G Stack by Garry Tan: 23 AI Specialists and 8 Power Tools for Your Coding Agent
G Stack turns Claude Code into a virtual engineering team with 23 specialists including CEO, QA lead, and security officer. Here's how to install and use it.
What G Stack Actually Is (And Why It Went Viral)
When Garry Tan, CEO of Y Combinator, shared his personal Claude Code configuration publicly, developers paid attention. Not because it was complicated — but because it was clever.
G Stack is a CLAUDE.md configuration file that transforms Claude Code from a single AI assistant into something that behaves more like a full engineering team. It defines 23 specialist personas and 8 power tools that Claude can draw from depending on what your project needs. You drop one file into your project root and your Claude workflows suddenly have access to a virtual CEO, QA lead, security officer, and 20 other domain experts.
This article breaks down exactly how G Stack works, what the 23 specialists do, how to install it, and when to reach for it.
The Core Idea: One Agent, Many Hats
Most people use Claude Code as a capable generalist — paste some context, ask a question, get an answer. That works fine for quick tasks. But large codebases and complex builds need more than a generalist. They need someone who thinks like a security engineer when reviewing authentication logic, like a QA lead when writing test coverage, and like a database architect when designing schemas.
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
G Stack solves this by giving Claude Code a structured identity system. Each specialist is defined in the CLAUDE.md file with a specific role, focus area, and thinking style. Claude doesn’t switch between literal AI models — it switches between defined perspectives and priorities encoded in your system prompt.
The result: you get contextually appropriate responses depending on the task at hand, without managing multiple AI tools or writing long prompts from scratch every time.
The 23 Specialists: What Each One Covers
The 23 specialists in G Stack cover the full arc of a software project — from high-level strategy to line-by-line code review. Here’s a breakdown of the categories:
Executive and Strategic Roles
- CEO/Founder — Thinks about product direction, prioritization, and what actually ships. Useful for scope decisions.
- CTO — Focuses on technical strategy, architectural choices, and long-term maintainability.
- Product Manager — Translates business requirements into technical specs. Flags when engineering is over-building.
Engineering Specialists
- Senior Full-Stack Engineer — General-purpose implementation, best practices, code quality.
- Frontend Specialist — UI logic, component architecture, accessibility, performance.
- Backend Specialist — API design, service architecture, data flow.
- API Architect — Focuses specifically on interface contracts, versioning, and inter-service communication.
- System Architect — High-level structural design, scalability patterns, dependency management.
- Performance Engineer — Identifies bottlenecks, optimizes hot paths, thinks about load and latency.
- Database Administrator — Schema design, indexing strategies, query optimization.
- DevOps/Infrastructure Engineer — CI/CD pipelines, containerization, environment configuration.
- Site Reliability Engineer (SRE) — Uptime, monitoring, incident response patterns.
- Machine Learning Engineer — Model integration, data pipelines, inference optimization.
Quality and Safety Roles
- QA Lead — Test strategy, edge cases, regression risk, coverage gaps.
- Security Officer — Threat modeling, vulnerability identification, authentication review, dependency audits.
- Compliance Officer — Regulatory considerations, data handling, audit trails.
- Code Reviewer — Line-level feedback, readability, convention consistency.
- Debugging Specialist — Root cause analysis, tracing, systematic isolation of issues.
Support and Communication Roles
- UX/UI Designer — User flow logic, interface clarity, interaction patterns.
- Data Engineer — ETL logic, data modeling, pipeline reliability.
- Technical Writer — Documentation quality, README clarity, inline comment standards.
- Agile Coach — Sprint planning support, story sizing, velocity thinking.
- Documentation Specialist — API docs, architecture decision records, onboarding materials.
Not every project needs all 23. The value is that they’re available — when you need a security-minded review, you can invoke that frame explicitly instead of hoping a generalist catches what an expert would.
The 8 Power Tools
Beyond the specialist personas, G Stack defines 8 slash commands (power tools) that give you structured workflows for common development tasks. These are defined directly in the CLAUDE.md file and can be invoked during any Claude Code session.
- /plan — Generate an implementation plan before writing code. Breaks down the task, flags dependencies, and identifies risks upfront.
- /review — Triggers a structured code review using the Code Reviewer and Security Officer lenses simultaneously.
- /debug — Activates the Debugging Specialist persona for systematic root cause analysis.
- /test — Invokes the QA Lead to generate test cases, identify edge cases, and flag coverage gaps.
- /security — Runs a security-focused audit on specified code. The Security Officer and Compliance Officer perspectives both apply.
- /refactor — Initiates a structured refactoring pass with an eye on readability, performance, and long-term maintainability.
- /docs — Generates or updates documentation, README content, inline comments, or API references.
- /deploy — Produces a deployment checklist through the DevOps and SRE lenses, covering environment config, rollback plans, and monitoring hooks.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Each tool is essentially a pre-written prompt structure that ensures Claude approaches the task with the right frame. You’re not losing flexibility — you’re adding consistency.
How to Install G Stack
Installation is intentionally simple. G Stack is delivered as a single CLAUDE.md file.
Step 1: Get the File
Garry Tan’s G Stack configuration is publicly available on GitHub (search for the repository directly). Copy the raw contents of the CLAUDE.md file.
Step 2: Place It in Your Project Root
your-project/
├── CLAUDE.md ← paste it here
├── src/
├── package.json
└── ...Claude Code automatically reads CLAUDE.md files in the project directory when you start a session. No additional configuration required.
Step 3: Verify It’s Loading
Start a Claude Code session in your project and ask: “What specialists do you have available?” Claude should respond with the list of 23 roles defined in your CLAUDE.md.
Step 4: Customize for Your Stack
The real power comes from editing the CLAUDE.md to match your project. Garry’s version is a starting point. Modify specialist definitions to reflect your tech stack (e.g., give the Backend Specialist knowledge of your specific framework), add project-specific conventions, and remove roles you’ll never use.
How to Use It in Practice
Having the configuration in place is one thing. Using it well is another.
Invoking a Specialist Directly
You can explicitly call on a specialist in your prompt:
“As the Security Officer, review this authentication middleware for JWT validation issues.”
This signals to Claude which lens to apply. The response will prioritize security-relevant concerns rather than general code quality.
Using Power Tools in Sequence
A common workflow for new features looks like:
/plan— get a structured implementation plan- Build the feature with the Senior Engineer perspective active
/test— generate tests and catch edge cases/security— audit the new code/docs— update documentation
This mirrors how a real engineering team would work through a feature — just compressed into one tool.
Combining Perspectives
Some tasks benefit from multiple specialists. Ask for it explicitly:
“Review this database migration script as both the DBA and the DevOps Engineer. Flag risks from both perspectives.”
Claude will structure its response to address both angles.
Common Mistakes to Avoid
Don’t invoke every specialist for every task. The system adds value through specificity. Using a generalist prompt when you have specialists available wastes the configuration.
Don’t skip the /plan step. It’s tempting to jump straight to implementation, but the planning pass often catches scope issues or architectural conflicts before they become expensive.
Do keep the CLAUDE.md updated. As your project evolves, update the specialist definitions to reflect new conventions, new tech, and lessons learned. A stale CLAUDE.md is worse than no CLAUDE.md.
Why the CLAUDE.md Approach Works
Claude Code’s instruction system is straightforward: whatever is in CLAUDE.md gets loaded as persistent context for every session in that directory. This makes it a powerful configuration surface that most developers barely use.
G Stack is a demonstration of what’s possible when you treat CLAUDE.md as a first-class engineering asset — not just a place to put “always use TypeScript” but a place to encode team knowledge, role definitions, and process structure.
The underlying mechanism maps to how Anthropic describes Claude’s contextual behavior — Claude performs better when given structured roles and clear task framing rather than vague open-ended instructions.
This is also why G Stack can be adapted. The 23 specialists aren’t sacred. They’re defaults based on Garry Tan’s experience building and funding startups. Your project might need a Blockchain Specialist or a Mobile Platform Engineer instead of an Agile Coach. The template is the pattern, not the specific roles.
Where MindStudio Fits for Teams Who Want This Without Code
G Stack is a great approach if you’re already using Claude Code directly. But not every team works that way — and not every AI automation use case starts from a terminal.
MindStudio offers a no-code platform where you can build the same kind of specialist-routing logic into visual AI workflows. Instead of encoding 23 personas in a CLAUDE.md file, you build an orchestration layer where different AI agents — each configured with a specific role and system prompt — handle tasks based on what’s being asked.
The Agent Skills Plugin is particularly relevant here. If you’re building a Claude Code agent or any custom coding assistant, MindStudio’s npm SDK (@mindstudio-ai/agent) lets your agent call 120+ typed capabilities as simple method calls — handling rate limiting, retries, and auth in the background. Your agent focuses on reasoning; MindStudio handles the infrastructure plumbing.
For teams building AI-powered automation workflows that span more than just code generation — including document handling, email triggers, Slack notifications, and API integrations — MindStudio gives you the specialist-routing model G Stack demonstrates, extended to the full range of business operations.
You can start building free at mindstudio.ai.
FAQ
What is G Stack?
G Stack is a CLAUDE.md configuration created by Garry Tan, CEO of Y Combinator. It installs 23 AI specialist personas and 8 slash-command power tools into Claude Code, turning it into a multi-role AI engineering team. You install it by placing the CLAUDE.md file in your project root directory.
What is CLAUDE.md?
CLAUDE.md is a markdown file that Claude Code reads automatically when starting a session in a given directory. It works as a persistent system prompt — any instructions, conventions, or persona definitions you place there apply to every interaction in that project. G Stack uses this file to define its 23 specialists and 8 tools.
Do you need to pay for Claude to use G Stack?
You need access to Claude Code, which is available through Anthropic’s platform. Claude Code itself requires an Anthropic account, and usage is billed based on tokens. The G Stack configuration file itself is free and open source.
Can you customize G Stack for your own project?
Yes — and you should. G Stack is a starting template. The 23 specialists and 8 tools are well-designed defaults, but every project has different needs. You can add specialists for your specific domain (e.g., iOS Engineer, Blockchain Developer, Data Scientist), update existing role definitions to match your tech stack, and remove roles that don’t apply.
How is G Stack different from just writing a good system prompt?
G Stack is more systematic than a one-time system prompt. It gives Claude a structured vocabulary of roles to draw from, defines consistent behaviors for each role, and provides repeatable tools (slash commands) for common tasks. This makes it useful over the lifetime of a project — not just for a single session.
Can G Stack be used with AI tools other than Claude Code?
The CLAUDE.md format is specific to Claude Code, but the underlying approach — defining specialist personas and structured workflows in a system prompt — can be adapted to any AI coding assistant or agent framework that supports persistent system instructions. The pattern is broadly applicable even if the exact file format isn’t.
Key Takeaways
- G Stack is a CLAUDE.md configuration by Garry Tan that gives Claude Code 23 specialist personas and 8 power tools.
- Installation takes under a minute: drop the file in your project root and Claude Code loads it automatically.
- The 23 specialists cover the full engineering lifecycle — from architecture and security to QA and documentation.
- The 8 power tools (/plan, /review, /debug, /test, /security, /refactor, /docs, /deploy) give you structured workflows for common tasks.
- The real value comes from customizing the configuration to your project’s specific stack and conventions.
- For teams who want specialist-routing logic without writing CLAUDE.md files from scratch, MindStudio offers a visual alternative for building multi-agent AI workflows that span the full scope of a project — from code generation to business process automation.
