Beginner's Guide
to GStack

Welcome to the "Team of One." This guide will help you install and run GStack, the leading role-based agentic framework. This isn't just a coding tool; it’s an automated factory for your product.

Phase 0: Ground Zero (Installation)

Open your terminal. We are going to "hire" your entire staff in about sixty seconds.

Clone the Framework:

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

Onboard your Project: Navigate to your own project folder and link GStack:

cd path/to/your-startup

Phase 1: The "Lazy" Constitution (CLAUDE.md)

In GStack, you don't write your company's technical rules; you interview for them.

Action: Run /office-hours.

The Example: * Agent (CEO Persona): "I see you want to build a Fitness App. Who is the target user? Are we doing mobile-first or web? What’s the tech stack?" * User: "It's for powerlifters and heavy lifters. Web-first using Next.js, Tailwind, and Supabase. Keep it minimalist, high-contrast (gym-friendly), and fast."

The Command: After the chat, say: "Update my CLAUDE.md based on this conversation."

The Result: The AI generates a file that explicitly enforces your "High-contrast" and "Fast" philosophy in every code block it writes, ensuring large, tappable buttons for sweaty hands.

Phase 2: Design First (/design-consultation)

Before building, you need a "Visual Law." This prevents "AI Slop" and ugly UI.

Action: Run /design-consultation.

• The Example: "Design a 'Rugged Fitness' dashboard. It needs to look like a high-performance tool, dark mode by default, with neon accents for Personal Records (PRs)."
• The Output: The agent creates a DESIGN.md file and a Style Guide Preview Page.

The Impact: When the Engineering Agent later builds a "Log Workout" button, it must pull the padding, border-radius, and hex code directly from this DESIGN.md.

Phase 3: The Execution Loop

This is the standard rhythm you will follow to build any new feature. Let's look at a "Max PR Tracker" feature as an example:

1. Product Planning (/plan-ceo-review):
   • Focus: Proposing the idea and validating the business logic. This is where you pitch the "Vibe" and the high-level feature set.
   • Result: The agent lists the user-facing changes and product flow. You approve the vision.
2. Technical Planning (/plan-eng-review):
   • Focus: Feasibility and architecture. The "Engineering Manager" critiques the product plan, identifies technical debt, and locks in the implementation strategy.
   • Result: A technical roadmap including database schemas, API route logic, and potential edge cases.
3. Build (/engineer):
   • Example: The agent writes the PRCard.tsx component and the Supabase query. It automatically uses the "Electric Lime" color defined in your DESIGN.md.
4. Test (/qa):
   • Example: GStack launches an advanced Visual QA Agent that has the visual context of a browser powered by a series of "diffs" sent at highly-frequent intervals, reacting to the page in real-time as the cursor moves and clicks to verify a set of requirements.
5. Audit (/review):
   • Result: The Review Agent flags that the PR text isn't using the "Heavy Bold" font weight. It forces the Engineering Agent to fix the styling before you ever see the code.
6. Deploy (/ship):
   • Result: The Release Lead writes: "Added PR tracking logic. Verified real-time dashboard updates. Ready to merge."

Uniquely GStack: The LLM-Powered Visual QA Agent (Browser Daemon)

While specialized agents for planning, engineering, and shipping are becoming standard in most agentic frameworks, the unique differentiator of GStack is its enhanced perceptual awareness.

The core engine driving this is the LLM-powered Visual QA Agent (aka Browser Daemon). Unlike standard test scripts that only "read" the DOM, this is a vision-aware autonomous observer. It has real-time context of your specific browser, allowing it to see and react to the UI exactly like a human would—moving the cursor, clicking elements, and verifying the "vibe" as it happens.

Key Features

• GStack Browser Server: A specialized wrapper around Playwright/Chromium that manages the persistent Session Lifecycle, intercepts the render pipeline to inject diagnostics and cursor rendering, and handles complex authentication flows.
• Real-time Visual Reactive Testing: The agent doesn't just check the DOM; it "watches" the screen for layout shifts, color inaccuracies, and flickering in real-time.
• Cursor Overlay: During execution, you can watch the AI's "thought process" via a virtual cursor overlay that highlights elements before the agent interacts with them.
• Automatic Authentication & Anti-Bot: By leveraging your active browser session, the agent bypasses 2FA, Captchas, and login walls seamlessly by inheriting your existing browser cookies.

How It Works Under the Hood

GStack uses a proprietary "perceptual relay" to keep the agent in sync without burning through thousands of tokens:

Optimal Use Cases

• Visual Regression: Catching accidental CSS breaks in complex, nested components.
• Multi-step Onboarding: Testing long, state-dependent user journeys that are usually too brittle for standard Selenium or Playwright scripts.
• Design Audit: Ensuring that the final rendered output matches the DESIGN.md "Visual Law" with 100% fidelity.

Pro-Tips for Solo Success

• Real-Time Context: Use /setup-browser-cookies early. It allows the AI to "see" through your eyes, meaning it won't get stuck at login screens during testing.
• Deterministic Backup: For your most critical paths (like User Signup or Payment Processing), use GStack to generate Playwright test scripts. This gives you a permanent, script-based "Safety Net."

Conclusion: The Implications of the Agentic Shift

The emergence of frameworks like GStack isn't just a technical upgrade; it is a fundamental restructuring of the tech industry.

1. The Hyper-Productive Founder

The most immediate impact is a massive surge in individual productivity. Tasks that previously required a weekly sync can now be completed in a single "Loop." This collapses the time-to-market from months to days.

2. From Task to Job

We are witnessing a shift in what it means to be a "tech worker."

Yes many people's work around doing "tasks" will be reduced. They will increasingly need to move up to do the higher level "job" of what's needed, leaving the "tasks" to AI to do.

This will "free" up tech workers, but in reality this means they will simply be asked to handle broader scope and spend more of their time doing QA, checking the work of AI for meeting their quality bar.

3. Emerging Chasm

The entire AI Transformation of the workplace will empower CEOs/Stakeholders and increase the need for QA (to check for AI Slop). In between, however, will emerge a chasm. Lower-level CEOs (call them entry/mid-level PMs) will no longer be needed. The same for lower-level designers and engineers, etc.

In their place will be a small cadre of "Principal-level" Orchestrators, who can lead AI teams to build products faster and better than ever before. They will be the most valuable people in the world.

Folks who are in the bottom of the pipeline -- in QA -- are still vital as they will need to test for quality. However, being able to "graduate" higher up will be increasingly difficult as the chasm will not make that bridge smooth or easy. That gap is emerging in for all industries and room at the top is limited. The only other way to the top is for those left behind to form their own startups -- to solve new problems. And that is the real test -- how many employees displaced by AI will be able to find new problems to solve in the world -- and create their own startup. That is the question.

4. Product is NOT the bottleneck anymore

Legacy startups raised millions to hire engineers for basic infrastructure. The agentic founder can now reach Product-Market Fit (PMF) with nearly zero overhead. It used to take months to get engineers to build anything substantial. That is now turned to days/weeks.

The new bottleneck is around getting customers -- business development, customer acquistiion, sales, marketing. While AI will transform these areas too, they are strongly protected against AI automation. You will need humans to reach out to potential customers to get feedback to refine the product. You will need humans on the other side to agree with their company to spend money for your product. You will still be selling to a human decision maker -- not an AI bot. Driving engagement and revenue will increasingly be held up by the human element.

Those who excel at communication, relationship building, and dare I say it, charm and attracion -- will be the next set of valued human employees.

The "Team of One" is here -- and is transforming your workplace. Upskill today.