Every Claude Code session, every project, every token — tracked horizontally. Not a status update. The actual operating dashboard.
| ID | Project | Stream | Status | Progress | Target | WIP |
|---|
Which agent owns what — active project assignments across all 18 agents. WIP bar shows capacity utilization (limit 3/agent). Click a project pill to jump to portfolio.
| Session | Date | Model | Tokens (est.) | Ctx % | Duration | Files | Tasks |
|---|
ノードをクリックして選択。役割・KPI・連携先が展開されます。
Toyota-style VSM. PT = Process Time (value-add work). WT = Wait Time (queue / blocked). The timeline bar shows where time is actually spent. %C/A = completeness & accuracy from upstream. Toggle to see improvement targets.
3 variables per project (Delay · Resource % · Scope %) + 2 global variables (Energy level · External blocker). Run to see cascading timeline, cost, and quality impact across all agents.
After each Claude Code session, update the DATA object in the <script> block:
1. Session Log — Add new entry to DATA.sessions[] with date, model, tokens, duration, tasks, outcome.
2. Token count — Read from Claude Code's context indicator (e.g. "857k / 200.0k"). Record as tokensApprox.
3. Projects — Update status and pct when a project phase changes.
4. Milestones — Change status: "pending" → "done" when reached.
Future: Cloud Function auto-export from BigQuery usage logs → this page auto-updates from CDN JSON.