secucheck - OpenClaw Security Audit

Comprehensive security audit skill for OpenClaw deployments. Analyzes configuration, permissions, exposure risks, and runtime environment with context-aware recommendations.

---

Summary

secucheck performs read-only security audits of your OpenClaw setup:

• 7 audit domains: Runtime, Channels, Agents, Cron Jobs, Skills, Sessions, Network
• 3 expertise levels: Beginner (analogies), Intermediate (technical), Expert (attack vectors)
• Context-aware: Considers VPN, single-user, self-hosted scenarios
• Runtime checks: Live system state (network exposure, containers, privileges)
• Dashboard: Visual HTML report with security score
• Localized output: Final report matches user's language

Never modifies configuration automatically. All fixes require explicit user confirmation.

---

Quick Start

Installation

clawhub install secucheck

Usage

Ask your OpenClaw agent:

• "security audit"
• "secucheck"
• "run security check"

Expertise Levels

When prompted, choose your level:

1. Beginner - Simple analogies, no jargon
2. Intermediate - Technical details, config examples
3. Expert - Attack vectors, edge cases, CVEs

All levels run the same checks—only explanation depth varies.

Dashboard

"show dashboard" / "visual report"

Opens an HTML report in your browser.

---

Example Output

🔒 Security Audit Results

🟡 Needs Attention

| Severity | Count |
|----------|-------|
| 🔴 Critical | 0 |
| 🟠 High | 0 |
| 🟡 Medium | 2 |
| 🟢 Low | 3 |

### 🟡 Agent "molty": exec + external content processing
...

---

Features

• 🔍 Comprehensive: Channels, agents, cron, skills, sessions, network, runtime
• 👤 3 Expertise Levels: Beginner / Intermediate / Expert
• 🌏 Localized: Final report in user's language
• 🎯 Attack Scenarios: Real-world exploitation paths
• ⚡ Runtime Checks: VPN, containers, privileges, network exposure
• 🎨 Dashboard: Visual HTML report with security score

---

Agent Instructions

Everything below is for the agent executing this skill.

---

When to Use

Trigger this skill when:

• User requests security checkup/audit
• Auto-trigger: Installing skills, creating/modifying agents, adding/modifying cron jobs
• Periodic review (recommended: weekly)

Expertise Levels

Level	Identifier	Style
Beginner	1, beginner	Analogies, simple explanations, no jargon
Intermediate	2, intermediate	Technical details, config examples
Expert	3, expert	Attack vectors, edge cases, CVE references

Execution Flow

Step 1: Ask Level (before running anything)

Present options in user's language. Example (English):

What level of technical detail do you prefer?

1. 🌱 Beginner - I'll explain simply with analogies
2. 💻 Intermediate - Technical details and config examples
3. 🔐 Expert - Include attack vectors and edge cases

📌 All levels run the same checks—only explanation depth varies.

STOP HERE. Wait for user response.

Step 2: Run Audit

bash ~/.openclaw/skills/secucheck/scripts/full_audit.sh

Returns JSON with findings categorized by severity.

Step 3: Format Output

Parse JSON output and format based on user's expertise level. Final report must be in user's language.

Report Structure (Organize by Category)

🔒 Security Audit Results

📊 Summary Table
| Severity | Count |
|----------|-------|
| 🔴 Critical | X |
| ...

⚡ Runtime
- [findings related to RUNTIME category]

🤖 Agents  
- [findings related to AGENT category]

📁 Workspace
- [findings related to WORKSPACE category]

🧩 Skills
- [findings related to SKILL category]

📢 Channels
- [findings related to CHANNEL category]

🌐 Network
- [findings related to NETWORK category]

Group findings by their category field, not just severity. Within each category, show severity icon and explain.

Step 4: Auto-Open Dashboard

After text report, automatically generate and serve dashboard:

bash ~/.openclaw/skills/secucheck/scripts/serve_dashboard.sh

The script returns JSON with url (LAN IP) and local_url (localhost). Use the url field (not localhost) when telling the user — they may access from another device.

Example:

📊 대시보드도 열었어요: http://192.168.1.200:8766/secucheck-report.html

If running in environment where browser can be opened, use browser tool to open it.

Cross-Platform Support

Scripts run on Linux, macOS, and WSL. Check the JSON output for platform info:

{
  "os": "linux",
  "os_variant": "ubuntu",
  "in_wsl": false,
  "in_dsm": false,
  "failed_checks": ["external_ip"]
}

Platform Detection

Field	Values
os	linux, macos, windows, unknown
os_variant	ubuntu, arch, dsm, wsl, version string
in_wsl	true if Windows Subsystem for Linux
in_dsm	true if Synology DSM

Handling Failed Checks

If failed_checks array is non-empty, run fallback commands based on platform:

Network Info Fallbacks

Platform	Command
Linux	ip addr show or ifconfig
macOS	ifconfig
WSL	ip addr show (or check Windows via cmd.exe /c ipconfig)
Windows	PowerShell: Get-NetIPAddress
DSM	ifconfig or /sbin/ip addr

Gateway Binding Fallbacks

Platform	Command
Linux	ss -tlnp | grep :18789 or netstat -tlnp
macOS	lsof -iTCP:18789 -sTCP:LISTEN
Windows	PowerShell: Get-NetTCPConnection -LocalPort 18789

File Permissions Fallbacks

Platform	Command
Linux/macOS	ls -la ~/.openclaw
Windows	PowerShell: Get-Acl $env:USERPROFILE\.openclaw

Windows Native Support

If os is windows and scripts fail completely:

1. Use PowerShell commands directly:

# Network exposure
Get-NetTCPConnection -LocalPort 18789 -State Listen

# File permissions
Get-Acl "$env:USERPROFILE\.openclaw"

# Process info
Get-Process | Where-Object {$_.Name -like "*openclaw*"}

2. Report what you can check and note Windows-specific limitations.

Minimal Environments (Docker, DSM)

Some environments lack tools. Check output and supplement:

Missing Tool	Fallback
curl	wget -qO-
ss	netstat
ip	ifconfig or /sbin/ip
pgrep	ps aux | grep

Agent Decision Flow

1. Run full_audit.sh
2. Check "failed_checks" in output
3. For each failed check:
   a. Identify platform from os/os_variant
   b. Run platform-specific fallback command
   c. Incorporate results into report
4. Note any checks that couldn't complete

Dashboard Generation

When user requests visual report:

bash ~/.openclaw/skills/secucheck/scripts/serve_dashboard.sh

Returns:

{
  "status": "ok",
  "url": "http://localhost:8766/secucheck-report.html",
  "pid": 12345
}

Provide URL directly to user.

Detailed Check References

Read these only when deep explanation needed:

File	Domain
checks/runtime.md	Live system state
checks/channels.md	Channel policies
checks/agents.md	Agent permissions
checks/cron.md	Scheduled jobs
checks/skills.md	Installed skills
checks/sessions.md	Session isolation
checks/network.md	Network configuration

Attack Scenario Templates

Use these for expert-level explanations:

File	Scenario
scenarios/prompt-injection.md	External content manipulation
scenarios/session-leak.md	Cross-session data exposure
scenarios/privilege-escalation.md	Tool permission abuse
scenarios/credential-exposure.md	Secret leakage
scenarios/unauthorized-access.md	Access control bypass

Risk Levels

🔴 Critical - Immediate action required. Active exploitation possible.
🟠 High     - Significant risk. Should fix soon.
🟡 Medium   - Notable concern. Plan to address.
🟢 Low      - Minor issue or best practice recommendation.
⚪ Info     - Not a risk, but worth noting.

Risk Matrix

                Tool Permissions
              Minimal       Full
         ┌──────────┬──────────┐
Exposure │   🟢     │   🟡     │
  Low    │  Safe    │  Caution │
         ├──────────┼──────────┤
         │   🟡     │   🔴     │
  High   │ Caution  │ Critical │
         └──────────┴──────────┘

Exposure = Who can talk to the bot (DM policy, group access, public channels)
Tool Permissions = What the bot can do (exec, file access, messaging, browser)

Context-Aware Exceptions

Don't just pattern match. Consider context:

Context	Adjustment
Private channel, 2-3 trusted members	Lower risk even with exec
VPN/Tailscale only access	Network exposure less critical
Self-hosted, single user	Session isolation less important
Containerized environment	Privilege escalation less severe

Always ask about environment if unclear.

Applying Fixes

CRITICAL RULES:

1. Never auto-apply fixes. Always show suggestions first.
2. Warn about functional impact. If a fix might break something, say so.
3. Get explicit user confirmation before any config changes.

Example flow:

Agent: "Changing this setting will disable exec in #dev channel.
        If you're using code execution there, it will stop working.
        Apply this fix?"
User: "yes"
Agent: [apply fix via gateway config.patch]

Language Rules

• Internal processing: Always English
• Thinking/reasoning: Always English
• Final user-facing report: Match user's language
• Technical terms: Keep in English (exec, cron, gateway, etc.)

Auto-Review Triggers

Invoke automatically when:

1. Skill installation: clawhub install <skill> or manual addition
2. Agent creation/modification: New agent or tool changes
3. Cron job creation/modification: New or modified scheduled tasks

For auto-reviews, focus only on changed component unless full audit requested.

Quick Commands

User Request	Action
"check channels only"	Run channels.md check
"audit cron jobs"	Run cron.md check
"full audit"	All checks
"more detail"	Re-run with verbose output

Trust Hierarchy

Apply appropriate trust levels:

Level	Entity	Trust Model
1	Owner	Full trust — has all access
2	AI Agent	Trust but verify — sandboxed, logged
3	Allowlists	Limited trust — specified users only
4	Strangers	No trust — blocked by default

Incident Response Reference

If compromise suspected:

Containment

1. Stop gateway process
2. Set gateway.bind to loopback (127.0.0.1)
3. Disable risky DM/group policies

Rotation

1. Regenerate gateway auth token
2. Rotate browser control tokens
3. Revoke and rotate API keys

Review

1. Check gateway logs and session transcripts
2. Review recent config changes
3. Re-run full security audit

Files Reference

~/.openclaw/skills/secucheck/
├── SKILL.md              # This file
├── skill.json            # Package metadata
├── README.md             # User documentation
├── scripts/
│   ├── full_audit.sh     # Complete audit (JSON output)
│   ├── runtime_check.sh  # Live system checks
│   ├── gather_config.sh  # Config extraction (redacted)
│   ├── gather_skills.sh  # Skill security scan
│   ├── gather_agents.sh  # Agent configurations
│   ├── serve_dashboard.sh # Generate + serve HTML report
│   └── generate_dashboard.sh
├── dashboard/
│   └── template.html     # Dashboard template
├── checks/
│   ├── runtime.md        # Runtime interpretation
│   ├── channels.md       # Channel policy checks
│   ├── agents.md         # Agent permission checks
│   ├── cron.md           # Cron job checks
│   ├── skills.md         # Skill safety checks
│   ├── sessions.md       # Session isolation
│   └── network.md        # Network exposure
├── scenarios/
│   ├── prompt-injection.md
│   ├── session-leak.md
│   ├── privilege-escalation.md
│   ├── credential-exposure.md
│   └── unauthorized-access.md
└── templates/
    ├── report.md         # Full report template
    ├── finding.md        # Single finding template
    └── summary.md        # Quick summary template

Security Assessment Questions

When auditing, consider:

1. Exposure: What network interfaces can reach this agent?
2. Authentication: What verification does each access point require?
3. Isolation: What boundaries exist between agent and host?
4. Trust: What content sources are considered "trusted"?
5. Auditability: What evidence exists of agent's actions?
6. Least Privilege: Does agent have only necessary permissions?

---

Remember: This skill exists to make OpenClaw self-aware of its security posture. Use regularly, extend as needed, never skip the audit.