feat(cli): add nemoclaw reconnect command for post-reboot recovery

[prekshivyas]

Description

After a host reboot, users must manually run multiple commands to reconnect to their sandbox: start Docker, restart the OpenShell gateway, check sandbox state, then connect. This is documented in the troubleshooting guide but could be a single command.

A nemoclaw reconnect command would automate the recovery sequence:

1. Verify the container runtime is running
2. Start the OpenShell gateway if it's not running (openshell gateway start --name nemoclaw)
3. Wait for the gateway to become healthy
4. Check sandbox state (openshell sandbox list)
5. Start port forward and connect to the default (or named) sandbox

Context

• Issue Please add restart commands to documentation #469 requested documentation for reconnecting after a reboot — addressed by the docs PR
• PR docs(troubleshooting): explain reconnect flow after host reboot #904 (by @WuKongAI-CMU) also attempted to document this but was closed due to PR limit
• The building blocks already exist in the codebase:
  • sandboxConnect() in bin/nemoclaw.js handles port forwarding and connection
  • bin/lib/onboard.js has gateway start, health check, and CoreDNS fix logic
  • registry.listSandboxes() provides the default sandbox name

Open questions

• Does openshell gateway start --name nemoclaw work correctly when the gateway container exists but is stopped? Or does it require a destroy/recreate cycle?
• Is the CoreDNS fix needed again after a reboot, or only on first setup?
• Should this be a new nemoclaw reconnect command, or should nemoclaw <name> connect be smarter about detecting and recovering from a stopped gateway?

Suggested implementation

async function reconnect(sandboxName) {
  // 1. Check gateway
  const gwInfo = runCaptureOpenshell(["gateway", "info", "-g", "nemoclaw"], { ignoreError: true });
  if (!gwInfo || !gwInfo.includes("Running")) {
    console.log("  Starting OpenShell gateway...");
    runOpenshell(["gateway", "start", "--name", "nemoclaw"]);
    // wait for health
  }
  // 2. Find sandbox
  const name = sandboxName || registry.listSandboxes().defaultSandbox;
  if (!name) {
    console.error("  No sandbox registered. Run `nemoclaw onboard` first.");
    process.exit(1);
  }
  // 3. Connect
  sandboxConnect(name);
}

Related

• Please add restart commands to documentation #469 — original user request for reboot recovery docs
• PR docs(troubleshooting): explain reconnect flow after host reboot #904 — previous docs-only attempt (closed, PR limit)

[WuKongAI-CMU]

Opened PR #960 to add a global nemoclaw reconnect [name] command that reuses the existing gateway recovery and sandbox connect path for post-reboot recovery.

[dknos]

Full post-reboot recovery script + systemd auto-start on WSL2

Until a nemoclaw reconnect command exists, here's a complete startup script we use that handles the full recovery sequence automatically on every boot.

~/start-all.sh — handles:

1. Waits for Docker + OpenShell gateway to be ready (retry loop, up to 5 min)
2. Starts the sandbox if not running
3. Waits for sandbox to reach Ready state (retry loop, up to 3 min)
4. Clears stale session lock files
5. Re-sets the inference provider (survives sandbox recreates)
6. Restores custom files (skills, env, model config) via SSH
7. Starts bridge processes via pm2

#!/usr/bin/env bash
set -uo pipefail

source "$HOME/.nemoclaw_env"
SANDBOX="${SANDBOX_NAME:-my-assistant}"

# Wait for Docker + OpenShell gateway
for i in $(seq 1 30); do
  if openshell gateway status 2>/dev/null | grep -qiE "running|ready|ok"; then break; fi
  if docker info >/dev/null 2>&1; then break; fi
  echo "($i/30) Docker not ready — waiting 10s..."
  sleep 10
done

# Start sandbox if not running
nemoclaw status --json 2>/dev/null | grep -q '"running":true' || nemoclaw start 2>/dev/null &

# Wait for sandbox ready
for i in $(seq 1 36); do
  nemoclaw status --json 2>/dev/null | grep -q '"running":true' && break
  echo "($i/36) Sandbox not ready — waiting 5s..."
  sleep 5
done

# Clear stale lock files
SSH_CONF=$(mktemp); openshell sandbox ssh-config "$SANDBOX" > "$SSH_CONF"
ssh -T -F "$SSH_CONF" "openshell-${SANDBOX}" \
  'find /sandbox/.openclaw-data -name "*.lock" -delete 2>/dev/null'
rm -f "$SSH_CONF"

# Re-set inference
openshell inference set --provider your-provider --model your-model --no-verify 2>/dev/null || true

# Start bridges
pm2 start ecosystem.config.js
pm2 save

Auto-start on WSL boot via systemd user service (~/.config/systemd/user/nemoclaw.service):

[Unit]
Description=NemoClaw AI Assistant
After=network.target

[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/home/%u/start-all.sh
TimeoutStartSec=600

[Install]
WantedBy=default.target

Enable with:

systemctl --user enable nemoclaw.service
loginctl enable-linger $USER

This gives full hands-off recovery after every reboot — no manual steps needed.