# better-gstack

> A minimal fork of gstack providing two AI agent skills—office-hours and plan-ceo-review—for pre-build product validation and plan review, without the surrounding system overhead.

better-gstack is a MIT-licensed fork of Garry Tan's gstack, created by Michael Ryaboy (mrmps), that strips the original 80+ skill suite down to just two skills: `office-hours` and `plan-ceo-review`. It is installable via the skills.sh CLI or by cloning the repository directly, and works with any AI agent that reads `SKILL.md` files, including Claude Code.

## What It Is

better-gstack is a minimal agent skill package targeting the two highest-leverage moments in software development: before any code is written (product validation) and before a plan is executed (scope review). The project is explicitly not affiliated with Garry Tan or Y Combinator — it is an independent fork that preserves the core skill methodology while removing the state management, telemetry, update checker, session tracking, and onboarding prompts that shipped with the original gstack. What remains is approximately 1,400 lines across four markdown files.

## The Two Skills

**office-hours** runs a YC-partner-style diagnostic before you build. It asks six questions sequentially, each pushed until the answer becomes specific:
- Who would be upset if the product vanished tomorrow?
- What workaround are users living with, and what does it cost them?
- Name the actual human — not a category like "marketing teams"
- What's the smallest version someone pays for this week?
- Have you watched someone use it without helping them?
- In three years, does this get more essential or less?

The questions adapt by stage: pre-product gets questions 1–3, paying-customer stage gets 4–6. A second mode exists for side projects, hackathons, and open source, which shifts from interrogation to finding the most interesting version of the idea. Both modes write a design doc to `docs/design/` in the current repo.

**plan-ceo-review** takes a plan and evaluates whether it is the right size. The user selects one of four postures — scope expansion, selective expansion, hold scope, or scope reduction — and the skill runs an 11-section review covering architecture, error and rescue mapping, threat model, edge cases, code quality, tests, performance, observability, deployment, long-term trajectory, and design. Every scope change requires an explicit opt-in; the skill does not add or remove work autonomously. It reads whatever `office-hours` wrote to `docs/design/`, so running them in sequence produces better output than running either alone.

## What Was Removed from gstack

The fork deliberately deleted the surrounding system that gstack ships with:
- `~/.gstack/` state directory and home-directory database
- Telemetry and analytics logging
- Update checker, version pinning, and upgrade prompts
- Setup script and symlink management
- Session tracking, builder profiles, and relationship tiers
- ~30 first-run onboarding prompts that fired before every invocation
- Cross-skill handoff notes, review dashboards, and decision logs
- The YC application pitch in the closing

Design docs now write to `docs/design/` and `docs/ceo-plans/` inside the current repo, so they are committed alongside the code they describe and are readable by the whole team.

## Installation and Setup

The primary install path uses the skills.sh CLI:

```
npx skills add mrmps/better-gstack
```

Individual skills can be added selectively. Alternatively, the repo can be cloned and the skill files copied manually into `~/.claude/skills/`. The skill structure is two directories, each containing a `SKILL.md` and a `sections/` subdirectory for the longest review phase.

## Current Status

The repository was created in August 2026 and had 27 stars and 0 forks at last update. It is actively maintained under the MIT license, with original skill content credited to Garry Tan and modifications credited to Michael Ryaboy. The project welcomes contributions from founders, operators, angels, and partners who have pattern-matched across many companies — specifically skills with proprietary insight that a model would not produce on its own.

## Features
- office-hours: YC-partner-style pre-build product diagnostic
- plan-ceo-review: 11-section plan scope review with four postures
- Writes design docs to docs/design/ and docs/ceo-plans/ in current repo
- Stage-adaptive questioning (pre-product vs paying customers)
- Side project / hackathon / open source mode for office-hours
- Explicit opt-in for every scope change in plan-ceo-review
- Works with any agent that reads SKILL.md files
- Installable via npx skills add or manual clone
- No state directory, telemetry, or session tracking
- MIT licensed, ~1400 lines across four markdown files

## Integrations
skills.sh CLI, Claude Code, Any AI agent supporting SKILL.md

## Platforms
WEB, API, CLI

## Pricing
Open Source

## Version
main

## Links
- Website: https://github.com/mrmps/better-gstack
- Repository: https://github.com/mrmps/better-gstack
- EveryDev.ai: https://www.everydev.ai/tools/better-gstack