Run a Company Solo

Does this sound familiar?

You want to build a product on your own, but you don’t know what to do first.
“I need to finish this one thing this week,” you tell yourself, without knowing where to begin.
Start coding and you can have 2,000 lines in an hour.
But is it the right direction? You’re still not sure.

You open a PR without knowing who should review it.
Design? Engineering management? Security? Developer experience?
When you write and merge everything alone, bugs keep reaching production.
QA gets squeezed in at the end, and your users find the problems first.

There’s no second perspective to ask, “What am I missing?”

The real problems

1. AI writes faster than you can review manually.
   It produces 200 lines in five minutes, while reading them carefully and finding bugs takes an hour.

2. Too much manual work piles up before release.
   PR descriptions, documentation, design review and QA all still need your attention.

3. Working alone makes a fresh perspective hard to find.
   When you build everything yourself, it’s difficult to see the product differently.

About the skill

gstack is a collection of skills that Y Combinator’s Garry Tan uses every day.
Install it in Claude Code, then move through /office-hours to clarify the idea, /plan-ceo-review to review scope, /review to inspect code, /qa to test in a browser, and /ship to prepare the PR for release.
The ./setup script detects your environment and installs it.

Recommended for:

• Solo founders who want a faster release cycle
• Senior engineers who want consistent code review, QA, release checks and security audits on each PR
• People who work with multiple AI coding tools

Key features

• /office-hours
  Reframe the idea through six questions.
  “I want a daily briefing app” might become “I’m building a personal AI assistant.” The result is a design document that the later skills can read.

• /plan-ceo-review
  Review the plan from a CEO’s perspective.
  Choose among four modes: expand scope, selectively expand, hold scope or reduce it.

• /plan-eng-review
  Review the architecture as an engineering manager.
  Covers architecture, data flow, edge cases and the test plan, including ASCII diagrams.

• /review
  Inspect a PR from a staff engineer’s perspective.
  Fix straightforward issues automatically and ask for approval on larger decisions.

• /qa
  Open a real Chromium browser and click through the user flows.

• /ship
  Sync with main, run tests, audit coverage and write the PR description in one workflow.

Workflow

CH 01. Install in 30 seconds

git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup

Requires Claude Code, Git and Bun v1.0 or later.
Windows also requires Node.js.

CH 02. First use: clarify the idea

Describe what you want to build in one sentence and invoke /office-hours. Discuss the result you want, as if you were talking through it with a YC founder.

CH 03. Review the plan

Run /plan-ceo-review, then use /plan-eng-review to decide on the architecture.
For a PR with a UI, /plan-design-review follows automatically. Developer tools use /plan-devex-review.

CH 04. Build and review

Write the code, then inspect it with /review.

CH 05. QA and release

Use /qa to open a browser and exercise the user flows.
Fix discovered bugs in commits and add regression tests.
Finally, /ship completes the release workflow.

Is it really free? Is there a separate payment or subscription?

It’s completely free under the MIT license, with no premium tier.

Does it work with agents other than Claude Code, such as Cursor or OpenAI Codex?

Yes. ./setup detects the environment automatically. To choose explicitly, use a command such as ./setup --host codex.

Supports nine hosts: Claude Code, OpenAI Codex CLI, OpenCode, Cursor, Factory Droid, Slate, Kiro, Hermes and GBrain. Skill-folder paths vary by host—for example, ~/.codex/skills/gstack-*—and setup installs to the appropriate location.

What are the environment and disk requirements?

Requires Claude Code, Git and Bun v1.0 or later. macOS and Linux use Bun; Windows also needs Node.js. Allow about 700 MB, including the Chromium cache used by /qa and /browse. Installation requires an internet connection; routine operation runs locally.

Where does it send data? Does company code leave the machine?

Code, file paths and repository names are not sent out. On first use, it asks about telemetry covering skill names, run time, crashes and an anonymous device ID. You can choose anonymous aggregate reporting or disable it entirely.

Change the setting at any time with gstack-config set telemetry off. Browser tasks such as /qa also run on your own machine.