Stop Wasting the First Hour of Every Project

Every new project starts the same way.

I have an idea. I know what I want to build.

Then comes the part where I have to decide what to include or what the structure of the project should be.

"Which frontend or backend to go with?" "Should the auth be set up from the start or can it be done later?" "Which database would be best for this? Do I use an ORM?"

Most of the time, I either spend too long comparing options I don't really need yet, or I pick something fast and keep adding things on as I go.

That second approach always costs more time later. You might end up retrofitting auth into something that wasn't designed for it, or realising the database you picked doesn't play well with the ORM you added three weeks in.

What Better T Stack does

Better T Stack, created by Aman Varshney, provides a stack builder where you can configure the frontend, backend, database, ORM, auth, and other add-ons before you write a single line of code.

You can also check whether a technology is compatible with the rest of your stack before committing to it. That alone saves a round of research every time a new project starts.

So instead of starting with a big template and removing things I do not need, I can pick the parts that actually fit the app.

It does not make the decisions for you. But it puts all the choices in one place, before you've already started building around them.

Here is everything it lets you choose from:

The compatibility check is the part I found most useful. If a combination of database, ORM, and backend does not work together, it tells you before you start, not halfway through.

What I used for a demo project

Here is the command I ran to set up a full-stack project with the stack I wanted:

1bun create better-t-stacklatest AppDemo 2  --frontend tanstack-router 3  --backend express 4  --runtime bun 5  --api trpc 6  --auth better-auth 7  --payments none 8  --database postgres 9  --orm prisma 10  --db-setup prisma-postgres 11  --package-manager bun 12  --git 13  --web-deploy vercel 14  --server-deploy vercel 15  --install 16  --addons fumadocs mcp oxlint pwa skills turborepo 17  --examples ai todo

That gives me a Turborepo monorepo with a TanStack Router frontend, an Express backend running on Bun, tRPC for the API layer, Prisma connected to a PostgreSQL database on Prisma Postgres, Better Auth for authentication, and Vercel as the deployment target for both the web and server.

The add-ons include Fumadocs for documentation, an MCP server, oxlint for linting, PWA support, and Turborepo for managing the monorepo. There are also two example apps included: an AI chat interface and a todo list, both useful for seeing how the stack fits together before writing any of your own code.

0:00

0:00

0:00

The skills addon

One of the add-ons in that command is --addons skills.

When you include it, Better T Stack pulls in a set of agent skills tailored to the technologies in your project. If you chose Prisma, you get Prisma-related skills. If you chose tRPC, those skills are included too. The idea is that your AI coding tool starts the project with better context about the stack you're actually using, rather than defaulting to generic patterns it has seen across millions of other codebases.

This matters because the first version of any AI-generated code is usually a starting point. The agent knows a lot, but it does not know that this specific project uses tRPC with Bun, or that the auth layer is Better Auth, or how those pieces are expected to work together. Skills give it that context upfront.

When the built-in skills are not enough

The skills that come with Better T Stack cover the stack you configured. That is a good start, but depending on what you are building, you might need more.

I found skills.sh when I started thinking about this more carefully.

It is a library of agent skills organised by topic. Frontend and React, Next.js, Design and UI, Mobile, Agent workflows, Database, Testing, Marketing, and more. You can pick the skills that match what you are building and install them before the agent starts writing code.

1npx skills add Lombiq/Tailwind-Agent-Skills2npx skills add vercel-labs/agent-skills --skill web-design-guidelines

The difference between starting with and without these skills is visible in the output. Without them, the agent tends to repeat what it has already done for every other project. With them, it has a clearer set of constraints and patterns specific to the work you are asking it to do.

What the combination looks like in practice

Better T Stack handles the stack selection before the project starts. The built-in skills add context for the specific tools you chose. skills.sh fills in the gaps for anything more specialised, like UI patterns, design guidelines, or domain-specific workflows.

None of this replaces the actual work. The agent still needs direction, and the code it generates still needs review. But starting with the right stack and the right skills means the first round of output is closer to something you can actually use.

The alternative is spending the first hour of every project on decisions that could have been made in five minutes, and then spending another hour explaining your stack to an agent that has no idea what you chose.

Sources & Links

• Better-T-Stack on GitHub — CLI and stack generator created by Aman Varshney
• Aman Varshney's Portfolio / Website
• skills.sh — Directory of agent skills