# gstack: Turn Claude Code Into a Full Engineering Team > Source: > Published: 2026-11-05 19:06:23+00:00 Y Combinator CEO Garry Tan shipped three production services and over 40 features in 60 days — part time, while running YC full-time. The tool making that possible is a 23-command Claude Code skill pack he open-sourced in March. It has [139,000 GitHub stars](https://github.com/garrytan/gstack). Here is how it works and whether the numbers hold up. ## What gstack Is gstack is a collection of slash commands for Claude Code. Each command instills a specific engineering role’s mindset and checklist — CEO, Engineering Manager, QA Lead, Designer, Chief Security Officer, Release Engineer — before Claude writes anything. That distinction matters. Telling Claude “act as a CEO” produces generic results. Running `/plan-ceo-review` loads a full prompt scaffold that asks whether you are building the right feature at all, not just whether you are building it correctly. It is MIT-licensed, requires Claude Code and Bun v1.4.2 or newer, and installs in one command: ``` git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup ``` After that, add one line to your project’s `CLAUDE.md` pointing gstack’s `/browse` skill at web lookups, and you are done. No configuration files, no API keys, no subscriptions beyond your existing Claude Code plan. ## The Five Commands Worth Learning First gstack ships 23 commands. Start with five: - **/plan-ceo-review** — Challenges scope before you build. The command asks whether the literal request is even the right feature, and whether a simpler or deeper version serves users better. - **/plan-eng-review** — Locks architecture, data flow, and edge cases. Surfaces hidden assumptions and writes a test plan before implementation starts. - **/review** — Senior engineer bug review. Runs after implementation, before merge. - **/qa** — Opens a real browser and tests APIs, CLI behavior, webhooks, and jobs against a staging URL. Not a mock. Actual requests. - **/cso** — OWASP and STRIDE security audit with an explicit coverage report. Finds what generic code review misses. The other 18 commands — `/office-hours`, `/plan-design-review`, `/ship`, `/autoplan`, and more — extend the same pattern into documentation, design assessment, and automated sprint orchestration. ## The Workflow The sequence matters as much as the commands. gstack’s intended flow: 1. Run `/office-hours` to write a design doc for the feature. 2. Run `/plan-ceo-review` to challenge whether that feature is correct. 3. Run `/plan-eng-review` to lock architecture and produce a test plan. 4. Implement the feature with Claude Code. 5. Run `/review` to catch bugs before staging. 6. Run `/qa` against a staging URL. 7. Run `/ship` to merge. `/autoplan` chains steps two, three, and the review into a single invocation if you want to skip the manual sequencing. The full sprint runs in one pass. ## On the 810x Claim Garry Tan reports his 2026 development pace at 810 times his 2013 baseline — 11,417 logical lines of code per day versus 14. That number is everywhere, and it is not quite what it appears. First, “logical LOC” measures new behaviors added, not raw lines written or reformatted — a more honest metric than it sounds. Second, Tan is comparing his own pre-AI productivity to his post-AI productivity, not a controlled study, but a first-person data point from someone with 60 days of continuous output to back it up. Third, [Tan’s own documentation](https://github.com/garrytan/gstack/blob/main/docs/ON_THE_LOC_CONTROVERSY.md) acknowledges that if you use a more conservative 50 LOC/day baseline, the multiple drops to 228x. Independent developers report 1.5x to 3x gains, not 800x. That is still significant. The more defensible framing: gstack does not make Claude write faster. It makes Claude more accountable — plans reviewed before implementation, security audited before merge, QA run against real infrastructure. The uplift is as much quality as speed. ## Who Should Use It gstack is built for solo founders, indie hackers, and small teams of two to five who use Claude Code daily and want planning and review discipline without hiring for it. If you regularly ship features without a design review or security audit, gstack adds those gates for free. It is a poor fit for teams that already have structured processes — a PM, an Eng Manager, and a QA cycle. Adding gstack to an existing process creates overlap, not leverage. It also adds limited value in hardware-adjacent development or regulated industries where human review is mandatory. The honest verdict: install it, run `/qa` once on a live feature, and decide from there. The browser-based QA alone finds issues most developers catch only after users report them. The [full command reference is at gstacks.org](https://gstacks.org/). For a complementary planning-first approach, [GitHub Spec Kit](https://github.com/github/spec-kit) handles the spec lifecycle before gstack takes over execution.