Inbox Triage - moved to a live URL

Live dashboard

First visit on each device needs ?key=... - after that a 90-day cookie keeps you signed in. The same key works for every /d/* URL.

Why the data was stale

Gmail thread classification depends on a Mac-dependent pipeline:

• The Gmail MCP only exists inside your Claude Desktop session - GitHub Actions cannot reach it.
• Each morning a Cowork scheduled task on your Mac pulls last 24h of threads, writes state/inbox_threads_raw.json, commits, and pushes.
• GitHub Actions inbox-triage.yml then classifies (NEEDS REPLY / APPROVAL / FYI / LIKELY AUTOMATED, importance 1-5) and regenerates the dashboard.
• If your Mac was asleep or Claude Desktop was closed at 5:30 AM ET, no new data was written and the workflow used synthetic sample data instead.

The "freshness" you see on the live page reflects the most recent successful Mac-side pull. The dashboard intentionally surfaces staleness instead of pretending to be live.

The fix - a daily Cowork scheduled task

A new scheduled task spec lives at scheduled-tasks/inbox-refresh.json in the repo. It runs daily at 5:30 AM ET (before the 6 AM morning digest):

1. Calls Gmail MCP search_threads with query newer_than:1d in:inbox.
2. For each thread, calls get_thread for full headers + body.
3. Writes state/inbox_threads_raw.json and pushes.
4. Triggers inbox-triage.yml via gh workflow run to classify immediately.

See MANUAL-STEPS.md for the one-time Claude Desktop sidebar setup (single copy-paste block). After that, this dashboard refreshes automatically every morning - as long as your Mac is on at 5:30 AM ET.