Garry Tan Open-Sources His Exact Claude Code Setup: 23 Agents Serving as CEO, Designer, Eng Manager, QA, and More

Garry Tan Open-Sources His Exact Claude Code Setup: 23 Agents Serving as CEO, Designer, Eng Manager, QA, and More

Garry Tan — President and CEO of Y Combinator — published gstack to GitHub this morning: his complete, production Claude Code configuration, MIT licensed and ready to drop into ~/.claude/skills/. The repo contains 23 role-based agents organized by function — CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA — each packaged as a CLAUDE.md-compatible skill with slash commands, hooks, and subagent delegations. The README opens with the context that makes this credible: Tan has shipped 600,000+ lines of production code in 60 days, 10,000–20,000 lines per day, part-time. "I don't think I've typed like a line of code probably since December" — gstack is the how behind that claim. The repo is already trending with 210 HN points and 119 comments within hours of publishing, and the HN thread is dominated by founders sharing modified versions.

What makes gstack worth actual study rather than just starring: these aren't toy examples. The Eng Manager agent alone — covering sprint planning, code review routing, and incident triage — represents weeks of CLAUDE.md design work that teams would otherwise have to build themselves. The Release Manager agent handles changelog generation, version tagging, and deploy sequencing. Each agent is wired with specific subagent delegations so the right tool tier (Haiku vs. Sonnet vs. Opus) handles each task type, which matters significantly for cost management in high-volume founder workflows. Peter Steinberger's OpenClaw is cited as inspiration in the README.

For founders and small engineering teams: this is the most credible "solo developer shipping at team scale" proof point published to date, backed by specific output numbers from someone who has worked with thousands of startups. The 23 agents are a canonical starting point that the community will fork and improve rapidly — now is the right time to grab a copy and adapt it to your stack before the conventions solidify around someone else's opinions.

Read the full repo at GitHub (Garry Tan) →