DEV Community

Cover image for What Is gstack? This will Change How you Code Forever
Preecha
Preecha

Posted on

What Is gstack? This will Change How you Code Forever

TL;DR / Quick Answer

gstack is Garry Tan’s open-source system that turns Claude Code into a virtual engineering team of 20 specialists. As Y Combinator’s President and CEO, Garry ships 10,000–20,000 lines of production code per day (35% tests) while running YC full-time. gstack uses structured slash commands such as /office-hours for product strategy, /plan-ceo-review for scope validation, /review for bug detection, /qa for browser testing, and /ship for deployment. Install it with git clone ~/.claude/skills/gstack && ./setup. It is free and MIT licensed.

Try Apidog today

Introduction

“I don’t think I’ve typed like a line of code probably since December, basically, which is an extremely large change.”

When Andrej Karpathy said this on the No Priors podcast in March 2026, he described a major workflow shift: one developer using AI agents can ship at the pace of a much larger team.

Peter Steinberger demonstrated a similar model by building OpenClaw—247,000 GitHub stars—essentially solo with AI agents.

Garry Tan built gstack around that workflow. As President and CEO of Y Combinator, he works with startups such as Coinbase, Instacart, and Rippling at their earliest stages. Before YC, he was an early engineer at Palantir, cofounded Posterous (sold to Twitter), and built Bookface, YC’s internal social network.

gstack answers a practical question: how can one developer run a repeatable product, engineering, QA, security, and release process with AI agents?

According to Garry’s reported numbers from the last 60 days:

  • 600,000+ lines of production code
  • 35% test code
  • 10,000–20,000 lines per day
  • Work completed part-time while running YC full-time
  • A recent /retro across three projects reported 140,751 lines added, 362 commits, and approximately 115k net LOC in one week

This guide shows how to install gstack, choose the right skills, and run an implementation workflow from idea to production.

💡 Building API products? gstack can fit into an API testing and documentation workflow. For example, use /qa to validate API-backed browser flows and /document-release to keep API documentation aligned with shipped changes.

What Is gstack?

gstack is an open-source collection of 28 Claude Code skills. Each skill gives the agent a specialized role, such as:

  • Product strategist
  • CEO or founder reviewer
  • Engineering manager
  • Product designer
  • Staff engineer
  • QA engineer
  • Security officer
  • Release engineer
  • SRE
  • Technical writer

The skills are invoked as slash commands and defined in Markdown. gstack is free and MIT licensed.

The Core Workflow

Most AI coding workflows start with a prompt and immediately generate code. gstack encourages a staged process:

Think → Plan → Build → Review → Test → Ship → Reflect
Enter fullscreen mode Exit fullscreen mode

For example, instead of starting with:

Build a daily briefing app for my calendar.
Enter fullscreen mode Exit fullscreen mode

start with:

/office-hours

I want to build a daily briefing app for people who manage multiple
Google calendars and need better meeting preparation.
Enter fullscreen mode Exit fullscreen mode

The goal is to validate the problem before committing to implementation.

A typical feature workflow looks like this:

/office-hours
/plan-ceo-review
/plan-eng-review
/plan-design-review
# Implement the approved plan
/review
/qa https://staging.example.com
/ship
/document-release
Enter fullscreen mode Exit fullscreen mode

Each stage produces context for the next one:

  • /office-hours creates a product-oriented design document.
  • /plan-ceo-review validates scope.
  • /plan-eng-review defines architecture, edge cases, and tests.
  • /review catches code-level issues.
  • /qa validates user flows in Chromium.
  • /ship runs tests and opens a pull request.
  • /document-release updates documentation.

The 28 Skills Explained

Product and Strategy Skills

/office-hours — YC Office Hours

Role: YC Partner

Use /office-hours at the beginning of a new feature, product, or major refactor.

It asks forcing questions, challenges your initial framing, and produces alternatives before you write code.

Example:

You said "daily briefing app." But what you actually described is a
personal chief of staff AI.

Here are five capabilities implied by the user problem...

[challenges four assumptions]
[generates three implementation approaches with effort estimates]

RECOMMENDATION: Ship the narrowest wedge tomorrow and learn from real
usage. The complete vision is a three-month project. Start with the
daily briefing that actually works.
Enter fullscreen mode Exit fullscreen mode

Use it when:

  • Starting a new product idea
  • Planning a large feature
  • You have a feature request but unclear user pain
  • You want alternatives before implementation

/plan-ceo-review — CEO / Founder Review

Role: CEO who rethinks the product

Run this after /office-hours and before implementation.

It reviews the plan from first principles and selects one of four directions:

  1. Expansion — Make the idea significantly more ambitious.
  2. Selective Expansion — Expand only the highest-leverage parts.
  3. Hold Scope — Keep the current plan.
  4. Reduction — Cut most of the scope and ship a narrower wedge.

Use it when:

/office-hours
/plan-ceo-review
Enter fullscreen mode Exit fullscreen mode

This is useful when a plan has become too broad, too safe, or too implementation-driven.

/plan-design-review — Senior Designer

Role: Senior Product Designer

This skill rates design dimensions from 0 to 10, explains what a 10 would look like, and updates the plan.

It also includes AI-slop detection and uses one AskUserQuestion per design decision.

Use it when:

  • A feature includes UI or UX changes
  • You need to review interaction design before coding
  • You want to catch design debt before it becomes code debt 
/plan-design-review
Enter fullscreen mode Exit fullscreen mode

/design-consultation — Design Partner

Role: Design Partner

Use this when you need a complete design system or a more comprehensive design exploration than /plan-design-review.

It can:

  • Research comparable products
  • Propose creative risks
  • Define visual direction
  • Generate realistic product mockups 
/design-consultation

Create a design system for a calendar briefing product used by busy
startup founders. Prioritize dense information, calm hierarchy, and
fast scanning before meetings.
Enter fullscreen mode Exit fullscreen mode

Engineering and Architecture Skills

/plan-eng-review — Engineering Manager

Role: Engineering Manager

Run /plan-eng-review after product and design reviews, but before coding.

It makes architecture decisions explicit by generating:

  • Data-flow diagrams
  • State-machine diagrams
  • Error paths
  • Edge cases
  • Test matrices
  • Security considerations

Example output:

Architecture Review:

┌─────────────┐     ┌──────────────┐     ┌────────────┐
│   Client    │────▶│  API Gateway │────▶│  Database  │
└─────────────┘     └──────────────┘     └────────────┘
       │                    │
       ▼                    ▼
  [State Cache]      [Rate Limiter]

Test Matrix:
- Happy path: authenticated user, valid data
- Edge case: concurrent modifications
- Failure mode: database connection timeout
- Security: SQL injection, XSS, CSRF
Enter fullscreen mode Exit fullscreen mode

Recommended usage:

/plan-eng-review

Review the approved daily briefing plan. Include:
- calendar synchronization failure modes
- duplicate event handling
- rate limiting
- timezone handling
- test cases for stale event data
Enter fullscreen mode Exit fullscreen mode

/review — Staff Engineer

Role: Staff Engineer focused on production bugs

Run /review after implementation and before QA.

It looks for bugs that may pass CI but fail under real production conditions. It can auto-fix obvious issues and ask for approval on riskier changes.

Example:

[AUTO-FIXED] 2 issues:
- Missing null check in getUserById()
- Unhandled promise rejection in API handler

[ASK] Race condition in concurrent update flow

[COMPLETENESS GAP] No retry logic for transient failures
Enter fullscreen mode Exit fullscreen mode

Use it on a branch with changes:

/review
Enter fullscreen mode Exit fullscreen mode

/investigate — Root-Cause Debugger

Role: Debugger

Use /investigate for issues that need systematic diagnosis.

Its core rule is:

No fixes without investigation.

It traces data flow, tests hypotheses, and stops after three failed fixes to avoid random patching.

Use it when:

/investigate

The calendar sync job intermittently creates duplicate briefing records.
Trace the job lifecycle, identify the root cause, and propose a fix only
after validating the failure path.
Enter fullscreen mode Exit fullscreen mode

/codex — Second Opinion

Role: OpenAI Codex CLI

Use /codex as an independent review pass after /review, especially for critical paths.

Available modes include:

  • Review pass/fail gate
  • Adversarial challenge
  • Open consultation
  • Cross-model analysis when both /review and /codex have run 
/review
/codex review
Enter fullscreen mode Exit fullscreen mode

Use this for authentication, payments, authorization, migrations, or other high-risk code.

Testing and QA Skills

/qa — QA Lead

Role: QA Engineer with a real browser

/qa uses headless Chromium to execute real browser flows. It can find bugs, create atomic commits for fixes, re-verify them, and generate regression tests.

Example workflow:

1. Open the staging URL in Chromium
2. Execute the test plan from /plan-eng-review
3. Find a bug: submit button remains enabled while loading
4. Create an atomic fix commit
5. Re-run the flow and verify the fix
6. Generate a regression test
Enter fullscreen mode Exit fullscreen mode

Run it against staging:

/qa https://staging.example.com
Enter fullscreen mode Exit fullscreen mode

For API-backed applications, give QA concrete scenarios:

/qa https://staging.example.com

Test calendar connection, event sync, briefing generation, retry states,
empty-state UX, failed API responses, and logout/login behavior.
Enter fullscreen mode Exit fullscreen mode

/qa-only — QA Reporter

Role: QA Reporter

/qa-only uses the same browser-based testing approach as /qa, but does not modify code.

Use it when you need a report for:

  • QA handoff
  • Audit trails
  • Bug triage
  • A separate engineer responsible for fixes 
/qa-only https://staging.example.com
Enter fullscreen mode Exit fullscreen mode

/benchmark — Performance Engineer

Role: Performance Engineer

Use /benchmark before and after significant rendering or performance work.

It tracks:

  • First Contentful Paint (FCP)
  • Largest Contentful Paint (LCP)
  • Cumulative Layout Shift (CLS)
  • Time to Interactive (TTI)
  • Bundle sizes 
/benchmark https://staging.example.com
Enter fullscreen mode Exit fullscreen mode

Run it:

  • Before a major refactor
  • After performance changes
  • On PRs that modify rendering, routing, or bundle composition

/browse — Browser Automation

Role: Browser Automation

/browse provides direct Chromium automation. It is used internally by /qa.

Common commands:

goto <url>
click <selector>
type <selector> <text>
screenshot <name>
wait <selector>
Enter fullscreen mode Exit fullscreen mode

Example:

/browse goto https://staging.example.com/login
/browse type input[name="email"] dev@example.com
/browse type input[name="password"] password
/browse click button[type="submit"]
/browse wait [data-testid="dashboard"]
/browse screenshot dashboard-loaded
Enter fullscreen mode Exit fullscreen mode

/setup-browser-cookies — Session Manager

Role: Browser Session Manager

Use this before browser QA if staging requires authentication.

It imports cookies from Chrome, Arc, Brave, or Edge into the headless browser session.

/setup-browser-cookies
/qa https://staging.example.com
Enter fullscreen mode Exit fullscreen mode

Security and Compliance Skills

/cso — Chief Security Officer

Role: Chief Security Officer

/cso runs an OWASP Top 10 review and STRIDE threat model.

It uses a zero-noise approach:

  • 17 false-positive exclusions
  • 8/10 confidence threshold
  • Independent finding verification
  • Concrete exploit scenarios for findings

Example:

[CRITICAL] SQL Injection in /api/users?id= parameter

Exploit:
GET /api/users?id=1' OR '1'='1

Impact:
Full database read access

Fix:
Use parameterized queries

Confidence:
9/10
Enter fullscreen mode Exit fullscreen mode

Run it before releasing features that involve:

  • Authentication
  • User data
  • Payments
  • File uploads
  • Admin panels
  • Public APIs 
/cso
Enter fullscreen mode Exit fullscreen mode

Shipping and Deployment Skills

/ship — Release Engineer

Role: Release Engineer

/ship moves a tested branch toward a pull request.

It can:

  1. Sync main
  2. Create or update a feature branch
  3. Run tests
  4. Bootstrap a test framework if needed
  5. Audit coverage
  6. Push the branch
  7. Open a pull request

Example workflow:

1. git checkout main && git pull
2. git checkout -b feature/daily-briefing
3. npm test
4. Coverage audit: 42 tests → 51 tests (+9 new)
5. git push origin feature/daily-briefing
6. Open PR
Enter fullscreen mode Exit fullscreen mode

Run it after /qa clears the branch:

/ship
Enter fullscreen mode Exit fullscreen mode

/land-and-deploy — Deployment Engineer

Role: Deployment Engineer

Use /land-and-deploy after PR approval.

It can:

  1. Merge the pull request
  2. Wait for CI
  3. Wait for deployment
  4. Run production health checks
  5. Report deployment status 
/land-and-deploy
Enter fullscreen mode Exit fullscreen mode

It supports workflows involving platforms such as GitHub Actions, CircleCI, Vercel, Railway, and Fly.io.

/canary — SRE

Role: Site Reliability Engineer

Run /canary immediately after deployment for a 5–15 minute monitoring loop.

It monitors:

  • Browser console errors
  • API error rates
  • Page-load regressions
  • JavaScript exceptions 
/canary https://app.example.com
Enter fullscreen mode Exit fullscreen mode

/document-release — Technical Writer

Role: Technical Writer

Run /document-release after shipping or deployment to keep docs synchronized with released behavior.

Example:

[UPDATED] README.md — added new /qa command to docs
[UPDATED] CHANGELOG.md — v0.4.2 release notes
[CREATED] docs/qa-guide.md — new QA workflow guide
[FLAGGED] API.md — may need update for new endpoints
Enter fullscreen mode Exit fullscreen mode 
/document-release
Enter fullscreen mode Exit fullscreen mode

For API changes, use this after validating endpoint behavior and updating your API reference.

Reflection and Analytics Skills

/retro — Engineering Manager

Role: Engineering Manager

Run /retro at the end of the week to review shipping velocity, test health, and project trends.

/retro global can aggregate across projects and AI tools including Claude Code, Codex, and Gemini.

Example:

Week of March 17–23, 2026

Garry:
- 140,751 lines added
- 362 commits
- ~115k net LOC
- Test coverage: 35% (↑2% from last week)

Projects:
- gstack: 89 commits, 45k LOC
- ycombinator.com: 156 commits, 62k LOC
- internal-tools: 117 commits, 33k LOC

Shipping streak: 47 days
Enter fullscreen mode Exit fullscreen mode

Run:

/retro
/retro global
Enter fullscreen mode Exit fullscreen mode

Power Tools

/careful — Safety Guardrails

Use /careful before risky operations, including:

  • rm -rf
  • DROP TABLE
  • Force pushes
  • Production data changes 
/careful

I am about to run destructive migration and cleanup commands.
Warn me before any command that could delete data or rewrite history.
Enter fullscreen mode Exit fullscreen mode

/freeze — Edit Lock

Restrict edits to a specific directory.

/freeze /src/auth
Enter fullscreen mode Exit fullscreen mode

This is useful while debugging a focused area of the codebase.

/guard — Full Safety

/guard combines /careful and /freeze.

Use it for:

  • Production debugging
  • Database migrations
  • High-risk security fixes 
/guard /src/payments
Enter fullscreen mode Exit fullscreen mode

/unfreeze — Unlock

Remove the edit boundary after the focused task is complete.

/unfreeze
Enter fullscreen mode Exit fullscreen mode

/setup-deploy — Deploy Configurator

Run this once for each new project before using /land-and-deploy.

It detects your deployment platform, production URL, and deploy commands.

/setup-deploy
Enter fullscreen mode Exit fullscreen mode

/autoplan — Review Pipeline

Run the complete planning pipeline in one command:

/autoplan
Enter fullscreen mode Exit fullscreen mode

It runs CEO, design, and engineering reviews automatically, while surfacing taste decisions for your approval.

/gstack-upgrade — Self-Updater

Use this to update gstack.

/gstack-upgrade
Enter fullscreen mode Exit fullscreen mode

Run it monthly or after a new feature announcement.

Installation Guide

gstack installs inside .claude/. It does not modify your PATH or run background services.

Requirements

  • Claude Code
  • Git
  • Bun v1.0+
  • Node.js on Windows only

On Windows, Node.js is required because Bun has a known bug with Playwright’s pipe transport.

Step 1: Install gstack for Your User Account

Open Claude Code and paste:

Install gstack: run git clone https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup then add a “gstack” section to [CLAUDE.md](http://claude.md/?ref=apidog.com) that says to use the /browse skill from gstack for all web browsing, never use mcp__claude-in-chrome__* tools, and lists the available skills: /office-hours, /plan-ceo-review, /plan-eng-review, /plan-design-review, /design-consultation, /review, /ship, /land-and-deploy, /canary, /benchmark, /browse, /qa, /qa-only, /design-review, /setup-browser-cookies, /setup-deploy, /retro, /investigate, /document-release, /codex, /cso, /autoplan, /careful, /freeze, /guard, /unfreeze, /gstack-upgrade. Then ask the user if they also want to add gstack to the current project so teammates get it.
Enter fullscreen mode Exit fullscreen mode

Image

Step 2: Add gstack to a Repository

To commit gstack into a project so teammates receive it after cloning, paste this into Claude Code:

Add gstack to this project: run cp -Rf ~/.claude/skills/gstack .claude/skills/gstack && rm -rf .claude/skills/gstack/.git && cd .claude/skills/gstack && ./setup then add a “gstack” section to this project’s [CLAUDE.md](http://claude.md/?ref=apidog.com) that says to use the /browse skill from gstack for all web browsing, never use mcp__claude-in-chrome__* tools, lists the available skills, and tells Claude that if gstack skills aren’t working, run cd .claude/skills/gstack && ./setup to build the binary and register skills.
Enter fullscreen mode Exit fullscreen mode

Install for Codex, Gemini CLI, or Cursor

gstack works with agents that support the SKILL.md standard. Skills live in .agents/skills/.

Install Into One Repository 

git clone https://github.com/garrytan/gstack.git .agents/skills/gstack
cd .agents/skills/gstack && ./setup --host codex
Enter fullscreen mode Exit fullscreen mode

Install for Your User Account 

git clone https://github.com/garrytan/gstack.git ~/gstack
cd ~/gstack && ./setup --host codex
Enter fullscreen mode Exit fullscreen mode

Auto-Detect Available Agents 

git clone https://github.com/garrytan/gstack.git ~/gstack
cd ~/gstack && ./setup --host auto
Enter fullscreen mode Exit fullscreen mode

Troubleshooting

A Skill Is Not Showing Up

Re-run setup:

cd ~/.claude/skills/gstack && ./setup
Enter fullscreen mode Exit fullscreen mode

/browse Fails

Install dependencies and rebuild:

cd ~/.claude/skills/gstack
bun install
bun run build
Enter fullscreen mode Exit fullscreen mode

Your Install Is Stale

Run:

/gstack-upgrade
Enter fullscreen mode Exit fullscreen mode

Or enable automatic updates in ~/.gstack/config.yaml:

auto_upgrade: true
Enter fullscreen mode Exit fullscreen mode

Windows Setup

gstack works on Windows 11 through Git Bash or WSL.

Confirm that both bun and node are available:

bun --version
node --version
Enter fullscreen mode Exit fullscreen mode

Claude Cannot See the Skills

Add this to your project’s CLAUDE.md:

## gstack

Use /browse from gstack for all web browsing.
Never use mcp__claude-in-chrome__* tools.

Available skills:
- /office-hours
- /plan-ceo-review
- /plan-eng-review
- /plan-design-review
- /design-consultation
- /review
- /ship
- /land-and-deploy
- /canary
- /benchmark
- /browse
- /qa
- /qa-only
- /design-review
- /setup-browser-cookies
- /setup-deploy
- /retro
- /investigate
- /document-release
- /codex
- /cso
- /autoplan
- /careful
- /freeze
- /guard
- /unfreeze
- /gstack-upgrade
Enter fullscreen mode Exit fullscreen mode

Real-World Example: Build a Daily Briefing App

Here is a practical end-to-end workflow.

1. Start With the User Problem 

I want to build a daily briefing app for my calendar.
Enter fullscreen mode Exit fullscreen mode

2. Run Product Discovery 

/office-hours
Enter fullscreen mode Exit fullscreen mode

Provide concrete pain points:

I manage multiple Google calendars. Events often have stale information
or wrong locations. Meeting preparation takes too long, and the output
from existing tools is not useful enough.
Enter fullscreen mode Exit fullscreen mode

The agent may respond:

You said "daily briefing app." But what you actually described is a
personal chief of staff AI.

[extracts five implied capabilities]
[challenges four premises]
[generates three implementation approaches with effort estimates]

RECOMMENDATION: Ship the narrowest wedge tomorrow. Start with the daily
briefing that actually works.
Enter fullscreen mode Exit fullscreen mode

3. Challenge Scope 

/plan-ceo-review
Enter fullscreen mode Exit fullscreen mode

Use this to decide whether to:

  • Ship only calendar aggregation first
  • Include meeting preparation from day one
  • Add email and CRM context later
  • Cut features that do not prove user value

4. Lock Architecture and Tests 

/plan-eng-review
Enter fullscreen mode Exit fullscreen mode

Ask it to explicitly cover:

Include:
- OAuth token refresh failures
- Calendar API rate limits
- Duplicate events
- Timezone normalization
- Partial sync failures
- Retry behavior
- Test coverage for stale event metadata
Enter fullscreen mode Exit fullscreen mode

5. Implement the Approved Plan 

Approve plan. Exit plan mode.
Enter fullscreen mode Exit fullscreen mode

6. Review the Code 

/review
Enter fullscreen mode Exit fullscreen mode

Example result:

[AUTO-FIXED] 2 issues.
[ASK] Race condition in concurrent update flow.
[COMPLETENESS GAP] Missing retry logic for transient API failures.
Enter fullscreen mode Exit fullscreen mode

7. Run Browser QA 

/qa https://staging.myapp.com
Enter fullscreen mode Exit fullscreen mode

The QA flow can open a real browser, click through the application, find an issue, apply a fix, and verify it.

8. Ship the Pull Request 

/ship
Enter fullscreen mode Exit fullscreen mode

Example output:

Tests: 42 → 51 (+9 new)
PR: github.com/you/app/pull/42
Enter fullscreen mode Exit fullscreen mode

The key outcome is not just code generation. The workflow turns a vague request into a scoped, reviewed, tested, and shippable implementation.

Parallel Sprints: Run Multiple Agents at Once

gstack can support multiple Claude Code sessions running in parallel. Conductor runs sessions in isolated workspaces.

Example:

Session 1: /office-hours — refine the product spec
Session 2: /review — review yesterday's feature
Session 3: /qa — test the staging deployment
Session 4: Implementation — build an approved plan
Enter fullscreen mode Exit fullscreen mode

Use parallel work only when each session has a clear boundary.

A practical split:

Session Responsibility Output
Product /office-hours Design doc and scope
Architecture /plan-eng-review Technical plan and test matrix
Implementation Build approved tasks Feature branch changes
QA /qa Verified flows and regression tests
Release /ship Pull request

Without a process, multiple agents can create conflicting changes. With a sprint structure, each agent has a defined stage and deliverable.

Privacy and Telemetry

gstack includes opt-in usage telemetry.

Default Behavior

Telemetry is disabled by default. Nothing is sent unless you explicitly opt in.

Data Sent If You Opt In

  • Skill name
  • Duration
  • Success or failure
  • gstack version
  • Operating system

Data Not Sent

  • Source code
  • File paths
  • Repository names
  • Branch names
  • Prompts
  • User-generated content

Disable telemetry at any time:

gstack-config set telemetry off
Enter fullscreen mode Exit fullscreen mode

Telemetry data is stored in Supabase. The schema is available in the repository for inspection. The Supabase publishable key is public, while row-level security policies restrict access to insert-only operations.

For local analytics without remote data, run:

gstack-analytics
Enter fullscreen mode Exit fullscreen mode

Who Should Use gstack?

Founders and Technical CEOs

Use gstack to add structure to product planning, engineering review, QA, and release work without building a large internal process from scratch.

First-Time Claude Code Users

The slash-command workflow provides guardrails instead of requiring a perfect open-ended prompt.

Tech Leads and Staff Engineers

Use individual skills such as /review, /qa, /cso, and /document-release to strengthen existing PR workflows.

Solo Builders

gstack provides specialized roles for planning, debugging, review, testing, and release management.

YC Startups

Garry built gstack for YC founders, making it a natural fit for early-stage teams that need to move quickly with limited engineering capacity.

Who Should Skip the Full Workflow?

Teams With Mature Processes

If you already have strong design review, code review, CI/CD, QA, and release processes, use individual skills instead of adopting the full sprint.

For example:

/review
/qa https://staging.example.com
/cso
/document-release
Enter fullscreen mode Exit fullscreen mode

Developers Who Prefer Freeform AI Workflows

gstack is intentionally structured. If you prefer unbounded exploration and loose prompting, the workflow may feel restrictive.

Non-Claude Code Users

gstack supports Codex, Gemini CLI, and Cursor through SKILL.md, but it is optimized for Claude Code.

The Philosophy Behind gstack

Boil the Lake

Do not half-complete a task that requires a full solution. Partial work often creates more follow-up work than completing the important path correctly.

Search Before Building

Before generating new code, check whether the repository or ecosystem already has a solution.

This reduces duplicate implementations and avoids code that does not need to exist.

Three Layers of Knowledge

  • Explicit knowledge — Documentation, comments, and written rules
  • Tacit knowledge — Experience, intuition, and engineering judgment
  • Unknown knowledge — Blind spots you do not know to investigate

gstack attempts to encode tacit knowledge into reusable skills.

For example, /review is not just “look for bugs.” It applies a structured production-review process based on engineering experience.

The Iron Law of Debugging

No fixes without investigation.

After three failed attempts, stop and reassess.

This prevents both humans and AI agents from applying random changes without understanding the root cause.

Conclusion

gstack is Garry Tan’s system for using AI agents as a structured engineering team.

The practical workflow is:

/office-hours        # Reframe the problem
/plan-ceo-review     # Challenge scope
/plan-eng-review     # Lock architecture and test plan
/plan-design-review  # Review UX and design quality
# Build the approved plan
/review              # Find implementation issues
/qa                  # Test real browser flows
/ship                # Push and open a pull request
Enter fullscreen mode Exit fullscreen mode

Start small. You do not need to adopt every skill on day one.

A good first workflow is:

/review
/qa https://staging.example.com
/ship
Enter fullscreen mode Exit fullscreen mode

Then add planning skills when you need more rigor before implementation.

Next Steps

  1. Install gstack: 
   git clone https://github.com/garrytan/gstack.git ~/.claude/skills/gstack
   cd ~/.claude/skills/gstack
   ./setup
Enter fullscreen mode Exit fullscreen mode

  1. Run /office-hours on your next feature idea.

  2. Run /review before opening your next pull request.

  3. Run /qa against staging before release.

  4. Use /document-release to keep project and API documentation synchronized with shipped behavior.

FAQ

Is gstack free?

Yes. gstack is free and MIT licensed. There is no premium tier or waitlist.

Do I need Claude Code to use gstack?

gstack is optimized for Claude Code, but it also works with agents that support the SKILL.md standard, including Codex CLI, Gemini CLI, and Cursor.

How long does installation take?

About 30 seconds:

git clone https://github.com/garrytan/gstack.git ~/.claude/skills/gstack
cd ~/.claude/skills/gstack
./setup
Enter fullscreen mode Exit fullscreen mode

Can I use individual skills without the full sprint?

Yes. The sprint is a recommended workflow, not a requirement.

For example:

/review
/qa https://staging.example.com
Enter fullscreen mode Exit fullscreen mode

Does gstack work with private repositories?

Yes. Skills can live in .claude/skills/gstack inside your repository. Commit that directory to share the setup with teammates.

Does gstack work on Windows?

Yes. gstack works on Windows 11 through Git Bash or WSL. Install both Bun and Node.js, and ensure they are available on your PATH.

How does /browse work?

/browse uses Playwright to control a headless Chromium browser. Commands execute in approximately 100ms. Use /setup-browser-cookies to test authenticated pages.

Can I customize gstack skills?

Yes. Skills are Markdown files. Edit them to match your repository conventions, team process, and release workflow.

What is the difference between /qa and /qa-only?

  • /qa finds bugs and can auto-fix them with atomic commits.
  • /qa-only finds bugs and reports them without changing code.

Use /qa-only when you need an audit trail or separate ownership for fixes.

How does telemetry work?

Telemetry is opt-in. If enabled, gstack sends skill name, duration, success or failure, version, and OS. It does not send code, file paths, repository names, branch names, or prompts.

Disable it with:

gstack-config set telemetry off
Enter fullscreen mode Exit fullscreen mode

What if I find a bug in gstack?

Run /investigate against gstack’s codebase or open an issue on GitHub.

Can I run gstack skills in parallel?

Yes. With Conductor, you can run multiple Claude Code sessions in isolated workspaces. Assign each session a clear stage, such as planning, implementation, review, or QA.

Top comments (0)