name	run-website
description	Run, smoke-test, screenshot, and deploy the casparsadenius.de website (Django + PostGIS). Use when asked to run, start, test, verify, restart, or screenshot the website, diary map, or its APIs.

Run: Website (casparsadenius.de)

Django 5.2 + PostGIS travel-diary site. The web and db containers are managed by the dashboard stack (/root/dashboard/docker-compose.yml); ~/website/docker-compose.yml only runs nginx + certbot. Code is live-mounted into the container (/root/website:/app) — no image rebuild for code changes. Paths below are relative to /root/website/.

Smoke test (agent path — run this first)

.claude/skills/run-website/smoke.sh            # full run incl. screenshots (~45 s)
SKIP_UI=1 .claude/skills/run-website/smoke.sh  # containers + HTTP + ORM only (~10 s)

Checks containers, public HTTPS endpoints (/, /diary/, diary APIs), manage.py check, a read-only ORM query, and renders /diary/ headless (screenshot → /tmp/shots/website-diary.png, blank-detection, MapLibre <canvas> present in DOM — WebGL works via SwiftShader in zenika/alpine-chrome).

Direct invocation (most backend changes need only this)

docker exec website-web python manage.py shell -c \
  "from apps.diary.models import Trip; print(Trip.objects.count())"
docker exec website-web python manage.py check

The container runs the live working tree, so an edited module is importable immediately — but running Gunicorn workers keep old code until restarted (see Deploy). manage.py shell starts a fresh process, so it always sees your edit.

Deploy a code change

Documented path (from CLAUDE.md + compose ownership; restart not re-run in the authoring session — it briefly interrupts the live site):

cd /root/dashboard && docker compose restart web     # note: dashboard dir, not ~/website

Needed for: any .py change (Gunicorn), any template change (cached loader in production).
Migrations + collectstatic run automatically on container start (compose command).
What IS verified from the authoring session: recreating web via the dashboard stack brings the site back healthy (HTTP 200) within seconds.

Gotchas

cd ~/website && docker compose restart web does not work — web is not in that compose file (its own CLAUDE.md deployment section predates the move to the dashboard stack). Use /root/dashboard.
A plain docker compose up -d --build in /root/dashboard rebuilds the website image and recreates the live web container whenever /root/website changed. Scope compose commands to one service.
Media files live in the named volume website_media_volume, not in ~/website/media/.
/tmp/shots must be chmod 777 (Chrome runs as uid 1000).
/admin/, /diary/manage/ are nginx basic-auth'd — smoke only covers public routes; management UI testing needs the user's browser.

Troubleshooting

Symptom	Fix
Public URLs 502	`docker logs website-web --tail 50` — Gunicorn may still be starting after a restart
ORM query prints nothing	Run without `tail -1` to see the traceback; check `DJANGO_SETTINGS_MODULE=config.settings.production`
Diary screenshot blank	Raise `--virtual-time-budget` (map tiles + GeoJSON take a few seconds)

Run: Website (casparsadenius.de)

Smoke test (agent path — run this first)

.claude/skills/run-website/smoke.sh # full run incl. screenshots (~45 s) SKIP_UI=1 .claude/skills/run-website/smoke.sh # containers + HTTP + ORM only (~10 s)

Direct invocation (most backend changes need only this)

docker exec website-web python manage.py shell -c \ "from apps.diary.models import Trip; print(Trip.objects.count())" docker exec website-web python manage.py check

Deploy a code change

Documented path (from CLAUDE.md + compose ownership; restart not re-run in the authoring session — it briefly interrupts the live site):

cd /root/dashboard && docker compose restart web # note: dashboard dir, not ~/website

Needed for: any .py change (Gunicorn), any template change (cached loader in production).

Migrations + collectstatic run automatically on container start (compose command).

What IS verified from the authoring session: recreating web via the dashboard stack brings the site back healthy (HTTP 200) within seconds.

Gotchas

cd ~/website && docker compose restart web does not work — web is not in that compose file (its own CLAUDE.md deployment section predates the move to the dashboard stack). Use /root/dashboard.

A plain docker compose up -d --build in /root/dashboard rebuilds the website image and recreates the live web container whenever /root/website changed. Scope compose commands to one service.

Media files live in the named volume website_media_volume, not in ~/website/media/.

/tmp/shots must be chmod 777 (Chrome runs as uid 1000).

/admin/, /diary/manage/ are nginx basic-auth'd — smoke only covers public routes; management UI testing needs the user's browser.

Troubleshooting

Symptom

Fix

Public URLs 502

docker logs website-web --tail 50 — Gunicorn may still be starting after a restart

ORM query prints nothing

Run without tail -1 to see the traceback; check DJANGO_SETTINGS_MODULE=config.settings.production

Diary screenshot blank

Raise --virtual-time-budget (map tiles + GeoJSON take a few seconds)