Stuff we've built

boilerplate_sveltekit

A front end foundation, built for coding agents.

SvelteKit with Svelte 5, strict TypeScript, Firebase Auth, and shadcn-svelte components. It's opinionated on purpose: the constraints that would slow a person down are the ones that help a coding agent get it right first time.

Like its backend sibling, it ships as npm packages, so improvements reach every project built on it.

Public · GitHubView on GitHub

How it works

Install what you need

Start with the core package, then add only the utilities your app uses: authentication, a typed API client, role-based access, and more. Their dependencies sort themselves out.

Run setup

A setup wizard asks for your project's name, Firebase project, roles, and features, then writes the configuration. An agent or a build pipeline can skip the questions and hand it a JSON file instead.

Add components

Pull components and whole page blocks from the registry. They're copied in as your own source, so you can change them freely, with nothing to fork.

Three tiers of components

Three tiers of componentsBasic atoms such as buttons, cards, and inputs are simple enough for an agent to write to a fixed pattern. Extended atoms such as calendars, data tables, and form wizards come from the registry. Blocks, hundreds of page-level compositions across more than fifty categories, also come from the registry. All three sit on the @xbg.solutions packages.Basic atomsButtonCardInputBadgeCODED TO A PATTERNExtended atomsCalendarData tableForm wizardCode inputFROM THE REGISTRYBlocksSign-inDashboardPricingSidebarHUNDREDS, IN 50+ CATEGORIES@xbg.solutions packagesAUTH, API CLIENT, ACCESS CONTROL, AND MORE · UPDATED THROUGH NPM

Built for agents to work in

The repo carries its own skills: written guidance on Svelte 5, the component system, Firebase, and the project's own rules, for a coding agent to read before it writes anything. Configuration lives in one object, so there's one place to change a role or a feature flag, not five.

Hardened before you start

The app talks to your API, never to the database directly, so access is decided in one place. It ships with a strict content security policy that allows no inline scripts, and protection against cross-site requests and unsafe input. Sign-in is passwordless, by email link or phone.

What building it taught us

We rebuilt it three times. The first version was built straight from what we wanted, and let the model make too many of the design choices. It didn't run at all. What changed for the second and third versions is how we now work with agents on any project.

Architecture first

Decide the standards before any code: what's a utility and what's a service, how errors and logs work, and what a good test looks like. An agent works like a capable senior developer who needs a junior's guardrails.

Test what, not how

Left alone, models over-mock and test the steps a function takes rather than what it does, so the tests break on every change. Spell out the principles, with examples of both good and bad tests.

Don't trust agreement

A model that answers "you're absolutely right" to a deliberately wrong suggestion can't be trusted on the right ones. We make honest disagreement a rule, and treat easy agreement as a sign the agent has lost the thread.

Get started

Install the core, then run setup:

npm install @xbg.solutions/bpsk-core
npx @xbg.solutions/bpsk setup

The code is public on GitHub, under the Business Source License 1.1.

Often goes with

For the API behind it, boilerplate_backend follows the same pattern, and the two run side by side in one project.

Ready for a conversation?

No sales pitch. No obligation. Just a chat about what you're building and where the opportunities might be.