Claw Sos

Install, run, and manage the SOS emergency recovery tool for OpenClaw instances. Use when (1) the bot stops responding on Telegram/Discord and needs diagnosis or recovery, (2) you need to install SOS on a new or existing machine, (3) autofix, rollback, network check, or Telegram test is needed, (4) the user mentions "sos", "recovery", "bot is down", "not responding", or "emergency fix".

Emergency recovery tool for OpenClaw. Diagnoses and fixes unresponsive bots via SSH.

Install

The script is bundled at scripts/sos.sh. Install locally:

bash

cp scripts/sos.sh /usr/local/bin/sos
chmod +x /usr/local/bin/sos

To install on a remote machine:

bash

scp scripts/sos.sh root@<IP>:/usr/local/bin/sos
ssh root@<IP> "chmod +x /usr/local/bin/sos"

Alternative — install from GitHub:

bash

curl -fsSL https://raw.githubusercontent.com/clawsos/claw-sos/main/install.sh | bash

Usage

Interactive (human via SSH)

bash

sos          # whiptail arrow-key menu

Non-interactive (agent or cron)

bash

sos auto     # autofix: diagnose → doctor → restart → force → nuclear
sos net      # network check only
sos tg       # telegram test message
sos --version
sos --help

When the bot is down — decision tree

1. Run sos auto — handles 95% of cases automatically
2. If autofix fails → run sos net — if network is broken, fix DNS/internet first
3. If network is fine but bot still down → run sos tg — check Telegram delivery
4. If Telegram fails → check bot token validity
5. Last resort → sos menu option 9 (Nuclear) — kills everything and starts fresh

Menu Options

#	Name	What it does	Safe?
1	Check status	Gateway running, RAM, disk, version	✅ Read-only
2	Restart	Graceful restart	✅ Safe
3	Force kill	Kill process + restart	⚠️ Drops active sessions
4	Rollback	Revert to previous version + config	⚠️ Downgrade
5	View logs	Last 50 lines of gateway log	✅ Read-only
6	Full diagnostic	RAM, disk, Telegram, sessions, process	✅ Read-only
7	Backup config	Save config + version before changes	✅ Safe
8	Self-test	Verify SOS script works	✅ Read-only
9	Nuclear	Kill ALL openclaw processes, reload	🔴 Last resort
10	Autofix	Escalating auto-repair	⚠️ May restart
11	Network check	DNS, internet, Telegram API, Anthropic	✅ Read-only
12	Telegram test	Send real test message	✅ Safe

Autofix Escalation Order

1. Diagnose: process, RAM, disk
2. Clean RAM/disk if critical
3. openclaw doctor --fix
4. Graceful restart → wait 15s → check
5. Force kill + restart → wait 15s → check
6. Nuclear (kill all + reload)

Stops as soon as gateway is healthy.

Log Location

All actions logged to ~/.openclaw/backups/sos.log. Read this to understand what happened during recovery:

bash

cat ~/.openclaw/backups/sos.log

Platform Support

• Linux: Full support (systemd, journalctl, free, /proc)
• macOS: Full support (launchctl, log show, vm_stat, purge, dialog fallback)
• Docker: Partial — no systemd, manual fallback works

Credentials

This skill uses no API keys or tokens of its own. It reads existing OpenClaw config to:

• Check gateway status (local process, no network)
• Send Telegram test messages (uses the bot token already configured in openclaw.json)
• Check Anthropic API reachability (HTTPS ping only, no auth)

No credentials are stored, transmitted, or required to install or run SOS.

Source

GitHub: https://github.com/clawsos/claw-sos