/reboot — a controlled restart of a fleet node

The pain: rebooting blind = a lost unsynced file (sync-conflict), a false SEV1 alarm from the peer watchdogs ("the node is dead!"), a robot that never came back after boot, and — as with the HP Wolf driver on 2026-07-24 — "I rebooted and the driver is still in memory", because Fast Startup does not clear RAM on an ordinary shutdown.

Reboot ONLY when: (a) there is a reason (evict a hook/driver from RAM, apply an update, unstick something), and (b) autostart is armed → the machine will come back on its own.

STEP 1 — PRE-FLIGHT (before the reboot, in order)

1. Reason + confirmation. One line: why the reboot and what it fixes. If it is risky (away-mode, a machine without auto-login, FileVault/BitLocker with a manual password at boot) → ask the operator first.
2. Save state. The TurnState black box writes on every turn by itself (it survives the reboot → /1 will bring it back). Any active uncommitted work in code or the vault → commit/save it now.
3. Finish the Syncthing sync (otherwise: loss/conflict). Wait for need=0 on every share:

   powershell -NoProfile -ExecutionPolicy Bypass -File "$IMPORTS_ROOT/sync_check/sync_check.ps1"

   need>0 → wait for the sync / raise it (/raise-sync); do NOT reboot with a pending need.
4. Warn the peers (otherwise the observers raise a false offline SEV1) — into the bus + the fleet log chat, BEFORE the reboot:

   python "$HOME/.claude/scripts/bus_send.py" --text "🔁 <host> going down for a reboot (~N min), reason: <...>. I'll come back on my own."

5. Verify autostart is armed (the machine must come back WITHOUT hands):
   • PC: Claude Desktop = a single launcher %APPDATA%\...\Startup\Claude-Autostart.lnk (shell:AppsFolder\Claude_pzs8sxrjxfjjc!Claude), NOT a broken HKCU\Run pointing at an old version ([[claude-desktop-autostart-race]]); Syncthing = start-syncthing.vbs in Startup; the watchdog tasks in place. On the hub, also auto-login + hub_boot_report.py ([[hub-boot-selfreport]]).
   • Mac: Login Items / launchd agents for Syncthing + Claude; a FileVault password at boot is manual (without it an unattended reboot will not come back — only do it with a human present).

STEP 2 — THE REBOOT (with the right command)

⚠️ A FULL Restart, not a Shutdown — Fast Startup (hiberboot) on Windows does NOT clear the kernel/drivers/RAM on shutdown /s; Restart (/r) always does a full cycle and clears them. To "evict a driver/hook from memory" (HP Wolf, an antivirus, a stuck driver) ONLY a full Restart works.

• PC (Windows):

  shutdown /r /t 0

  (a guaranteed clean exit even with Fast Startup ON; the Claude Code session ends with it).
• Mac:

  osascript -e 'tell app "System Events" to restart'   # or: sudo shutdown -r now

STEP 3 — POST-REBOOT (once it's up — through /1)

The machine is back → in a new session run /1 (resurrection): RECALL from the black box + a health ping (arch/sync/mcp) + the previous session's full history into the buffer. Then:

• Confirm autostart fired: Syncthing connected, Claude Desktop = a single instance (no race), connectors green.
• Check that the reason for the reboot was actually achieved (e.g. the black windows are gone / the driver is no longer in memory) — by looking or measuring, not on faith ([[prichina-kak-claim]]).
• A boot self-report into the bus (the hub does it itself; on a peer — a short "✅ <host> is back, all green").

Boundaries

• READ-only right up to the reboot itself; the reboot is a deliberate action with a stated reason.
• ⛔ Away-mode / no auto-login / a manual FileVault-BitLocker prompt at boot → reboot only with the operator's approval (risk: it never comes back without hands).
• A peer reboots ITSELF ([[peers-own-outbound-local-full-member]]); rebooting the hub while away — carefully ([[away-mode-45-days]], BIOS "Restore on AC Power Loss").
• The folder name is reboot; /restart is a text trigger for the same skill (case does not matter).

Like this skill? It is one of 100 in second-brain-starter-kit: the second brain we built for ourselves and run every day at Palo Alto AI Research Lab. Install the whole set with npx skills add tonydzi/second-brain-starter-kit. Everything is open source and free, so take what you need.

Flagships worth a look on their own: secondop-panel (a second opinion from a panel of external models), claude-memory-tidy (stop your agent's memory from rotting), telegram-mcp-kit (your own Telegram over MCP in about 15 minutes).

Author: Anton Dziatkovskii, Palo Alto AI Research Lab. Telegram @tonydzi - WhatsApp +1 341 222 9178 - X @Tony_Stef_

Engineers: want to test-drive this setup? Message me. I hand out free starter seeds to engineers who test and report back, and custom skill requests are welcome.