Lift + tint on hover, visible focus ring, aria-current for the dashboard,
colourful icons on a black-and-white surface, prefers-reduced-motion respected.
- + button next to Dashboard (desktop and mobile sheet), lazy-loaded native <dialog>
- POST /trackers/tasks/quick inserts source 'quick-add', notes, due date, priority
- tasks.tasks.auto_handle (default false) with a Daisy badge in the tracker
- DAISY_TASK_WEBHOOK_URL/KEY/HEADER: fire-and-forget POST {task_id, title}, 3 s timeout
TrackerIcon + #lib/tracker-icons: one gradient rounded tile and a few stroke paths per
tracker (tasks, money, books, cars, music, health, groceries, houses, energy, games, skins,
pets, lists), distinct colours, readable in light and dark. No icon library. Test checks
every tracker has a unique tile and tasks/money/books stay first.
/start is no longer immersive, so it loads the lazily imported SiteChrome like other pages.
Only the book reader still skips the chrome. hidesFooter() keeps the footer off /start.
- music schema: artists/albums/tracks (+album art), plays with source + dedupe, now_playing,
spotify_auth, import_state; plays_deduped view hides Last.fm scrobbles Spotify also has
- in-app Spotify poller (60s, leased) started from hooks init; no new role or env needed
- /api/spotify/connect + callback (owner only), /trackers/music public stats
- scripts: music-poll, music-import-lastfm (resumable), music-import-history
- deploy/music-collector: optional host timer with its own role
- Header/footer moved to #lib/components/SiteChrome.svelte, returned by a universal
+layout.ts load only for non-immersive pages, so /start and the book reader never
download icon libraries, dropdowns or the bits-ui Sheet. /start: 36 -> 18 JS files,
133 -> 78 KB gzip.
- Header logo is a hashed asset (inlined), so it is cached with the build.
- inlineStyleThreshold 4096: tiny page/component CSS is inlined into the HTML.
- Umami is not loaded on /start.
- home schema: home.links (seeded, owner-editable) and home.snapshots (written by a host collector)
- deploy/home-collector: stdlib Python + systemd user timer (60s) that snapshots /proc stats,
docker containers, systemd timers, service HTTP checks and ntfy topics via a limited role
- extras/newtab-extension: MV3 new-tab override, zipped to static/newtab-extension.zip
- owner-only pages skip Google Analytics; /trackers/health#body anchor for weigh-ins
The nightly scraper and worker API now take over an existing row (e.g. source=manual)
whose link is the same listing (normalized URL, or the Pararius key / Funda, Kamernet,
Vesteda id in it) instead of inserting a copy; typed values are kept where the scrape has
none. house-listing update/add give manual rows with such links the site key unless
another row has it, update accepts --source/--source-id, and lookups by link ignore
tracking params. scripts/house-rekey.ts backfills existing manual rows.
Lists contracts/pension docs with written summaries, an inline viewer and
download at /trackers/money/documents. Files and summaries live on the host
(MONEY_DOCUMENTS_ROOT, bind-mounted read-only by deploy-remote.sh), never in
git or static/; the file endpoint is admin-only (401/403) and private, no-store.
- scripts/house-scrape.ts: polite nightly scrape (honest UA, 6-10 s apart, stops a source on 403/challenge), ingest + price history + removals, ntfy alerts on a separate topic (max 5 + summary, first run seeds)
- address_key + duplicate_of: copies on another source link to one listing, manual wins
- /projects/house-tracker -> /trackers/houses (308), owner only (404 otherwise); filters for min m2, rooms, source, Gouda region default; notes, add-by-hand form, total rent, also-on links
- scripts/house-db.sh: run the house-listing CLI as the limited house_cli role
- deploy/systemd/house-scrape.{service,timer}
scripts/house-listing.ts (bun run house add|update|list|remove|restore) upserts
house.listings on (source, source_id) derived from the listing URL or address,
derives city_slug, records price_history, and accepts flags or JSON.
pdf.js 6 removed PDFDocumentProxy.destroy(), so the PdfReader teardown threw
"destroy is not a function" on every unmount, breaking the client-side
navigation back to the book page (and sometimes showing the error page).
Close the document through its loading task instead (closePdf), which also
cancels a load still in flight. Test covers the pdf.js API and closePdf.
- Zoom buttons + Ctrl/Cmd +/-/0, Ctrl+wheel/trackpad pinch, touch pinch on PDFs;
PDF scales the pdf.js render (sharp re-render after it settles), EPUB scales text.
Remembered per format, and per book.
- One/two pages side by side: PDF rows with the cover alone, foliate max-column-count.
- Scroll vs Pages mode for both engines; PDF paged shows one page/spread fitted to the
screen, prev/next move a whole spread; spreads stack when scrolling.
- Reading position is kept across zoom/spread/mode changes; progress keeps saving.
- layout.ts helpers with unit tests; old flow/fontSize/pdfZoom settings migrate.
Adds money.accounts.include_in_total (default true). Excluded accounts keep their own balance and line, but are left out of net worth, invested, this month, the projection and the total/All accounts lines.
Run #79 failed at "sudo tee /opt/personal-website/compose.books.yml": the deploy
account may only sudo-tee .env. Write the override to ~deploy instead; the
compose CLI runs as root and reads it from there.
Run #78 skipped the /books mount because [ -d /home/stan/books/library ] runs as
the deploy account, which has no access to /home/stan (0750). Check inside a
container with the dir bind-mounted instead; the mount itself is done by dockerd.
