OpenClaw Security Check

Security self-check for OpenClaw deployments. Audits openclaw.json config and host security in one pass: gateway exposure, auth mode, token strength, channel...

Security Scan

The skill's checks, required files, and commands match its stated purpose (config + host audit); it is read-only by default but includes optional auto-fix recipes that modify the system and require sudo, so review and control how fixes are executed before enabling automated use.

Purpose & Capability

The name/description state a fast OpenClaw config + host security audit; the bundled script reads ~/.openclaw/openclaw.json, inspects firewall/SSH/listening ports, and scans channel policies and file permissions — all consistent with the stated checks. There are no unrelated credentials, binaries, or external endpoints requested.

Instruction Scope

The shipped script is read-only and implements the 10 checks described in SKILL.md. However SKILL.md also documents an 'Auto-Fix Flow' with concrete commands (apt install ufw, chmod, sed edits to /etc/ssh/sshd_config, firewall enable, openclaw gateway restart) that modify system state and require sudo. The documentation says to confirm with the user first, but that is a policy-level instruction — the agent could be asked to run fixes unless you enforce confirmation. Also the SKILL.md suggests adding the script to periodic heartbeat/cron which would make the checks automatic; ensure fixes are not automated without explicit approval.

Install Mechanism

No install spec; this is an instruction-only skill with a bundled shell script. That is low-risk from an installation perspective — nothing is downloaded or written during install.

Credentials

The skill requires no environment variables or external credentials. It reads HOME and system files (/etc/ssh/sshd_config, ~/.openclaw/openclaw.json), which is proportionate to auditing host and config. The auto-fix commands require elevated privileges (sudo) which is expected for the kinds of system changes suggested but increases potential impact if executed without supervision.

Persistence & Privilege

always:false (no forced inclusion) and default autonomous invocation is allowed (disable-model-invocation:false). Autonomous invocation alone is normal, but combined with documented auto-fix recipes increases blast radius: if the agent is permitted to run fixes automatically, it could install packages or modify SSH/firewall settings. Prefer to require explicit user confirmation before any fix actions and avoid scheduling auto-fix via heartbeat/cron unless you trust the environment.

Assessment

This skill appears to do what it claims (a quick OpenClaw config + host audit) and the bundled script is readable — good signs. Before installing or enabling automated invocation: 1) Review the full scripts/security-check.sh yourself to verify behavior (it is included). 2) Use the script in read-only/reporting mode first (scripts/security-check.sh or --json) to see findings. 3) Do NOT allow the agent to run auto-fix commands without explicit, interactive confirmation; the fixes use sudo, apt, chmod, sed and can lock you out if misapplied. 4) If you plan to schedule checks, schedule only read-only reports; never auto-apply fixes from cron/heartbeat. 5) Backup ~/.openclaw/openclaw.json and ensure you have a second active SSH session before applying SSH/firewall changes. 6) Prefer running the audit in a staging environment first. If you want, restrict the skill so it can only produce reports and not execute remediation steps autonomously.

Fast 10-point security audit for OpenClaw config + host. Read-only by default, optional auto-fix.

Quick Start

Run the bundled script for a non-interactive report:

scripts/security-check.sh        # human-readable
scripts/security-check.sh --json # structured output

Or tell the agent: "run a security check" / "audit my OpenClaw config".

What It Checks

#	Check	Severity if failed	What it looks at
1	Gateway Bind	CRITICAL	gateway.bind — must be loopback, not 0.0.0.0
2	Gateway Auth	CRITICAL	gateway.auth.mode — must not be off/none
3	Token Strength	HIGH	gateway.auth.token — must be ≥32 chars
4	DM Policy	HIGH	Per-channel dmPolicy — open without allowFrom is dangerous
5	Group Policy	HIGH	Per-channel groupPolicy — open/any allows strangers to trigger the agent
6	Config Permissions	MEDIUM	File mode of openclaw.json — should be 600 or 400
7	Plaintext Secrets	MEDIUM	Scans config values for keys matching password/secret/apiKey/privateKey
8	Host Firewall	HIGH	UFW or firewalld must be installed and active
9	SSH Hardening	MEDIUM	PasswordAuthentication and PermitRootLogin in sshd_config
10	Exposed Ports	MEDIUM	Count of non-loopback listening ports (>8 = FAIL)

Auto-Fix Flow

If any item is FAIL or WARN, offer fixes. Always confirm with the user first.

Fix Recipes

#1 Gateway Bind → FAIL: Set gateway.bind to "loopback". Use openclaw CLI if available, otherwise edit openclaw.json.

#2 Gateway Auth → FAIL: Set gateway.auth.mode to "token". Generate a strong token if missing:

openssl rand -hex 24

#3 Token Strength → FAIL/WARN: Replace with a new 48-char hex token: openssl rand -hex 24. Warn user that paired clients will need the new token.

#4 DM Policy → FAIL: Set affected channels to "dmPolicy": "pairing", or add specific IDs to allowFrom.

#5 Group Policy → FAIL: Set affected channels to "groupPolicy": "allowlist".

#6 Config Permissions → FAIL/WARN:

chmod 600 ~/.openclaw/openclaw.json

#7 Plaintext Secrets → WARN: Cannot auto-fix safely. Advise moving secrets to environment variables or .env.local.

#8 Host Firewall → FAIL:

sudo apt install ufw -y
sudo ufw default deny incoming
sudo ufw default allow outgoing
# IMPORTANT: Allow SSH before enabling!
sudo ufw allow from <trusted_ip_or_subnet> to any port 22 proto tcp
sudo ufw enable

#9 SSH Hardening → WARN:

sudo cp /etc/ssh/sshd_config /etc/ssh/sshd_config.bak
sudo sed -i 's/^#*PasswordAuthentication .*/PasswordAuthentication no/' /etc/ssh/sshd_config
sudo sed -i 's/^#*PermitRootLogin .*/PermitRootLogin no/' /etc/ssh/sshd_config
sudo sshd -t && sudo systemctl reload ssh

CRITICAL: Ensure key-based SSH access works in a separate session before closing current one.

#10 Exposed Ports → WARN/FAIL: Review with ss -ltnp, close unnecessary services, or restrict with firewall rules.

Fix Rules

• Backup first: cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak
• Merge, don't overwrite: Modify only the specific keys, preserve everything else.
• SSH changes need special care: Always test access in a second session before closing the first.
• Firewall: allow SSH first, enable second. Getting this backwards locks you out.
• After config changes: openclaw gateway restart to apply.
• Re-run the check after fixes to confirm everything passes.

Integration

Heartbeat

Add to HEARTBEAT.md for periodic checks:

- Every heartbeat: Run scripts/security-check.sh, alert on any FAIL

Cron

Schedule via OpenClaw cron for standalone audits:

openclaw cron add --name "security-check" --schedule "0 8 * * *" --task "Run scripts/security-check.sh and report results"

Combining with healthcheck skill

This skill focuses on fast config + host audit (10 checks, <5 seconds). The built-in healthcheck skill provides a full hardening workflow (risk profiling, remediation planning, guided execution). Use this skill for quick checks; escalate to healthcheck for comprehensive hardening.