One-time PocketBase migration that replaces every article's slug with
article-1, article-2, ... ordered by publishedAt (oldest first),
requested explicitly by the site owner in place of the existing
descriptive slugs.
The language switch overlapped the centered logo on narrow viewports
(~390px), and article card covers used an overly aggressive 4:5 crop
that sliced through text baked into cover images.
It was deliberately left Italian-only when /en support was first added;
now translated and following the same thin-page + shared-view pattern
as every other route, with the nav link and language switch no longer
special-casing it.
Blog and category archives now show a numbered page list (first, last,
current +/-1, collapsing gaps into an ellipsis) instead of only
previous/next — see shared/utils/pagination.ts's paginationRange().
Renders page links via NuxtLink's custom/v-slot API rather than letting
it render its own <a>: RouterLink's built-in active-class/aria-current
detection compares only the route's path, not its query string, so
every ?page=N link was incorrectly marked as the current page.
The language switch now shows both "Italiano / English" at all times,
with the current one as plain non-interactive text and the other as a
link, moved to the header's top-right corner instead of stacked under
the logo.
PocketBase gains optional, manually-authored English fields on articles
and categories (same record, same slug), and the frontend serves an /en
counterpart of every dynamic route via thin page wrappers around shared
view components — no i18n library, consistent with the project's existing
minimalism stance for a two-locale site with ~20 UI strings.
An article/category with no translation 404s cleanly on its own /en
detail page and is filtered out of /en listings, and the header's
language-switch link falls back to the English blog index rather than a
dead link; resolving that requires a global route middleware, since the
layout's header renders before the page content in document order and so
can't react to state a page component sets during its own async setup.
PocketBase silently enforced its own default max on unbounded text
fields, which broke saving a real production article longer than that.
Make the limit explicit and generous enough for the content this blog
actually publishes, and add a test asserting the exact boundary.
Makes local test data (e.g. content pulled from production for testing)
directly inspectable and swappable on the host filesystem, without
affecting the production volume-based setup.
tests/{unit,integration,e2e}/, at the repo root (not under frontend/,
since e2e exercises the frontend and PocketBase together) — all
against real code rather than mocks:
- unit/: Vitest (nuxt environment) — pure functions in server/utils
and shared/utils, plus components/composables via
@nuxt/test-utils/runtime's mountSuspended.
- integration/: Vitest (node environment) — @nuxt/test-utils/e2e's
setup() builds and runs the real Nitro server against a real
ephemeral PocketBase instance, exercising the actual /content/* and
/admin routes end to end.
- e2e/: Playwright, a real browser against the built app and another
ephemeral PocketBase, covering the user-facing flows (home, blog
archive, article detail incl. JSON-LD, category filter, 404s, the
/admin redirect).
The Vitest/Playwright tooling (and node_modules) stays in frontend/,
the repo's only npm project; its configs just point at ../tests/.
Playwright resolves packages from the node_modules nearest each spec
file, so `npm run test:e2e` first symlinks tests/node_modules to
frontend/node_modules (pretest:e2e, idempotent, gitignored).
tests/support/pocketbase.ts is the shared piece behind integration
and e2e: it downloads the same pinned PocketBase binary
pocketbase/Dockerfile uses, starts it against the real
pocketbase/pb_migrations/ (not a copy), and seeds it from
tests/support/fixtures.ts — so these tests run against the exact
schema and API rules that ship to production, catching the class of
bug that only shows up when PocketBase actually enforces them (e.g. a
wrong filter/fields query string silently returning nothing, or too
much).
CLAUDE.md and docs/frontend.md document the new npm scripts and the
single-test commands for each layer.
Wires up @nuxt/eslint (the official Nuxt module) so `npm run lint`,
already documented in CLAUDE.md's pre-completion checklist, actually
exists. It auto-generates a flat config aware of Nuxt's auto-imports,
so it doesn't flag composables and utils that Nuxt injects without an
explicit import.
Strapi + Postgres are gone in favor of PocketBase: a single Go binary
with embedded SQLite, built-in admin UI and per-collection API rules.
No content existed yet, so this is a clean swap with no data migration.
Collections and rules are defined as code in pocketbase/pb_migrations/
and applied automatically on first boot. Draft & Publish has no native
PocketBase equivalent, so it's reproduced with a nullable `publishedAt`
field enforced by listRule/viewRule, matching the old Strapi semantics.
Routing flips: PocketBase's admin UI and REST/file API are hardwired to
`/_/` and `/api/*` at the domain root (its own dashboard assets and API
calls reference those paths directly, so a stripped path prefix like
`/admin/*` would break them). `/api` is therefore reserved for
PocketBase now, and the frontend's Nitro endpoints move to `/content/*`
(frontend/server/routes/content/, not server/api/). A `/admin` vanity
route in Nitro (not Caddy) redirects to `/_/`, so it works the same in
dev, where Caddy isn't part of the stack, and in production.
frontend/server/utils/strapi.ts becomes pocketbase.ts; queries.ts is
rewritten for PocketBase's filter/sort/fields/expand query syntax.
StrapiImage becomes MediaImage (no width/height — PocketBase file
fields don't store dimensions, and the cover images already reserve
their aspect ratio via CSS, so this is not a regression).
docs/*.md, CLAUDE.md and README.md are updated in the same commit.
Remove @strapi/plugin-cloud (deploy is self-hosted via Docker/Ansible,
never Strapi Cloud), disable the admin NPS/Enterprise-promotion flags
(single-admin panel, no upsell needed), disable Strapi's anonymous
telemetry, and prune dev dependencies from the production Docker image
layer.
Home now shows only the latest article, pointing to the full list on
/blog instead of duplicating recent articles in a grid. Nav trims to
Home, Tutti gli articoli, and the corralito page; the footer drops the
now-redundant Categorie column.
Turn the home page into a single-column feature: a lead image, the
opening statement as the page's sole h1, a second image, and the body
copy, with the existing latest-articles section still below. Add a new
static page for the crypto courses pitch ("come difendersi dal
corralito"), styled like an article with a cover image and a contact
CTA (email + Telegram) instead of the referenced but nonexistent
contact form. Link the new page from the header nav and footer.
The interface was translated in en/it/es/fr via @nuxtjs/i18n. Drop the
module, the locale files, and the language switcher: the UI is now
static Italian text, with translation for foreign visitors left to the
Google Translate browser extension instead of the app.
Strapi mounts each plugin's admin API at its own top-level path
(/content-manager, /upload, /i18n, ...), not nested under /admin as
previously assumed. That broke the dashboard widgets: recent/count
documents calls landed on the frontend instead of Strapi. List every
prefix actually used by the installed plugins (core-bundled ones plus
users-permissions and cloud from cms/package.json) explicitly in the
Caddyfile.
Replace the separate CMS subdomain with path-based routing in Caddy:
/admin and /uploads go to Strapi, everything else to the frontend.
Strapi 5 nests its whole admin panel (UI and API) under /admin, so this
one prefix is enough and never collides with the frontend's own /api
routes. Drops CMS_DOMAIN and the unused STRAPI_URL env var on the cms
service; PUBLIC_STRAPI_URL now points at the same origin as the site.
Authentication is unchanged: Strapi's own admin login still gates the
panel, this only changes how it's reached.
The host's systemd-resolved stub (127.0.0.53) is unreachable from inside
the container network namespace, so Caddy could not resolve Let's Encrypt
endpoints. Point it at public resolvers instead.
Mobile-first layout with breakpoints at 48rem (tablet) and 64rem (desktop):
stacked nav on phones, progressive column counts for the article grid and
footer, scrollable markdown tables, larger tap targets on nav links.
The marks come from the simple-icons package (CC0) and are inlined at build
time — the earlier hand-drawn glyphs were wrong, the X mark in particular is
not a plain cross. Importing the three SVGs individually rather than the
package index keeps the other icons out of the bundle: 10 kB, not 5 MB.
Rendered monochrome in the surrounding text colour, with the text label kept
next to each one.
Strapi already records which admin user created an entry, so the byline needs
no extra field to fill in: the Article content type sets populateCreatorFields
and the API layer exposes only that user's firstname and lastname. The name
also feeds the BlogPosting structured data.
Also fixes the gap below the article header, which until now came from the
cover image's margin and therefore vanished on articles without a cover,
leaving the body text against the byline.
The interface is translated through @nuxtjs/i18n: strings live in
i18n/locales/*.json, routes are prefixed except for the default English, and
useLocaleHead emits the lang attribute, the hreflang alternates and og:locale.
The switcher is a native select rather than a custom dropdown: keyboard
support, the platform picker on mobile and correct labelling come for free.
Article content stays single-language — this translates the site chrome only.
The seal already carries the name, so it replaces the text wordmark in the
header and opens the footer. Small icon sizes use the central emblem only: the
circular lettering is unreadable below 48px.
Centred wordmark over a rule of category links, a lead article above a grid of
cards, centred article headers over a narrow reading column, and a footer split
into columns. Neutral warm-grey palette with a serif display face for headings
and small uppercase labels for metadata.
README gets step-by-step instructions for both: secrets generation and the
local stack on one side, DNS records, firewall, environment variables, TLS and
backups on the other.
CLAUDE.md records the decisions that are not visible in the code: why the
browser never reaches Strapi, why the content model stays this small, and where
the public permissions are granted.
Production traffic enters through Caddy, which terminates TLS for the public
domain and the CMS subdomain and sets HSTS and the other security headers.
PostgreSQL, Strapi and Nuxt publish no ports of their own.
docker-compose.dev.yml publishes the ports on localhost and drops Caddy for
local testing. It is a separate file rather than an override.yml so it can only
be applied when passed explicitly, never by accident in production.
The browser never talks to Strapi: pages call Nitro endpoints under
server/api/, which are the only place Strapi queries are built. That keeps the
Strapi URL internal to the Docker network and avoids CORS entirely.
Article bodies are rendered from Markdown to HTML in the endpoint, so the
content is server-rendered and crawlable, marked stays out of the client
bundle, and the meta description is derived from the opening of the body.
Pages carry canonical URLs, Open Graph tags and BlogPosting structured data,
and handle empty, missing and failed states.
Article is deliberately minimal: title, slug, Markdown body, cover image and
category. No Author content type — the administrator is the only writer.
The bootstrap hook grants the Public role find/findOne on the two content types
and disables self-registration, so the public API is read-only and the site has
no front-end accounts.
davide