GStack Review: Garry Tan's 23-Tool Claude Code Setup (Is It Worth It?)
A. Frans
Published April 21, 2026
Table of Contents
Garry Tan runs Y Combinator, has backed hundreds of startups, and apparently also maintains a Claude Code skill setup with 77,000 GitHub stars. The man is either very productive or very good at delegating, possibly both.
GStack is his published Claude Code configuration: 23 opinionated tools organized around different "agent roles" — CEO, Designer, Engineering Manager, Release Manager, Doc Engineer, and QA. I used it for three weeks on a real project to figure out whether it's useful or just popular because of who made it.
What Is GStack?
GStack installs a suite of skills that turn Claude Code into something closer to a product team than a code assistant. Instead of just generating code on request, you can invoke specific agent personas with different priorities and mental models:
- CEO Agent: Strategic prioritization, roadmap planning, feature trade-off analysis
- Designer Agent: UI/UX recommendations, component design, user flow review
- Engineering Manager: Sprint planning, task decomposition, code review standards
- Release Manager: Deployment checklists, changelog generation, version bump coordination
- Doc Engineer: README generation, API docs, inline comment suggestions
- QA Agent: Test coverage analysis, edge case identification, regression checklists
The idea is that different stages of a project need different mental models. Specialization creates focus, and focus creates better output.
Install command:
claude skill add garrytan/gstack
GitHub stats (April 2026): 77,000+ stars, 150,000+ installs. Growing at roughly 3,900 new stars per week.
Quick Overview
| Feature | GStack |
|---|---|
| Author | Garry Tan (YC President) |
| Trust tier | Verified (community) |
| Skills included | 23 tools across 6 agent roles |
| Pricing | Free (MIT license) |
| Compatible agents | Claude Code |
| Security status | Community-reviewed |
| Last updated | April 19, 2026 |
My Three-Week Testing Experience
I used GStack on a SaaS dashboard project: Next.js frontend, Postgres backend, single developer (me). Here's what each agent role actually delivered.
The CEO Agent: I invoked this at project kickoff to help prioritize which features to build first. It asked clarifying questions about my target user persona and revenue goals, then helped me sort a backlog by impact-versus-effort. useful for the "am I working on the right thing?" anxiety that plagues solo builders.
It also pushed back when I tried to add scope. I proposed building a CSV export feature before fixing the onboarding flow. The CEO Agent told me that was the wrong call. It was right.
The Designer Agent: I used this to review a component layout before writing the code. It pointed out that my proposed navigation pattern would be confusing on mobile and suggested a bottom navigation alternative. The feedback was specific and actionable, more useful than asking vanilla Claude Code to "review my design."
The Engineering Manager: Sprint planning mode surprised me. I gave it 14 tasks and a 5-day deadline, and it produced a day-by-day breakdown with risk flags on the items most likely to cause delays. Having the plan externalized helped me stick to a sequence rather than context-switching randomly.
The Release Manager: This is where GStack earns its install for most developers. I got: a deployment checklist customized to my stack, a changelog draft generated from my git diff, and release notes that I cleaned up in about 10 minutes and published directly. Saved 30-45 minutes on a release I'd otherwise have rushed through.
The Doc Engineer: Generated a solid README from my codebase in one shot. The structure was accurate, the code examples were correct, and the Getting Started section needed minor editing. About 70% of the content was publication-ready.
The QA Agent: Most surprising value. It gave me a test coverage analysis and suggested edge cases I hadn't considered, including a potential null reference issue in my authentication flow that would have caused real production bugs. Worth installing for this alone.
The Core Insight Behind GStack
Most skill packs just add more commands. GStack's insight is that mental models matter more than tools.
By having Claude adopt different agent personas with different priorities, you get better-calibrated responses for each phase of development. The CEO Agent doesn't worry about whether the code compiles. The QA Agent doesn't care about the strategic roadmap. That specialization is the point.
This mirrors how good product teams work: a PM thinks about features differently from a QA engineer thinking about edge cases. GStack tries to replicate that cognitive diversity within a single AI assistant.
Security Considerations
GStack is not an official Anthropic skill, it's a community skill from Garry Tan's personal GitHub. MIT licensed, public source code. Security status: "community-reviewed," meaning other developers have examined it, but it hasn't gone through Anthropic's formal audit process.
Before installing, check: 1. Read the SKILL.md files in the repository 2. Verify the install command points to garrytan/gstack, not a fork or clone 3. Review whether any tools make external API calls or access files beyond normal Claude Code scope
From my review: GStack appears to be context-injection only. It doesn't make external API calls or access your filesystem beyond what Claude Code normally does. Installation risk: low for developers comfortable reading open-source code.
Limitations Worth Knowing
It's still one model: The agent personas don't create separate AI instances. It's Claude with different system context. The persona guides framing, not capability. Don't expect magic.
Overlap with SP skills: If you already use Anthropic's official SP skills (sp-writing-plans, sp-executing-plans, sp-finishing-branch), some of GStack's functionality overlaps. Both are worth having, but be aware of the redundancy in project planning workflows.
Requires active invocation: You have to remember to invoke the right agent for the right phase. If you write a normal Claude Code prompt, you're not getting GStack's value. It rewards intentional use.
Solo builder bias: GStack is clearly designed for solo founders and small teams handling multiple roles. If you're on a real team with dedicated PMs and QAs, the agent personas add less value.
GStack vs. Anthropic SP Skills
| Capability | GStack | SP Skills |
|---|---|---|
| Source | Garry Tan (community) | Anthropic (official) |
| Trust tier | Verified | Official/Audited |
| Focus | Product team simulation | Development workflow |
| Best for | Solo founders, indie hackers | All developers |
| Stars | 77K | 120K+ (parent repo) |
| Security | Community-reviewed | Audited |
Who Should Install This?
Strong yes:
- Solo founders or indie hackers managing all product decisions
- Developers who frequently second-guess prioritization or architecture choices
- Anyone building a product with multiple concerns (design, code, ops) managed alone
- People who release software regularly and want a repeatable deploy checklist
Maybe:
- Small teams where some roles are already filled by humans
- Developers primarily building internal tools with low product management overhead
Skip if:
- You're on a larger team with dedicated PMs, designers, and QA engineers
- You primarily do contract work where specs arrive fully formed
- You want maximum simplicity in your Claude Code setup
My Verdict
GStack is one of the better community skills I've tested. The agent persona idea is clever, and it addresses a real problem: solo builders need to think like a product team, not just a coder.
The 77,000 stars aren't purely a function of Garry Tan's following (though that helped with discovery). The tool is actually useful, especially the Release Manager and QA Agent functionality.
Rating: 4/5 — Especially recommended for solo founders and indie hackers. Read the source before installing, then add it to your permanent setup.
FAQ
Does GStack work with Cursor or other Claude Code alternatives? The docs indicate Claude Code as the primary compatible agent. Some features may work in Cursor, but it's optimized for Claude Code specifically.
How do I invoke a specific agent persona? Check the repository's README for the current slash command list. Commands typically follow a pattern like /ceo, /qa, or /release. The SKILL.md files in each subfolder document the specifics.
Is it safe to install without sandboxing? From my code review: yes, GStack is context-injection only with no external API calls. That said, I'd always recommend reading the source before installing any community skill. Repository at github.com/garrytan/gstack.
Will it conflict with other installed skills? Unlikely. Skills generally don't interfere unless they modify the same configuration. Test on a non-critical project first if you're cautious.
Does the CEO Agent actually help with prioritization, or is it just generic advice? Better than generic when you provide proper context. Give it your target user, current traction, and full backlog with rough estimates, the output becomes noticeably more calibrated than asking "what should I build next?"
How often is GStack updated? As of April 2026, the last update was April 19th. The repository shows active maintenance with regular commits. Star growth of ~3,900/week indicates a healthy community.
Share this article
⚙Related Tools
📄Related Articles
Get More AI Tool Guides
New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.