How GStack Turns One Developer into a Full‑Scale Software Team
GStack, an open‑source AI‑driven virtual software factory created by Y Combinator CEO Garry Tan, transforms Claude Code into a role‑based development team, offering 30+ slash commands that automate the entire product lifecycle—from requirement gathering to release—enabling a single developer to achieve the productivity of a twenty‑person team.
What is GStack?
GStack is an open‑source project created and maintained by Garry Tan, the current CEO of Y Combinator. Its core mission is to turn Claude Code from a general‑purpose AI assistant into an on‑demand professional development team.
GitHub: https://github.com/garrytan/gstack (68.8k stars)
Positioning and Principles
GStack is positioned as an AI‑driven “virtual software factory” that provides a complete engineering workflow rather than a collection of prompts. It decomposes a large model into specialized expert roles, each with fixed responsibilities, processes, and outputs.
Effectiveness
According to the author’s own test, a single developer using GStack can achieve productivity comparable to a 20‑person traditional team, writing up to 60 000 lines of production code per day (35 % covered by tests).
Key Features
Role Coverage : 26 skills map to 26 professional roles covering CEO, designers, engineers, QA, security, documentation, etc.
Closed‑Loop Process : From requirement exploration (/office‑hours) → planning (plan‑*) → design → development → testing → release → retro, forming a complete loop.
Decision Layers : Simulates real‑team decision‑making (CEO sets direction, designers control experience, engineers ensure quality).
Multi‑Agent Collaboration : Supports multiple AI agents working together via /pair-agent.
Comparison with Traditional AI Coding
Traditional AI coding relies on a single generic assistant with mixed thinking patterns, ad‑hoc prompts, and limited scope (code generation only). GStack replaces this with 26 specialized roles, structured workflows, end‑to‑end coverage from requirement to release, and a simulated full‑stack team.
Core Workflow
The core of GStack is a set of “skills” (Markdown prompt files) that implement a role‑based workflow. The typical sprint sequence is:
Think
Plan
Build
Review
Test
Ship
Reflect
Each skill invokes the next one using slash commands such as /office-hours, /plan-ceo-review, /plan-eng-review, /qa, /ship, etc.
Installation
Open Claude Code and run the following command to clone and set up GStack:
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setupCommon pitfalls: the setup script requires bun to be installed and added to the PATH. Install bun separately if the automatic step fails.
Main Commands
/office-hours– CEO/Product Manager: initiates a design document through six mandatory questions, turning vague ideas into a concrete plan. /plan-ceo-review – CEO/Founder: refines product ideas, exploring four planning modes (expand, selective expand, maintain scope, shrink). /plan-eng-review – Engineering Manager: defines architecture, data flow, edge cases, and test strategy, exposing hidden assumptions. /plan-design-review – Senior Designer: scores design dimensions from 0‑10, iterates to reach target quality, and runs low‑quality content detection. /review – Senior Engineer: finds production‑critical bugs, auto‑fixes obvious issues, and marks integrity gaps. /qa – QA Lead: tests the application, creates atomic fixes, and generates regression tests automatically. /ship – Release Engineer: syncs the main branch, runs tests, audits coverage, pushes a PR, and can bootstrap a test framework if none exists. /canary – SRE: monitors post‑deployment health, errors, performance regressions, and page failures. /cso – Chief Security Officer: runs OWASP Top 10 + STRIDE threat modeling with low false‑positive noise.
Technical Architecture
Skill System : Markdown prompt files integrated via Claude Code’s custom command feature.
Browser Automation : Playwright‑based persistent Chromium sessions with cookie import and real‑environment testing.
Multi‑Platform Support : Works with Claude Code, OpenAI Codex CLI, Cursor, OpenCode, and eight other AI coding tools.
Team Mode : Supports global and project‑level installation with automatic updates to ensure consistent versions across a team.
Target Audience
Technical founders and CEOs who want to be hands‑on in delivery.
First‑time Claude Code users seeking structured roles instead of free‑form prompts.
Tech leads and senior engineers who need strict PR review, QA, and automated release pipelines.
Project Information
GitHub: https://github.com/garrytan/gstack
License: MIT
Latest version (as of March 2026): v0.4.3
Installation: 30‑second clone into Claude Code’s skills directory.
Core idea: Structured role division and review workflow upgrade AI‑assisted development from vague conversation to a professional engineering pipeline, enabling a “one‑person army” level of efficiency.
Wukong Talks Architecture
Explaining distributed systems and architecture through stories. Author of the "JVM Performance Tuning in Practice" column, open-source author of "Spring Cloud in Practice PassJava", and independently developed a PMP practice quiz mini-program.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
