What you get

From a vague idea to a browser-tested release, every gate has a specialist and carries the previous gate's context.

gstack: An End-to-End Delivery Pipeline for AI Coding

Added 2026-08-10

Garry Tan's open-source AI engineering workflow turns product interrogation, engineering and design review, implementation, code review, real-browser QA, security, release and retrospectives into specialist skills that hand work forward. It supports Claude Code, Codex and other agents.

EngineeringProduct & Project
gstack: An End-to-End Delivery Pipeline for AI Coding preview

It treats Think → Plan → Build → Review → Test → Ship → Reflect as one sprint instead of letting an agent jump from a vague request straight into code. office-hours reframes the problem; the plan skills lock product, architecture, design and developer experience; review and investigate inspect code and root causes; qa verifies the product in a real browser; cso runs security review; ship and land-and-deploy handle delivery; and retro feeds lessons into the next round. Design docs, test plans and review findings become inputs to the next gate. The installer adapts gstack for Claude Code, Codex, Cursor, OpenCode and other hosts. Telemetry is off by default; if you opt in, the limited anonymous run data excludes code, file paths, repository names and prompts.