daily-startlisted

You open the working day for one orchestrator. You run once at session start. You are one half of a pair. `close-day` leaves one index behind it; you read that index and follow its ids. A handover produced every evening and never read the next morning is the cheapest waste in the fleet, and reading the WRONG one is worse — it looks like a normal day. Two modes: - **Human mode (Pi on the Chromebook)** — reports to the operator and advances the standing objective. - **Autonomous mode (every other orchestrator)** — chooses its own next task, proposes it to pi, and waits for a confirmation or a redirection before starting. **Every count you display is derived from a command whose output you paste. None is typed.** --- ## Step 1 — Identity and date (silent) Read the first 20 lines of `CLAUDE.md` for the orchestrator role. Derive `instanceId` from the hostname: `{role}-vps` or `{role}-chromebook`. Run `date -I` for the date. Human mode iff the header says "You are Pi" AND the workspace is the Chromebook Pi one. Otherwise autonomous. In doubt, autonomous — a silent orchestrator is a safer failure than one that interrupts a human. ## Step 2 — Register the message cron (autonomous and human alike) Crons do not survive a restart in this environment, so they are re-registered every session, here, once: ``` CronCreate cron='*/10 * * * *' prompt='/check-messages' recurring=true durable=true ``` Ten minutes. If a cron with that prompt is already live in this session, do not cr