The model: Departments don’t build automations themselves. Each dept’s only interface is its Telegram bot. To automate something, the dept hands EJ/operator the inputs, and EJ provisions it once in OpenClaw — then it runs hands-off.
What a department provides (the 4 inputs):
- Source — the exact Google Sheet ID + tab (or where the data lives) OpenClaw should read.
- Trigger — when it should run: a clock time (e.g. daily 6pm CT) or on the dashboard’s 15-min refresh.
- Output + delivery — what it produces and where it goes: a Telegram message, a dashboard panel, a written file.
- Recipients / approval — who receives it, and who signs off before it goes live.
What EJ / operator does: adds the source read + logic in OpenClaw config, sets the cron trigger, points delivery at the right Telegram bot/channel, approves + tests once. (On this VM, provisioning = operator task; depts never touch config.)
Good first automations to teach with: scheduled EOD reminders to non-filers · auto-refreshed dashboards · daily summary posts · flagging pending items · report/SOP drafting. Each rec above has a “🤖 set up OpenClaw” expander with its specific inputs.
🔒 Safety: read-only on source sheets by default; external sends need approval; one dept never sees another dept’s data.