Getting Started

Templates and starter kits for Claude Code

  • Tutorial
  • Tooling

Tour of the opinionated stacks of May 2026: gstack (Garry Tan), claude-code-templates, SuperClaude_Framework, get-shit-done, claude-on-rails. How to choose by profile.

TL;DR

  • Installing Claude Code "bare" gives you a capable but generic agent. A starter kit adds skills, agents, commands and hooks tailored to a real workflow.
  • 4 starter kits dominate in May 2026: gstack (Garry Tan, 23 opinionated tools running "team-of-20" style), claude-code-templates (catalog of 100+ components), SuperClaude_Framework (30 commands + 20 cognitive agents), get-shit-done / GSD (meta-prompting and sub-agent orchestration).
  • Pick by need: follow a tech leader → gstack ; pick à la carte → claude-code-templates ; full framework for a team → SuperClaude ; fight context rot solo → GSD.
  • All MIT-licensed. You can (and should) mix them.

Why use a template?

Vanilla Claude Code edits code, runs commands, makes commits. Enough for a one-off session. But on a project that lasts:

  1. You end up rewriting the same long prompts (code review, implementation plan, security audit)
  2. You manually re-do the same hooks (format before commit, block rm -rf)
  3. You reinvent the same agents (PM, designer, QA)
  4. You lose context across sessions

A starter kit ships all that packaged. Three main styles:

  • Skills + slash commands, gstack-style: /plan-ceo-review triggers a dedicated agent
  • Catalog à la carte, claude-code-templates-style: cherry-pick components you want
  • Integrated framework, SuperClaude or GSD style: the whole ecosystem comes together

The 4 must-have starter kits

gstack — Garry Tan's Claude Code setup

The starter kit that went viral in 2026. Garry Tan (Y Combinator) published his personal setup to help a solo founder "function like a team of 20".

What you get: 23 slash commands that embody full-team roles. Planning (/plan-ceo-review, /plan-eng-review, /plan-design-review), design (/design-consultation, /design-shotgun, /design-review), execution (/review, /ship, /qa, /land-and-deploy), security (/cso), polish (/freeze, /guard, /careful), research (/browse, /investigate), debug (/codex).

git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack
cd ~/.claude/skills/gstack
./setup

Best for: technical founder solo, staff engineer who wants an opinionated workflow, anyone who buys the "one human, many agents" philosophy. Less suited if you'd rather compose each tool yourself.

claude-code-templates — À la carte catalog (davila7)

Opposite approach: instead of a monolithic stack, a catalog of 100+ components you cherry-pick from.

What you get: domain-specialized agents (security audit, performance, etc.), custom slash commands, MCP integrations (GitHub, PostgreSQL, AWS, etc.), opinionated settings, automation hooks, skills with progressive disclosure.

# Interactive: choose what to install
npx claude-code-templates@latest


# Direct: install one specific agent
npx claude-code-templates@latest --agent security-auditor


# Install everything without asking
npx claude-code-templates@latest --yes

Best for: devs who want a library, cherry-picking per project, or testing an agent before adoption. Good fit for cautious profiles who don't want to ship 23 tools at once.

SuperClaude_Framework — Meta-programming framework

Framework-oriented approach: behavioral injection + component orchestration. The project positions itself as a "meta-programming configuration framework".

What you get: 30 slash commands covering the full dev cycle (brainstorming → deployment), 20 specialized cognitive agents (PM, security engineer, frontend architect, etc.), 8 MCP integrations, 7 behavioral modes that adapt to context.

pipx install superclaude
superclaude install

Best for: teams that want a cohesive framework, long-running projects with agent rotation, profiles who like structured frameworks (Spring Boot, Django…). Less suited if you'd rather add your own conventions one at a time.

get-shit-done (GSD) — Fight context rot

Minimalist approach centered on sub-agent orchestration: isolate heavy research into dedicated 200K-token contexts, keep the main window clean.

What you get: 6 commands orchestrating a research → plan → execute → verify cycle, persisted artifacts (PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md) that maintain context across sessions, multi-agent with fresh 200K contexts, interactive or autonomous ("yolo") mode.

npx get-shit-done-cc@latest

Compatible not only with Claude Code but also Cursor, Windsurf, and 15+ other runtimes. Best for: very long sessions (> 100K tokens), solo dev on a recurring product, anyone tired of /compact-ing every hour. See /advanced/memoire-persistante for the underlying pattern.

Honorable mention: framework-specific starter kits

claude-on-rails — For Ruby on Rails devs

The ecosystem is starting to spawn per-framework forks (Rails, Django, Next.js…). If you work in a specific stack, check first whether a dedicated starter exists before picking a generic kit.

Comparison table by use case

ProfileRecommendedWhy
Solo founder, "team of 20" in one persongstackOpinionated workflow, 23 roles already wired
Senior dev who picks à la carteclaude-code-templates100+ component catalog, targeted install
Team that needs consistencySuperClaude_FrameworkStructured framework, 20 shared cognitive agents
Very long sessions (context rot)get-shit-done200K sub-agents + persistent artifacts
Rails / Django / framework-specific devDedicated fork (e.g. claude-on-rails)Framework conventions integrated
Just want to test one featureNone, or single skillNo need to install 100 components for a test

How to choose

Three questions to ask:

  1. Do you have a clear workflow already? Yes (release process, structured code review) → gstack or SuperClaude. No → claude-code-templates à la carte.
  2. Solo or team? Solo → gstack or GSD (light). Team → SuperClaude (shared framework).
  3. Short or long sessions? Short (< 2h) → any kit. Long → GSD first for the sub-agent pattern.

Useful links

Previous

Frequently asked questions for beginners