Y Combinator CEO Garry Tan has open-sourced gstack—his personal collection of slash commands and skills for Claude Code. Instead of using Claude as a single, general-purpose assistant, gstack structures it into a virtual team with specialized roles, each invoked by a specific command. The repository gained over 56,000 stars almost immediately, signaling strong developer interest in this structured, multi-agent approach to coding.
What It Does — A Team in Your Terminal
gstack is essentially a configuration and prompt pack for Claude Code. It defines a set of roles, each with a specific responsibility and corresponding slash command (/). The core idea is to prevent context dilution and task-switching penalties by giving Claude a clear, singular focus for each interaction.
Here are the key roles and their commands:
/ceo: Focuses on product strategy, vision, and high-level planning./em(Engineering Manager): Provides architecture guardrails and technical oversight./designer: Reviews UI/UX, catches "AI slop," and improves user experience./reviewer&/qa: Finds bugs, suggests improvements, and can run tests in a real browser context./security: Conducts OWASP and STRIDE-based security audits./release: Handles release engineering, versioning, and deployment logic./retro,/doc: Facilitates retrospectives and documentation.
Tan claims this structured approach enables him to ship 10,000–20,000 lines of code per day while managing YC.
Setup — How to Install and Configure
Installation is straightforward via the Claude Code CLI. First, clone the repository:
git clone https://github.com/garrytan/gstack.git
cd gstack
Next, run the installation script. This will configure Claude Code to recognize the gstack slash commands by modifying your local Claude Code settings or CLAUDE.md file.
./install.sh
The script typically adds role definitions and system prompts to your Claude Code configuration, making the new commands available in your terminal session. You may need to restart your Claude Code terminal or source the new configuration.
When To Use It — Specific Workflows That Shine
gstack is not for every single-line edit. Its power is unlocked during larger, more complex coding sessions where different phases of work benefit from specialized focus.
- Starting a New Feature or Project: Use
/ceoto define the scope and vision, then immediately switch to/emto lay down the initial architecture and file structure. This separates the "what" from the "how" cleanly. - The Review & Audit Cycle: After writing a chunk of code, chain the commands. Run
/reviewerfor logic and bug checks, then/securityfor vulnerability analysis, and finally/designerfor any front-end polish. This creates a machine-driven QA pipeline. - Pre-Release Preparation: Use
/releaseto generate changelogs, bump versions, and check deployment readiness, then/docto ensure any new features are properly documented.
This follows a broader trend in the Claude Code ecosystem towards structured, multi-agent workflows, as seen in tools like terraphim-agent for verification sweeps, which we covered recently. The core value is enforcing a separation of concerns, which reduces prompt confusion and leads to more deterministic outputs from the AI.
gentic.news Analysis
Garry Tan's release of gstack directly taps into two major, converging trends for Claude Code. First, it exemplifies the move beyond a monolithic assistant towards a multi-agent framework, a concept core to Anthropic's own Claude Agent. This aligns with our recent coverage on adding machine-enforced rules with terraphim-agent—developers are increasingly seeking ways to compartmentalize AI tasks for better results and control.
Second, its viral adoption (56k+ stars) underscores a critical shift in developer mindset. Following incidents like the autonomous git reset --hard command that erased work, there's heightened demand for guardrails and structured processes. gstack provides this by design, embedding roles like "Security Officer" and "Engineering Manager" into the workflow. It’s a community-driven response to the need for safer, more predictable AI collaboration.
This development also highlights the growing importance of the Model Context Protocol (MCP), the architecture Claude Code is built upon. Packs like gstack are essentially sophisticated MCP configurations, showing how the protocol enables deep customization. As Claude Code's usage spikes—it appeared in 153 articles this week alone—the ecosystem around it, from MCP servers to skill packs, is becoming the real battleground for developer productivity.
