Gateway Keeper

OS-level watchdog for OpenClaw gateway. Runs outside the gateway process so it survives crashes.

How It Works

1. Health check — Runs openclaw gateway status every 15 minutes
2. Auto-restart — If gateway is down, runs openclaw gateway start
3. Recovery signal — Writes logs/gateway-recovery.json with crash timestamp
4. Session recovery — HEARTBEAT.md template detects recovery file, prompts agent to check incomplete work

Quick Setup

Install

Run the appropriate install script for your OS:

Windows (PowerShell as Admin):

powershell -ExecutionPolicy Bypass -File "<skill-dir>/scripts/install.ps1"

Linux/macOS:

bash "<skill-dir>/scripts/install.sh"

Uninstall

Windows:

powershell -ExecutionPolicy Bypass -File "<skill-dir>/scripts/uninstall.ps1"

Linux/macOS:

bash "<skill-dir>/scripts/uninstall.sh"

Recovery Protocol

After gateway restart, the agent should check logs/gateway-recovery.json:

{
  "crashed_at": "2026-02-26T00:00:00Z",
  "restarted_at": "2026-02-26T00:15:00Z",
  "restarted_by": "gateway-keeper"
}

Add to HEARTBEAT.md (done automatically by install script):

## Gateway Crash Recovery
If `logs/gateway-recovery.json` exists:
1. Read crash timestamp
2. List all active sessions/sub-agents
3. Check each for incomplete work
4. Resume or retry as needed
5. Delete the recovery file when done

Files

| File | Purpose | |------|---------| | scripts/check-gateway.ps1 | Windows health check + restart | | scripts/check-gateway.sh | Linux/macOS health check + restart | | scripts/install.ps1 | Windows Task Scheduler setup | | scripts/install.sh | Linux/macOS cron/systemd setup | | scripts/uninstall.ps1 | Windows cleanup | | scripts/uninstall.sh | Linux/macOS cleanup |

Customization

Edit check interval by modifying the scheduled task/cron entry. Default: 15 minutes.

To change recovery behavior, edit the HEARTBEAT.md recovery section.