gstack - AI programming workflow tool open-sourced by YC CEO

What is gstack

gstack is an Claude Code for AI programmingworkflow open-sourced by YC CEO Garry Tan, capable of turning an AI assistant into a virtual engineering team. The tool includes 15 expert roles (such as CEO review, Staff engineer, QA lead, etc.) and 6 enhancement tools, all invoked via slash commands. gstack supports running 10-15 sprints in parallel, and features real browser automation, dual-AI cross-review, automatic documentation updates, and more.

gstack's main features

Product refactoring：/office-hours Redefines the product through six mandatory questions, challenging assumptions and generating implementation plans.
Planning review：/plan-ceo-review Reviews requirements from a CEO perspective,/plan-eng-review locks down architecture and test matrix,/plan-design-review scores 0-10 and detects AI low-quality content.
Design system：/design-consultation Builds a complete design system from scratch, generating real prototype images and DESIGN.md documentation.
Code review：/review Identifies production environment risks and automatically fixes them,/investigate systematic root cause analysis, forcibly stops after three failures.
Design fixes：/design-review Personally fixes issues after auditing, atomic commits and generates before-and-after comparison screenshots.
Quality assurance：/qa Real browser testing, automatic fixes and generates regression tests,/qa-only only produces reports without modifying code.
Security audit：/cso Executes OWASP Top 10 and STRIDE threat modeling, providing specific attack scenarios.
Release deployment：/ship One-click sync, test, open PR;/land-and-deploy deploy and verify production health after merge;/canary monitor post-release errors and performance.
Browser automation：/browse Controls real Chromium to perform clicks and screenshots,/setup-browser-cookies imports real browser cookies.
Dual-AI review：/codex Introduces OpenAI Codex for independent review, cross-validating with Claude to find blind spots.
Team retrospective：/retro Generates weekly engineering reports, including personal data, release continuity and testing trends.
Security protection：/careful Warns before destructive commands,/freeze locks edit scope,/guard combines both to provide complete protection.
Efficiency tools：/autoplan One-click chaining of CEO-design-engineering review;/document-release automatically syncs documentation with code changes;/gstack-upgrade self-updates and displays changes.

gstack's key information and usage requirements

Author: Garry Tan (President and CEO of Y Combinator)
Positioning: An open-source toolset that turns Claude Code into a virtual engineering team
Core data: 600,000 lines of code in 60 days, daily output of 10,000-20,000 lines, running 10-15 sprints simultaneously
Composition: 15 expert roles + 6 enhancement tools, all invoked via slash commands
Required tools：Claude Code、Git、Bun v1.0+
Windows extra: Node.js (Bun has Playwright compatibility issues on Windows, automatically falls back to Node)
Installation time: About 30 seconds
Configuration: The project CLAUDE.md needs to add a gstack skills list
Browser features: Optional Chrome/Arc/Brave/Edge for cookie import

Core advantages of gstack

Structured workflow: Supports turning scattered AI prompts into a complete Sprint workflow (think→plan→build→review→test→ship), avoiding chaotic output.
Role specialization: 15 expert roles each focus on their own specialty — the CEO reviews direction, Staff engineers catch bugs, QA opens a real browser to test workflows, simulating real team collaboration.
Parallel capability: Supports running 10-15 independent Sprints simultaneously, managed the same way a CEO manages a team — stepping in for important decisions while the rest run automatically.
Real browser：/browse Controls a real Chromium rather than a simulation, with real clicks, screenshots, and CAPTCHA handling, letting the AI truly "see" the interface.
Dual-AI cross-validation：/codex Introduces OpenAI Codex for independent review, comparing with Claude to find blind spots and improve code quality.
Automated fix loop：/qa After finding a bug, automatically fixes it, generates regression tests, and re-verifies, completing the full fix process without human intervention.

How to use gstack

Install environment: Open Claude Code, paste and run the install command, and Claude will automatically clone the repository, compile the binary files, and register all skills locally.
Configure project: If team sharing is needed, copy gstack into the project .claude/skills/ directory and re-run setup, then add the skills list in CLAUDE.md for Claude to recognize.
Initialize product (/office-hours): Describe the product you want to build, and the AI will question your framing, challenge underlying assumptions, extract hidden requirements, and generate a design document that automatically flows into downstream skills.
Planning review (/plan-ceo-review, etc.): Run the CEO review to set direction, engineering review to lock the architecture, and design review to score it in sequence — three layers of checks to ensure the plan is feasible before entering development.
Code implementation (automatic or manual): After approving the plan, exit planning mode, and the AI automatically writes code based on the design document, or you develop manually and then enter the review stage.
Quality review (/review + /qa): Run /review to find production risks and automatically fix them, then run /qa to open a real browser for click testing, find bugs, automatically fix them, and generate regression tests.
Release and deployment (/ship): Execute /ship to sync the main branch, run tests, audit coverage, push code, and automatically open a Pull Request with one click, completing the release process.
Continuous maintenance (/retro + /gstack-upgrade): Periodically run /retro to review team data, and run /gstack-upgrade to self-update to the latest version to get new features.

gstack project address

GitHub repository：https://github.com/garrytan/gstack

Comparison of gstack with similar competitors

Dimension	gstack	OpenAI Codex	Devin (Cognition)
Positioning	Virtual engineering team workflow for Claude Code	OpenAI's official CLI coding assistant	Fully autonomous AI software engineer
Core form	15 expert roles + 6 tools, invoked via slash commands	Single conversational command-line tool	Independent cloud Agent, end-to-end development
Workflow	Structured Sprint (Think→Plan→Build→Review→Test→Ship)	Free-form conversation, no fixed process	Autonomous planning and execution, minimal human intervention
Parallel capability	Supports 10-15 Sprints running simultaneously	Single session	Single task
Code review	`/review` + `/codex` Dual AI cross-validation	Self-review	Autonomous test validation
Browser testing	`/browse` Real Chromium automation	None	Built-in browser automation
Team collaboration	Via `.claude/skills/` shared configuration	Personal use	Enterprise-grade, per-seat pricing

Use cases for gstack

Technical founders: Need to balance CEO responsibilities while maintaining code output, achieving the development efficiency of one person replacing twenty through a structured process.
First-time Claude Code users: Provides predefined expert roles and workflows, avoiding the confusion of facing a blank prompt with no idea where to start.
Tech Leads and Staff engineers: Need strict code review, QA, and release automation processes to ensure every PR goes through multiple layers of checks.
Parallel multi-project development: Need to advance 10-15 Sprints simultaneously, managing multiple AI sessions through a standardized process without falling into chaos.
Teams pursuing code quality: Need mechanisms such as dual AI cross-review, real browser testing, and automated regression testing to ensure production safety.