This is the source for Luca Pipolo's personal portfolio website, statically built with Astro.

Tech stack

• Astro
• TypeScript
• Tailwind CSS
• Alpine.js
• Sentry

Project structure

src/
├── pages/[...locale]/    routes, one tree per locale
├── content/               content collections: pages, settings, blog posts
├── components/
│   ├── common/            generic, reusable primitives
│   └── sections/          named page sections (Hero, Blog, Post, SiteHeader...)
├── layouts/main.astro     shared HTML shell
├── libs/                  i18n, content loaders, templating, utils
├── scripts/               client-side browser scripts
├── styles/                Tailwind config in CSS, plus partials
└── icons/                 hand-authored SVGs

middleware.ts              locale negotiation, at the edge
vercel.json                security headers
scripts/                   the Docker scripts below

Running this site locally

You have two ways to run this project: with Docker, or directly with bun installed on your system.

Option A: with Docker

1. Install Docker (or OrbStack as a macOS alternative) and make sure it's running.

2. Clone the repository:

   git clone git@github.com:LucaPipolo/lcp-web.git && cd lcp-web

3. Make .test domains resolve to your machine (I suggest using dnsmasq).

4. Run the setup script and open the site:

   ./scripts/setup.sh

5. Visit https://lcp-web.test.

The dev server is already running with hot reload with Vite, so any change you make is reflected straight away; there's nothing extra to start.

Caution

The Docker scripts in this repo (scripts/*.sh) currently only support macOS.

Note

On the first run, macOS asks for your password once, to trust the SSL certificate.

Commands you'll use day to day

Command	What it does
./scripts/up.sh	Start this checkout
./scripts/down.sh	Stop it, and the router if nothing else uses it
./scripts/setup.sh	Rewrite .env, then start. Safe to run any time

All three accept --verbose (or -v) for full output instead of one line per step.

To run project scripts, use docker compose exec against the app container, from this checkout's own root:

docker compose exec app bun run dev       # dev server
docker compose exec app bun run build     # production build
docker compose exec app bun run preview   # preview the build

Built for git worktrees

The Docker setup is designed around git worktrees: each worktree gets its own containers, port, and domain, so they never collide.

git worktree add ../my-feature
cd ../my-feature
./scripts/setup.sh

That checkout is now served at https://my-feature.lcp-web.test, alongside lcp-web.test and any other worktree, each running at the same time without conflicts. A single shared router handles HTTPS for all of them; down.sh only stops that router once nothing else is using it.

This makes the project a good fit for working with multiple AI agents or an ADE like Orca side by side. Each agent can work in its own worktree, with its own running instance, without stepping on the others.

I use Orca as my ADE and create a worktree there for each feature. Pointing its setup and archive scripts at this repo's own gives every new worktree a running Docker environment automatically, with no manual step in between:

Orca setting	Value
Setup script	$ORCA_WORKTREE_PATH/scripts/setup.sh
When to run	Run by default
Wait for setup to complete before starting agent	On
Archive script	$ORCA_WORKTREE_PATH/scripts/down.sh

Setup starts that worktree's own containers before the agent gets to work in it; archive stops them once the worktree is gone.

Day to day I work mostly inside Orca itself, including its per-worktree browser: its Design Mode turns the cursor into an element picker, hovering highlights the element under it, useful for pointing an agent at something on the page. For manual edits I switch to Zed: fast to jump between worktrees, and a solid editor on its own.

Option B: without Docker

1. Install bun.

2. Clone the repository:

   git clone git@github.com:LucaPipolo/lcp-web.git && cd lcp-web

3. Install dependencies and start the dev server:

   bun install && bun run dev

This serves the site at localhost:4321 with no HTTPS, but it is otherwise identical to Option A.

Caution

The git hooks run inside the Docker container, so without it running you'll need git commit --no-verify to commit.

AI tooling

This repo is set up to work with Claude Code. This section covers the parts meant to be reached for by hand, not the ones Claude uses on its own.

I run this repo on Sonnet as the base model, and set /advisor, a second opinion consulted before committing to an approach, to Opus rather than have Sonnet review itself. Anthropic's own guide to choosing a model covers that trade-off.

Commits made with AI assistance carry a Co-Authored-By: Claude Code <noreply@anthropic.com> trailer, plus a Model: <name> (<id>) line naming what actually did the work (see git log). I want that visible rather than folded into ordinary authorship: recording which model helped is how I track and improve how I use these tools over time. The next thing I want there is a token count per commit, so the cost side is visible too, once the tooling reliably exposes that.

Skills

Skill	What it does
/open-pr	Opens a GitHub PR for the current branch, writing the title and body to match this repo's template.
/address-pr-comments [PR]	Works through a PR's unresolved review comments: fixes what's valid, replies to what isn't. Defaults to the current branch's PR, or targets the given PR number.
/update-dependencies	Updates this repo's Bun dependencies one at a time, each in its own commit.

Note

I review PRs with ChatGPT Codex, not a human teammate, since I work alone on this project. /address-pr-comments only handles Codex's own bot account for that reason; extending it to other reviewers would be straightforward.

Note

/update-dependencies is deliberately not left runnable on its own (disable-model-invocation: true in its frontmatter). Wiring it into a scheduled cloud agent, a weekly cron run rather than a live /loop (which only polls while a session stays open), would be straightforward. I just haven't set it up yet.

MCP servers

Configured in .mcp.json, committed so every checkout gets them:

Server	What it's for
Vercel	Deployments, project and domain info, logs, for this project's own account
Astro docs	Up to date Astro documentation lookup
Sentry	This project's error reports, issues, and traces
Context7	Up to date documentation lookup for any library, including Tailwind CSS and Alpine.js, which have no official server of their own

Note

Vercel and Sentry authenticate through OAuth on first connection.

License

MIT