pulse

A zero-build GitHub Pages dashboard tracking velocity across GitHub trending repos, HuggingFace models, and HF Spaces — a scheduled Action scrapes every 3h, appends to a 90-day observation log, and the static page renders deltas as bar charts and per-card sparklines.

Tools (set_focus, open_url, filter_tab) register with the browser's AI context via WebMCP (navigator.modelContext), so an external agent can drive the view instead of an embedded chat.

GitHub Action (cron 0 */3 * * *)
  └─ scripts/fetch.js ──► data/{github,huggingface,spaces}.json
                          data/history.json   (90-day log: source→id→[{d,v,r}])
                          commit [skip ci]
                                   │
static page (index.html) ◄─ fetch() data/*.json on load
  ├─ velocity bar chart  (ECharts, observed span)
  ├─ per-card sparklines (plain SVG)
  └─ tools.js ──► navigator.modelContext.registerTool(...)

Run locally

node scripts/fetch.js          # populate data/ once (no npm deps, Node built-ins only)
python3 -m http.server 8080    # serve — fetch() won't work over file://

WebMCP tool registration requires Chrome 146+ Canary with chrome://flags/#webmcp-for-testing; the dashboard renders fine without it.

Layout

Flat files, one concern each. index.html shell, index.css styles, index.js UI wiring; {github,huggingface,spaces}.js card renderers; charts.js velocity bar + sparklines; tools.js the WebMCP surface. See CLAUDE.md for the per-file map, data-flow notes, and visualization guidance.