Build web apps the boring way.

What this is

1. The skill. Type /boring-stack in any Claude Code session and Claude loads opinionated defaults for that project. Pushes back on Postgres, Vercel, ORMs, Docker, and microservices when they don't fit the archetype. Two sentences plus a question per pushback. Always defers to your call.
2. The templates. The exact deploy.sh, Caddyfile, app.service, and Litestream config the Borela cloud build runs on (Borela is pre-launch — the templates have been live-fired against the dogfood deploy). Drop them in, replace the hostnames, ship. Zero magic.
3. The manifesto. Seven principles for shipping software that stays out of your way. Quote it, link it, fork it. Read the manifesto.

Install

One line. Works with every major AI coding tool.

$ curl -fsSL https://boringstack.org/install.sh | bash

That clones the repo to ~/.boring-stack and wires it into the tools that have a user-level config (Claude Code, Codex CLI). For tools that only support per-project rules, run this from inside any project where you want the boring stack defaults:

$ curl -fsSL https://boringstack.org/add.sh | bash

Auto-detects your project's existing AI tool configs and drops the right file in each. Falls back to AGENTS.md (the portable convention) if nothing's set up yet.

tip: the skill is MIT and stays MIT. Use it, fork it, ignore it, write a better one.

What the skill argues for

• Postgres SQLite + WAL + Litestream10k+ writes/sec on a $5 VPS. Continuous backup to R2 for $0.50/mo. Migration to Postgres later is pgloader + a connection string.
• Vercel Hetzner + Caddy$5/mo at every scale. Caddy auto-handles HTTPS via Let's Encrypt. No certbot, no cron, no nginx.
• Prisma / GORM stdlib + sqlcType-safe Go generated from .sql files. No runtime layer. Read your own queries.
• Docker / K8s systemd + one binaryAlready on every Linux box. Restart=on-failure is one line. journalctl is your log aggregator.
• Microservices Go monolith with internal packagesCompiler-enforced module boundaries. Extract behind HTTP later when team and traffic actually demand it.

Stay on the list

Why this exists

I'm building Borela. When it ships, Borela will be the babysitter for web apps running on the boring stack: it verifies your backups are actually restorable, drills the restore on a schedule, watches the agent heartbeat, and pages someone when the cron silently stops.

Pre-launch. No customers yet, no paid signups open, no public site. I'm publishing this skill before paid Borela ships, on purpose. If people install it and start running the boring stack on their side projects, some of those projects will grow up and need a babysitter. That's the bet.

This is not an original bet. Pieter Levels has been writing about this shape of stack for years. Steve Hanov runs $40k MRR across four products on a $20/month stack. DHH made a career out of refusing to add complexity that didn't earn its way in. I'm standing on shoulders.

What I want to add is the skill. Claude defaults to the modern hosted-everything pile (Next.js, Vercel, Supabase, fourteen npm packages) because that's what most of the training data looks like. The skill flips the default.

Install the skill and try it. If it argues with you in a way that doesn't fit your project, the archetype-fit table at the top of SKILL.md tells you when to ignore it.

Templates

The configs the skill ships with. Each one is the actual file running Borela in production, lightly edited to use placeholder hostnames.

• deploy.sh10 lines. build, scp, restart, healthz.
• Caddyfile8 lines. Auto-HTTPS reverse proxy.
• app.service22 lines. Hardened systemd unit with NoNewPrivileges, ProtectSystem=strict, ReadWritePaths scoped.
• litestream.service15 lines. Continuous SQLite replication to R2.
• litestream.yml10 lines. R2 destination, env-var secrets.