Wire StackPicks to your AI agent.
Install once, then point any agent at a bundle or an individual repo. The agent reads our curator takes and writes the integration.
Prerequisites (one-time)
You need three things on either OS: Node.js 20+, git, and a terminal. The install commands below are copy-paste-safe.
# Install Homebrew if you don't have it /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" # Then: brew install node@20 git pnpm
# Install winget if missing (Windows 11 has it by default) winget install --id OpenJS.NodeJS.LTS winget install --id Git.Git winget install --id pnpm.pnpm
Tip: on Windows, prefer WSL 2 + Ubuntu for a smoother dev loop. Then follow the Mac instructions inside WSL.
wsl --install -d Ubuntu
Verify after installing:
node -v # should print v20.x git --version pnpm -v
Claude Code
Terminal-first agent from Anthropic. Best at large-codebase reasoning and end-to-end tasks.
curl -fsSL https://claude.ai/install.sh | bash # OR via npm: npm install -g @anthropic-ai/claude-code # Then: claude --version
# Recommended path: install in WSL Ubuntu (see step 0) # Then in WSL: curl -fsSL https://claude.ai/install.sh | bash # Native Windows (PowerShell): npm install -g @anthropic-ai/claude-code
claude # opens a browser to log in once, then keeps the session
cd ~/projects/my-saas claude
The terminal becomes an interactive agent. It reads files, runs commands, edits code.
Inside the Claude prompt, paste:
I'm building a SaaS. Use this bundle as the source of truth for the stack: https://stackpicksdev-production.up.railway.app/build/ship-a-saas Set up Next.js 15 + shadcn/ui + better-auth + Supabase + Razorpay following the bundle order. Stop and confirm with me after each layer.
CLAUDE.md in the project root. Drop a one-page rules file (we keep one in the StackPicks repo). It will use the rules on every task.Cursor
VS Code-forked IDE with an AI panel. Best when you want a GUI and inline Tab autocomplete.
brew install --cask cursor
Or download the .dmg from cursor.com.
winget install --id Anysphere.Cursor
Open Cursor → settings → sign in. Pick claude-sonnet-4.5 or gpt-5 as the default model.
cd ~/projects/my-saas cursor .
Press ⌘L (Mac) orCtrl+L (Windows) to open the chat. Then:
@docs Add docs for shadcn-ui/ui (https://ui.shadcn.com) @docs Add docs for Razorpay (https://razorpay.com/docs) Now scaffold a Next.js app, install shadcn, wire Razorpay checkout per the docs above.
.cursor/rules directory. Create .cursor/rules/stack.mdwith your stack rules so every chat starts with context.OpenAI Codex
Browser-based agent that does long-running tasks on a virtual machine. Best for parallel work — fire off 10 PRs at once.
Codex is mainly cloud-based — no install needed. Go to chatgpt.com/codex.
For local CLI access:
npm install -g @openai/codex-cli codex login
npm install -g @openai/codex-cli codex login
On chatgpt.com/codex, click Settings → GitHub and authorize Codex on the repos you want it to act on.
Repo: jangirpiyush008-hash/my-saas Branch: feat/payments Task: Implement Razorpay subscription checkout. Reference the bundle: https://stackpicksdev-production.up.railway.app/build/ship-a-saas Use core/razorpay/ for the order + verification logic. Open a PR when done.
Google Antigravity
Agentic IDE from Google built around Gemini 3. Treats the IDE as a control surface for autonomous agents that drive editor, terminal, and browser.
# Download the macOS build from antigravity.google # Or via Homebrew once the cask lands: brew install --cask google-antigravity
Download the .exe installer from antigravity.google. Once installed, sign in with your Google account.
Antigravity has two surfaces: the IDE (familiar VS Code shell) and the Manager(where agents are dispatched, monitored, and review your work). Start in Manager.
In Manager, create a new task:
Build the auth layer for my SaaS. Reference stack: https://stackpicksdev-production.up.railway.app/build/ship-a-saas Specifically use: better-auth/better-auth + supabase/supabase. Open the browser tab on http://localhost:3000 to verify signup flow.
VS Code + AI extensions
The classic. Pair it with Copilot, Cline, or Continue for AI workflows.
brew install --cask visual-studio-code
winget install --id Microsoft.VisualStudioCode
Three solid choices, ranked by use case:
- GitHub Copilot — best inline autocomplete, $10/mo
- Cline — agentic, free + BYO API key, opens terminals and runs tools
- Continue — open-source alternative, BYO model
> Extensions: Install Extensions # Search and install: GitHub Copilot Cline Continue
Open Cline (left sidebar) and paste:
Read https://stackpicksdev-production.up.railway.app/build/ai-agent Scaffold a Next.js project, install pgvector locally via Docker, install Ollama, and create a /chat route that retrieves from pgvector and streams from Ollama.
Aider
Terminal AI pair-programmer. Edits files, commits to git automatically. Lightweight, model-agnostic.
brew install aider # Or via pip: python -m pip install aider-chat
python -m pip install aider-chat
# Anthropic: export ANTHROPIC_API_KEY=sk-ant-... # Or OpenAI: export OPENAI_API_KEY=sk-...
cd ~/projects/my-saas aider --model sonnet # or --model gpt-5
/add app/layout.tsx app/page.tsx Use https://stackpicksdev-production.up.railway.app/build/ship-a-saas as the canonical stack reference. Add a hero, pricing, and auth modal using shadcn/ui + better-auth.
How to feed StackPicks repos to any agent
Three patterns that work in every agent above:
Pattern A — paste a bundle URL
Reference stack: https://stackpicksdev-production.up.railway.app/build/ship-a-saas Implement the framework + UI + auth layers in that order.
The agent fetches the URL, reads the curator takes, picks the right repos. Works for any of the 12 bundles.
Pattern B — paste a single repo URL with the take
Use shadcn-ui/ui — see https://stackpicksdev-production.up.railway.app/preview/shadcn-ui/ui Install via npx shadcn@latest init. Add components: button, card, dialog, form, input.
Pattern C — clone & let the agent read locally
# Some bundles include self-hosted tools (Supabase, Medusa, Payload). # Clone them so your agent can read source directly: git clone https://github.com/supabase/supabase ~/refs/supabase git clone https://github.com/medusajs/medusa ~/refs/medusa # Then in your agent: # "Read ~/refs/supabase/apps/docs to understand the Auth API, # then implement auth in this project."
A real day-of-shipping flow
- Pick a bundle. Visit /build, open the one closest to what you're shipping (say, Ship a SaaS).
- Scaffold the project shell. In your terminal:terminal
pnpm create next-app@latest my-saas --typescript --tailwind --app cd my-saas git init && git add -A && git commit -m "Initial scaffold"
- Drop a CLAUDE.md / .cursor/rules / agents-md. Paste the bundle URL at the top so the agent always has stack context.
- Launch your agent of choice (sections 1-6 above) and ask it to implement layer 01 of the bundle. Verify it works locally before moving to layer 02.
- One layer at a time. Framework + UI → Auth → DB → Forms → Payments → Email. Resist the urge to install everything before the first
git commit. - Commit per layer. Easier rollback when one layer goes sideways.
- Ship to Railway / Vercel. Set the env vars, push the branch, smoke-test the live URL.