SHORE dashboard

Live portal for the Shellfish Hardening & Organismal Resilience Experimentation Center (Roberts Lab, UW). Published at https://robertslab.github.io/shore-line/.

Widget	Source	Refresh
Live Tankcam	gannet v1_web/webcam/latest.jpg (Last-Modified header) → data/webcam.json	site rebuild every ~10 min; page re-polls every 60 s
Puget Sound water temp	NOAA CO-OPS (Tacoma 9446484, fallback Port Townsend) → data/field-water.json + assets/water-banner.svg	every rebuild
Water-quality reports + per-tank readings	genefish RSS (Jesse Lowe) → data/reports.json	every rebuild
Facility temperatures	data/tanks.json (setpoints, placeholders); Left Blue is live from a seneye → seneye-data branch → data/seneye.json	collector every 15 min
Research from SHORE	data/research.json (hand-curated)	—

How it works

A static site (index.html, css/, js/) with no build step. .github/workflows/site.yml runs every 10 minutes. It fetches the live sources with the stdlib-only scripts in scripts/, writes the results into _site/, and deploys to GitHub Pages. Fetched data is not committed. If a source is down, the script republishes the copy that is currently live and marks it with an error, and the status dot for that feed turns amber or red.

Tankcam AI captions

scripts/caption_tankcam.py runs in every build after the webcam step. It asks Claude (claude-opus-5-5, low effort, structured JSON output) to caption the latest frame. That happens at most once per 2-hour slot from 6 AM to 6 PM Pacific (7 a day, ~$2–3/month), and only when the frame has changed. The rest of the time it carries the published caption forward. The caption, an approximate oyster count (only when oysters are visible), and anomaly flags are added to data/webcam.json and shown under the snapshot, labelled "AI caption".

• Needs: the ANTHROPIC_API_KEY Actions secret. Without it, the step skips quietly. If the key isn't scoped to a workspace, also set the ANTHROPIC_WORKSPACE_ID repo variable.
• Change the model: set the CAPTION_MODEL env var on the workflow step.
• Caption immediately, ignoring the schedule: python scripts/caption_tankcam.py --force. This needs the key in your environment and pip install anthropic.

Seneye (Left Blue tank)

The Seneye Web Server sits on a private UW address (172.25.149.26), which GitHub can't reach. So scripts/collect_seneye.py runs on a machine inside the network. It decodes the SWS WebSocket state, which has no JSON API, and keeps a 14-day series. scripts/seneye_publish.sh force-pushes that file as a single commit on the seneye-data branch. The site build mirrors it from the SENEYE_URL repo variable.

• Currently runs on: sr320's Mac, via launchd every 15 min. It only collects while that Mac is on and on the UW network. Log: /tmp/shore-seneye.log.
• Install / update (after editing either script):

scripts/install_seneye_agent.sh

• Uninstall: launchctl unload ~/Library/LaunchAgents/com.robertslab.shore-seneye.plist
• Moving to gannet later: run seneye_publish.sh from cron there. Alternatively, write the JSON to a web-served path and point SENEYE_URL at it.
• Device limits: the SWS allows one WebSocket client at a time, so the collector retries if someone has the device's page open. When the chemical slide is missing or expired, the device reports pH/NH₃ as 0. The collector publishes null for those instead.

Common edits

• Feature a paper: add an object to items in data/research.json (featured: true puts it in the top cards, max 3). Remove the placeholder entry once there's a real one.
• Tank list / setpoints: edit data/tanks.json. Set status to anything other than placeholder and fill latest / series when a logger feed exists.
• Change feeds: env vars WEBCAM_URL, REPORTS_FEED, SITE_LABEL.

Run locally

python3 scripts/update_webcam.py && python3 scripts/update_reports.py && python3 scripts/update_field_water.py

python3 -m http.server 8320

Then open http://localhost:8320. update_field_water.py --demo uses synthetic data when you have no network.

See PLAN.md for the roadmap.