rdmsm4x — Tyrell 0.2.0 build 7 on all six hosts; hub daemon 5c8f1b1 (rescan cadence, diagnosed store exit, headless --version)

2026-09-02 01:28 → 2026-09-02 01:40:01 EDT · rdmsm4x · Claude Code (session tyrell-c5), unattended — third and last iteration of the overnight loop

Changed

• Rescan cadence for every role (FEAT-20260902-01): TyrellConfig.rescanIntervalSeconds (default 3600, floor 300); ClientLoop.inventoryLoop replaces the scan-once-at-launch behaviour. Live on the hub; the five spokes still run their Sep 1 daemons until deploy_fleet.zsh runs in daytime. The build-6 self-inventory already completed a full pass on the hub and pruned 10,265 vanished rows; the hub's fleet row went from 4,388 min stale to current.
• Diagnosed exit on an unreadable store (ISSUE-20260902-07): counts(host:) throws instead of force-unwrapping; tyrelld exits 78 naming the DB path. Verified against a deliberately corrupted copy (64 KiB zeroed): exit 78 with the message, where build 5 trapped silently.
• Tyrell --version headless (Tyrell 0.2.0 (7)), because a bare --version launched a second GUI instance twice tonight.
• App build 7 on 6/6 hosts via the canary pipeline (probe PASS, receipt accepted, pending=none failed=none) and local install; every spoke verified through the new flag. Hub daemon releases/5c8f1b1 installed with Step 4 PASS.

Tests

248 + 65 + 51 = 364, 0 failures (360 at the start of the night).

Tickets

Resolved: ISSUE-20260902-07, TASK-20260902-04. Open with status recorded: FEAT-20260902-01 (spokes pending daytime). For other owners: ISSUE-20260902-06 (ecs0lib), FEAT-20260831-03 (project rooms).

Undo

Hub daemon: install_service.zsh --server --binary "~/Library/Application Support/Tyrell/releases/8fd31b1/tyrelld". App: per-host rollback dirs under /Applications/.tyrell-rollbacks/.

No secrets recorded.