Back

gstack Tutorial Series: Garry Tan's 23-Specialist AI Coding Framework

gstack is Garry Tan's open-source AI coding assistant framework with 23 specialist skills and 8 power tools. It turns Claude Code into a virtual engineering team—CEO, Designer, Eng Manager, Release Engineer, QA Lead, and more. This series covers the full sprint workflow: Think → Plan → Build → Review → Test → Ship → Reflect.

Series Preface

“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

When Karpathy said this, the AI coding world took notice. How does one person ship like a team of twenty? Peter Steinberger built OpenClaw — 247K GitHub stars — essentially solo with AI agents. The revolution is here.

I’m Garry Tan, President & CEO of Y Combinator. I’ve worked with thousands of startups — Coinbase, Instacart, Rippling — when they were one or two people in a garage. Before YC, I was one of the first eng/PM/designers at Palantir, cofounded Posterous (sold to Twitter), and built Bookface, YC’s internal social network.

gstack is my answer. In the last 60 days: 600,000+ lines of production code (35% tests), 10,000-20,000 lines per day, part-time, while running YC full-time. This is how I do it.

Why gstack?

Most AI coding assistants share the same weakness: they rush to write code. When you describe what you want, the model often starts generating implementation immediately—without deeply understanding the real goal, considering design tradeoffs, or ensuring test coverage.

gstack is built on a simple idea: think before you code.

It turns Claude Code into a virtual engineering team — a CEO who rethinks the product, an eng manager who locks architecture, a designer who catches AI slop, a reviewer who finds production bugs, a QA lead who opens a real browser, a security officer who runs OWASP + STRIDE audits, and a release engineer who ships the PR.

The Sprint Workflow

flowchart LR
    A[THINK] --> B[PLAN]
    B --> C[BUILD]
    C --> D[REVIEW]
    D --> E[TEST]
    E --> F[SHIP]
    F --> G[REFLECT]

    style A fill:#e3f2fd
    style B fill:#e8f5e9
    style C fill:#fff3e0
    style D fill:#fce4ec
    style E fill:#f3e5f5
    style F fill:#e0f2f1
    style G fill:#fff8e1

Each skill feeds into the next. /office-hours writes a design doc that /plan-ceo-review reads. /plan-eng-review writes a test plan that /qa picks up. /review catches bugs that /ship verifies are fixed. Nothing falls through the cracks.

The 23 Specialists

Skill Role What They Do
/office-hours YC Office Hours Six forcing questions that reframe your product before you write code
/plan-ceo-review CEO / Founder Find the 10-star product hiding inside the request. Four scope modes.
/plan-eng-review Eng Manager Lock in architecture, data flow, diagrams, edge cases, and tests
/plan-design-review Senior Designer Rate each design dimension 0-10, catch AI slop, interactive review
/plan-devex-review DX Lead Developer personas, TTHW benchmarks, magical moment, friction points
/design-consultation Design Partner Build a complete design system from scratch
/design-shotgun Design Explorer Generate 4-6 AI mockup variants, comparison board, taste memory
/design-html Design Engineer Turn mockups into production HTML with Pretext computed layout
/browse QA Engineer Real Chromium browser, actual clicks, screenshots, ~100ms per command
/open-gstack-browser Browser Launcher GStack Browser with sidebar, anti-bot stealth, auto model routing
/pair-agent Multi-Agent Coordinator Share browser with any AI agent. Scoped tokens, tab isolation.
/setup-browser-cookies Session Manager Import cookies from Chrome/Arc/Brave/Edge
/review Staff Engineer Find bugs that pass CI but blow up in production. Auto-fixes.
/codex Second Opinion Independent code review from OpenAI Codex CLI
/qa QA Lead Test your app, find bugs, fix with atomic commits, re-verify
/qa-only QA Reporter Same as /qa but report only, no code changes
/cso Chief Security Officer OWASP Top 10 + STRIDE threat model. Zero-noise design.
/design-review Designer Who Codes Same audit as plan-design-review, then fixes what it finds
/devex-review DX Tester Live developer experience audit. Compare against plan scores.
/ship Release Engineer Sync main, run tests, audit coverage, push, open PR
/land-and-deploy Release Engineer Merge PR, wait for CI, deploy, verify production
/canary SRE Post-deploy monitoring for console errors, performance regression
/benchmark Performance Engineer Baseline page load, Core Web Vitals, before/after comparison
/document-release Technical Writer Update all docs to match what shipped
/retro Eng Manager Weekly retro with per-person breakdowns, shipping streaks
/learn Memory Manage what gstack learned across sessions
/investigate Debugger Systematic root-cause debugging. No fixes without investigation.
/careful Safety Guardrails Warn before destructive commands
/freeze Edit Lock Restrict edits to one directory
/guard Full Safety /careful + /freeze combined
/unfreeze Unlock Remove the freeze boundary
/gstack-upgrade Self-Updater Upgrade gstack to latest

The 8 Power Tools

Tool What It Does
Real Browser Mode /browse gives the agent eyes. Actual Chromium, actual clicks, actual screenshots.
GStack Browser /open-gstack-browser launches branded browser with sidebar agent, anti-bot stealth.
Sidebar Agent Natural language in Chrome side panel. Auto-routes Sonnet for actions, Opus for analysis.
Design Pipeline /design-consultation/design-shotgun/design-html. Visual exploration to production HTML.
Security Audit /cso runs OWASP Top 10 + STRIDE. Zero-noise with 8/10+ confidence gate.
Cross-Model Review /codex gets second opinion from OpenAI. Cross-model analysis when both review.
Multi-Agent Coordination /pair-agent shares browser between Claude Code, Codex, Hermes, etc.
Parallel Sprints Run 10-15 Claude Code sessions simultaneously with Conductor.

Tutorial Structure

This series contains 12 detailed tutorials, organized by sprint stage:

Getting Started

  1. Install & Setup — 30-second install, multi-platform support, team mode
  2. office-hours — YC-style product interrogation
  3. plan-ceo-review — Strategic scope challenge

Core Skills

  1. Plan Reviews — Engineering, DevEx, and Design plan reviews
  2. Design Pipeline — Visual exploration to production HTML
  3. Browser Testing — Real Chromium, sidebar agent

Quality & Ship

  1. Code Review — Staff engineer + cross-model second opinion
  2. QA & Security — Live testing + OWASP+STRIDE audits
  3. Release Engineering — /ship, /land-and-deploy, /canary

Advanced

  1. Design Review Live — Live UX/DX audit
  2. Docs & Retro — Documentation and retrospectives
  3. Parallel Sprints — 10-15 parallel workers with Conductor

Who This Is For

  • Founders and CEOs — especially technical ones who still want to ship
  • First-time Claude Code users — structured roles instead of a blank prompt
  • Tech leads and staff engineers — rigorous review, QA, and release automation
  • OpenClaw users — gstack works directly in OpenClaw agents

Prerequisites

  • An active Claude subscription (for Claude Code)
  • Git configured on your machine
  • Bun v1.0+ installed
  • Node.js (Windows only, for Playwright)
  • Basic familiarity with your chosen AI coding assistant

How to Use This Series

  1. Read in order — skills build on each other
  2. Practice first — each tutorial includes hands-on examples
  3. Start with /office-hours — even if you skip setup, don’t skip product thinking

Platform Support

gstack works on 8 AI coding agents:

Agent Install Command
Claude Code ./setup
OpenAI Codex CLI ./setup --host codex
Cursor ./setup --host cursor
OpenCode ./setup --host opencode
Factory Droid ./setup --host factory
Slate ./setup --host slate
Kiro ./setup --host kiro
OpenClaw ClawHub marketplace

Series navigation:

Repository: github.com/garrytan/gstack

Discussion: Share your gstack experience or ask questions in the comments!