongest selling points is frictionless setup. For personal use, it’s a single paste command into your Claude Code terminal. For team adoption, there’s a second command that installs the skills into your repo’s .claude/skills/ directory, so every teammate gets the same setup.

The /browse skill requires a compiled Playwright binary, which is included. If you run into issues, a quick bun install && bun run build in the skills directory resolves it. The binary and node_modules are gitignored, so teammates just run a setup script once.

The entire project is MIT-licensed, meaning you can fork it, modify the prompts, strip out skills you don’t need, or build on top of it.

Who Is Garry Tan, and Why Does That Matter?

For those unfamiliar: Garry Tan is the President and CEO of Y Combinator. He studied Computer Systems Engineering at Stanford, started programming at 14, was an early designer and engineering manager at Palantir (where he designed their logo), co-founded Posterous (acquired by Twitter), and founded Initialized Capital (early investor in Coinbase, Instacart, and Flexport).

This matters because GStack isn’t a side project from a random developer. It’s the actual workflow of someone who has spent decades at the intersection of engineering and product leadership, and who currently advises hundreds of startups per year. When Garry Tan says “this is how I use Claude Code,” it carries weight — both because of his technical chops and because YC founders will inevitably adopt whatever he recommends.

The viral spread pattern confirms this: Garry tweeted about GStack on March 12, it hit Product Hunt on March 13 with 155 upvotes, appeared on Hacker News, and within 48 hours had accumulated around 170 GitHub stars. Not astronomical numbers, but the quality of attention — CTOs, YC founders, senior engineers — matters more than raw volume here.

The Bigger Picture: Are Skill Packs the Future of AI Coding?

GStack points to a trend worth watching. As AI coding tools mature, the differentiation is shifting from the models themselves to the workflows built on top of them. Claude Code’s skill system, Cursor’s rules, Copilot’s instructions — these are all mechanisms for encoding human expertise into AI behavior.

The interesting question is whether opinionated skill packs like GStack become the standard way developers configure their AI tools. Instead of every team reinventing their own prompts, you’d install a “stack” that reflects a particular engineering philosophy — much like how teams choose between different linting configs or CI templates.

If Garry Tan’s CTO friend is right that “over 90% of new repos will use gstack,” we’re looking at the emergence of AI workflow standards. That’s probably an exaggeration, but the direction is clear: the value in AI coding is moving upstream, from generating code to structuring how code gets planned, reviewed, and shipped.

Frequently Asked Questions

Is GStack free?
Yes. GStack is fully open source under the MIT license. There is no paid tier, no usage limits, and no account required. You can use it, modify it, and redistribute it freely.

Does GStack work with AI tools other than Claude Code?
Currently, GStack is designed specifically for Claude Code’s skill system. It won’t work directly with Cursor, GitHub Copilot, or Cline. However, the underlying prompt strategies could be adapted to other tools that support custom system instructions. The SKILL.md format is gaining cross-tool compatibility, so broader support may come in the future.

What are the prerequisites for installing GStack?
You need Claude Code installed and running. For the /browse skill specifically, you’ll need Bun v1.0 or later to compile the Playwright binary. All other skills work out of the box with just the one-line install command.

How does GStack compare to writing your own Claude Code skills?
GStack saves you the prompt engineering work and gives you a tested, coherent system. The trade-off is that you’re adopting Garry Tan’s specific workflow philosophy. Many developers use GStack as a starting point and customize the prompts to fit their own team’s needs — the MIT license makes this straightforward.

Can I use only some of the six skills?
Yes. Each skill is independent. You can use /review without ever touching /plan-ceo-review, or use /ship without /retro. The skills are designed to work as a pipeline but don’t require each other.


You Might Also Like


Discover more from Top AI Product

Subscribe to get the latest posts sent to your email.



Leave a comment