GStack is Garry Tan's (Y Combinator President & CEO) open-source Claude Code harness, released March 2026. It installs 23 specialist skills into Claude Code — turning a single AI coding assistant into a virtual engineering team that interrogates product decisions, reviews architecture, audits security, and runs QA before every merge. The headline feature is Conductor: a multi-agent orchestrator that runs multiple Claude Code sessions in parallel isolated workspaces, each with its own git worktree and context window. 82,700 stars and 12,000 forks in its first six weeks make it one of the fastest-growing developer tools in 2026. MIT licensed, installs via a single paste into Claude Code, and includes a team install option that adds the skills to a repo's CLAUDE.md so all teammates get the same setup.
GStack addresses a structural problem in solo and small-team development: there is no senior reviewer in the room. A developer building alone ships code without a security audit, architecture challenge, or QA pass — not because they are negligent, but because those reviewers don’t exist. GStack gives Claude Code the specialist personas that fill those roles.
The install takes 30 seconds. Open Claude Code and type Install GStack. One more paste installs the skills into your repo’s CLAUDE.md for the whole team.
Planning skills (/office-hours, /plan-ceo-review, /plan-eng-review, /autoplan) run before any implementation. /office-hours is intentionally adversarial — it challenges your idea with forcing questions before you write a line of code. /autoplan chains CEO, Engineering, and DevEx review in one command.
Security (/cso) applies OWASP Top 10 and STRIDE threat modeling as a mandatory step before /ship. Garry Tan shared a CTO testimonial where the skill caught a subtle XSS vulnerability the team was unaware of.
Quality (/qa, /qa-only) runs live browser testing. /qa applies fixes inline; /qa-only produces a bug report for developer review.
Release (/ship, /land-and-deploy, /document-release) handles PR creation, merge, deployment verification, and changelog updates.
Conductor runs multiple Claude Code sessions simultaneously in isolated workspaces. One session can be doing /office-hours on a new feature idea while another does /review on an open PR and a third is implementing a different branch. Each session gets its own git worktree, context window, and task scope — no cross-contamination between workstreams.
Garry Tan reports his 2026 development pace at approximately 810× his 2013 baseline (11,417 logical lines/day vs 14). This measures “logical LOC” — meaningful code changes, not raw lines — and represents a single developer’s experience building product software. Independent developers report genuine productivity improvements, particularly from the review-by-default pattern GStack enforces.
Claude Code users building product software who want structured review, security auditing, and QA without a dedicated team. Especially valuable for solo developers and early-stage startups where the default workflow skips review steps under deadline pressure.
Not recommended for teams that need multi-model routing (see oh-my-openagent) or developers working primarily on embedded systems or highly regulated domains where the web-product-oriented skills don’t apply.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.
Open-source AI pair programming tool that works in your terminal to edit code across your entire repository.
AWS's AI-powered coding assistant that helps developers build, deploy, and optimize applications on AWS with code generation and transformation.