garrytan/gstack

⭐ 92,193  ·  #5  ·  TypeScript

Use Garry Tan's exact Claude Code setup: 23 opinionated tools that serve as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA

TypeScript Webui

Project Analysis

🎯 Positioning	Visual interaction layer
💡 Core Value	Encapsulates Agent's command-line capabilities into a web interface, supporting session management, history, multi-model switching, etc., lowering the barrier for non-technical users
👥 Target Audience	Users unfamiliar with terminal operations, or scenarios requiring team collaboration with Agent

Why It's Worth Attention

92,193 Stars indicates this is a mature tool validated by a large user base. Developed in TypeScript. Core highlight: "I don't think I've typed like a line of code probably since December, basically, which is an extremely large change." — Andrej Karpathy, No Priors podcast, March 2026.

A CLI toolset that arms Claude Code as a virtual software factory.

Core Features

• 23 Expert Roles (Slash Commands): Define roles like CEO, Designer, Engineering Manager, QA, Security Officer via Markdown prompts. Each role has an independent workflow, e.g., /plan-ceo-review for product strategy review, /qa for end-to-end testing in a real browser.
• Complete Delivery Pipeline: Covers the entire software delivery lifecycle from product ideation (/office-hours), architecture review (/plan-ceo-review), code review (/review), quality assurance (/qa), to release management (/release).
• Automated Security Audit: Built-in /security-audit command automatically performs OWASP Top 10 and STRIDE threat modeling checks, outputting structured audit reports.
• Zero Cost, Zero Dependency: All tools are pure Markdown prompts + Claude Code native capabilities, requiring no additional SaaS subscriptions or infrastructure.
• MIT Open Source, Forkable for Customization: Prompts and toolchains are fully open. Developers can freely modify role definitions, adjust workflows, or extend new slash commands.

Technical Architecture

• Language & Runtime: Primarily TypeScript, depends on Claude Code CLI, Git, and Bun (Node.js only needed for Windows).
• Code Structure: Core assets are Markdown files in the prompts/ directory. Each role corresponds to a .md file (e.g., ceo-review.md, qa-engineer.md), loaded via Claude Code's Slash Command mechanism. No complex build systems or frameworks.
• Design Philosophy: Replaces traditional code development with "prompt engineering." Instead of writing business logic, it defines AI agent behavior boundaries, output formats, and interaction flows. Essentially an "operating system for AI agents"—orchestrating multi-agent collaboration through structured prompts.
• Highlight: Encapsulates AI uncertainty within fixed role boundaries. For example, the /release command not only creates a PR but also requires the agent to perform changelog generation, version number validation, rollback plan checks, etc., significantly reducing the risk of AI hallucinations.

Quick Start Guide

1. Install Dependencies: Ensure Claude Code, Git, and Bun (or Node.js for Windows) are installed.
2. Clone the Repository: Run the following command in the Claude Code terminal:

   git clone --single-branch --depth 1 https://github.com/garrytan/gstack

3. Run Your First Command: Start Claude Code in the project root directory, type /office-hours and describe your product idea.
4. Typical Workflow: /office-hours → /plan-ceo-review → Write code → /review → /qa → /release.

Strengths, Weaknesses, and Use Cases

Strengths:

• Extreme Efficiency: A single person can handle the workflow of a traditional 5-10 person team, especially suitable for prototyping and rapid iteration.
• Zero Learning Curve: For developers familiar with Claude Code, it's plug-and-play with no need to learn new frameworks or DSLs.
• Full Transparency: All prompts and processes are open source, auditable, and modifiable, avoiding the "black box" problem of commercial AI tools.
• Strongly Constrained Output: Role definitions and output templates significantly reduce the probability of AI generating meaningless content.

Weaknesses:

• Tightly Bound to Claude Code: Cannot be migrated to other models like GPT-4 or Gemini, posing a single-vendor risk.
• Dependent on AI Quality: The upper limit of effectiveness depends on Claude Code's actual capabilities. Complex business logic or rare edge cases may lead to hallucinations or errors.
• Lack of Visual Interface: Pure CLI operation is not user-friendly for non-technical team members.
• Not Suitable for Large Teams: Designed as a "one-person super team"; multiple people using Claude Code concurrently on the same repository may cause conflicts.

Use Cases:

• Solo Developers / Technical Founders: Want to quickly validate product ideas with minimal manpower.
• Small Teams (1-5 people): Need to cover the entire process but lack dedicated QA, security, or release engineers.
• AI Tool Explorers: Technicians interested in prompt engineering and AI agent orchestration.

Not Suitable For:

• Large enterprise projects (requiring strict compliance audits and multi-person approval).
• Scenarios with extremely high reliability requirements for model output (e.g., medical or financial core systems).
• Organizations with mature DevOps and QA infrastructure.

Community and Popularity

• Stars: 92,193 (as of analysis date), growing rapidly, reflecting significant interest in the AI agent tool space.
• Forks / Contributions: Due to the project's nature (a collection of prompts rather than a traditional codebase), forks are expected to be lower than stars, but "fork and customize" is the intended primary usage pattern.
• Recent Updates: Last updated on May 9, 2026, indicating active maintenance. Garry Tan himself continues to iterate, with impressive contribution data (1,237 contributions) shown in the README for 2026.
• Community Ecosystem: The project encourages users to share custom role prompts via Issues and PRs, potentially forming a "prompt marketplace"-style community resource pool.

Technical Information

• 💻 Language: TypeScript
• 📂 Topics:
• 🕐 Updated: 2026-04-19
• 🔗 Visit GitHub Repository

---

Data updated on 2026-05-09 · Star count based on actual GitHub data