gstack: AI Software Engineering Stack
AI coding agents can already write functions, modify repositories, run tests, and open pull requests. The harder problem is getting an agent to follow a repeatable engineering process before, during, and after the code is written. gstack, a project from Garry Tan originally built around Claude Code, takes a different route: instead of replacing your coding agent with another platform, it wraps…
gstack is an open-source AI software engineering workflow system originally built around Claude Code. It aims to help software development teams follow repeatable processes before, during, and after code writing. Rather than replacing existing coding agents, gstack wraps them with specialized skills, browser tooling, reviews, safety controls, and release processes.
The system is described as a virtual engineering team with 23 specialists and eight power tools, all accessible via slash commands in Markdown format and licensed under MIT.
gstack's capabilities stem from Markdown skill definitions. Each skill defines when to run, what context to inspect, what questions to ask, which tools to use, which commands to execute, what evidence to gather, which checks must pass, and how to structure the results. These skills function as documentation, reusable prompts, standard operating procedures, and executable workflow configurations.
The system presents these skills as a virtual engineering team: a CEO, an engineering manager, a designer, a reviewer, a QA lead, a security officer, and a release engineer. This structure allows each skill to hand its artifacts to the next in the order that a typical sprint runs, ensuring a structured approach to software development.
gstack is not a foundation model or a replacement for Claude Code; instead, it is a process layer running on top of an agent harness. The underlying mechanics involve TypeScript and Bun, a setup script, generated skill documentation, session hooks, state management under ~/.gstack/, a bundled browser, and standalone CLIs. The system is not limited to Claude Code, as it can detect various agents installed on the machine, including Codex CLI, OpenCode, Cursor, Factory Droid, Kiro, Slate, OpenClaw, and Hermes.
The project emphasizes discipline and adherence to processes, moving from idea to coding agent to code through stages such as planning, architecture review, design review, code review, browser QA, security analysis, release and deployment, and learning and memory. By challenging assumptions earlier, inspecting evidence, reviewing work from multiple perspectives, and verifying the application, gstack aims to change the probability distribution of AI mistakes.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
