Crusty Security 🛡️

On-host security monitoring for OpenClaw AI agents. Scans files and skills for malware. Monitors agent behavior for compromise indicators. Audits host security posture.

---

Why Your Agent Needs This

AI agents download files, install skills, and execute code — all with your system privileges. A single prompt injection can lead to:

• 🦠 Malware execution via downloaded files or malicious skill scripts
• 🔗 Data exfiltration through hidden webhook calls or reverse shells
• 🧬 Supply chain attacks from compromised ClawHub skills
• 🔑 Credential theft from exposed .env files and API keys
• 🧠 Agent hijacking via modified SOUL.md, AGENTS.md, or MEMORY.md

Crusty Security is the first security skill built specifically for the OpenClaw agent threat model.

Features

Feature	Description
File Scanning	ClamAV local scan with up-to-date signatures
Skill Auditing	Static analysis for reverse shells, crypto miners, data exfiltration, obfuscation
Host Audit	Cron jobs, open ports, SSH keys, file permissions, posture scoring (0-100)
Agent Monitoring	Detects modified config files, suspicious processes, unexpected outbound connections
ClawHub Sync	Tracks installed skill versions against ClawHub catalog, blocklist checking
Quarantine	Isolate threats with manifest tracking, never auto-deletes
Reports	Markdown security posture reports with recommendations

Quick Start

1. Install

clawhub install crusty-security

That's it. ClamAV is auto-installed on first scan if it's not already present. No separate setup step needed.

2. Start Scanning

# Scan a file
bash scripts/scan_file.sh /path/to/suspicious-file.pdf

# Scan your entire workspace
bash scripts/scan_file.sh -r /data/workspace

# Audit a skill before installing
bash scripts/audit_skill.sh /path/to/skill/

# Full host security audit
bash scripts/host_audit.sh

That's it. Crusty Security works immediately with ClamAV — no API keys required.

How It Works: Skill + Dashboard

Crusty Security has two parts that work together — but the skill works great on its own too.

┌─────────────────────────────┐         ┌──────────────────────────────┐
│   YOUR OPENCLAW AGENT       │         │   CRUSTY SECURITY DASHBOARD  │
│   (your machine / VPS)      │         │   (crustysecurity.com)       │
│                             │  HTTPS  │                              │
│  Crusty Security Skill      │ ──────► │  Web Dashboard               │
│  ✓ Scans files locally      │  POST   │  ✓ View all scan results     │
│  ✓ Audits skills            │         │  ✓ Manage alerts & threats   │
│  ✓ Monitors agent behavior  │         │  ✓ Track agent health        │
│  ✓ Checks host security     │         │  ✓ Multi-agent fleet view    │
│                             │         │  ✓ Email/Slack notifications │
└─────────────────────────────┘         └──────────────────────────────┘
        runs on YOUR machine               runs at crustysecurity.com

Key concepts:

• The skill does all the actual security work — scanning, auditing, monitoring. It runs on your OpenClaw agent's machine.
• The dashboard is a web app where you view results, manage alerts, and monitor multiple agents from one place.
• Data flows one way only: skill → dashboard. The dashboard never connects to your agent. Works behind firewalls, NATs, VPNs — no port forwarding needed.
• The skill is fully functional without the dashboard. The dashboard is optional for users who want centralized visibility.

Connecting Your Agent to the Dashboard (3 steps, 2 minutes)

Step 1: Sign up at crustysecurity.com, go to Agents → + Add Agent. Your API key is generated automatically — copy it.

Step 2: The Agents page shows a full copy-pasteable instruction block. Paste it to your OpenClaw agent, or manually set:

export CRUSTY_API_KEY="cg_live_xxxxxxxxxxxxxxxxxxxx"
export CRUSTY_DASHBOARD_URL="https://crustysecurity.com"

Step 3: Restart your OpenClaw agent. It appears in the dashboard within 5 minutes.

What works without the dashboard

Feature	Without Dashboard	With Dashboard
File scanning (ClamAV)	✅	✅ + history
Skill auditing	✅	✅ + tracking
Host security audit	✅	✅ + trends
Agent monitoring	✅	✅ + alerts
Quarantine	✅	✅ + remote view
Email/Slack alerts	❌	✅
Multi-agent fleet view	❌	✅
Scan scheduling from web	❌	✅
Security posture score	Local only	✅ aggregated

For OpenClaw agents (reading this as SKILL.md)

If you're an OpenClaw agent and your human has configured CRUSTY_API_KEY, you should:

1. Include --push when running scans to send results to the dashboard
2. The heartbeat runs automatically every 5 minutes when the key is set
3. Use python3 scripts/clawhub_sync.py --push to sync skill inventory to dashboard

All Commands

Command	Description
bash scripts/install_clamav.sh	Install and configure ClamAV
bash scripts/scan_file.sh <path>	Scan a file with ClamAV
bash scripts/scan_file.sh -r <dir>	Recursive directory scan
bash scripts/scan_file.sh --incremental -r <dir>	Skip unchanged files
bash scripts/scan_file.sh --quarantine <path>	Quarantine a file
bash scripts/audit_skill.sh <dir>	Audit a skill for threats
bash scripts/host_audit.sh	Host security audit
bash scripts/host_audit.sh --deep	Deep host audit (includes file modifications)
bash scripts/monitor_agent.sh	Agent behavior integrity check
bash scripts/generate_report.sh	Generate security posture report
python3 scripts/clawhub_sync.py	Sync installed skills against ClawHub catalog

All commands output JSON. All support --help.

Scanning Stack

                    ┌──────────────────┐
   File arrives →   │  ClamAV (local)  │  ← Free, instant, signature-based detection
                    └──────────────────┘

ClamAV handles zip, rar, 7z, tar, gz archives natively. Encrypted archives are flagged as "unscanned."

Skill Auditing — What It Catches

Static analysis specifically tuned for the OpenClaw threat model:

Severity	Pattern
🔴 Critical	curl | sh, reverse shell patterns, crypto mining indicators
🟠 High	eval/exec with dynamic input, base64 decode chains, webhook.site/ngrok exfil, credential harvesting, binaries in skill dirs
🟡 Medium	Hidden files, system file access, hardcoded IPs, obfuscated code, persistence mechanisms (cron, systemd)
🔵 Info	Large skill size, credential references in docs

Host Audit Scoring

The host audit produces a posture score from 0-100:

Score	Rating	Meaning
90-100	🟢 Excellent	Minimal risk
70-89	🟡 Good	Minor issues to address
50-69	🟠 Fair	Several findings, take action
0-49	🔴 Poor	Significant security issues

Deductions: Critical (-25), High (-15), Medium (-10), Low (-5).

Agent Behavior Monitoring

Detects indicators of agent compromise:

• Modified AGENTS.md, SOUL.md, MEMORY.md, TOOLS.md (config tampering)
• Unexpected cron jobs or scheduled tasks
• Suspicious outbound connections (IRC, Tor, backdoor ports)
• Files created outside workspace (/tmp executables, home directory changes)
• Suspicious processes (crypto miners, netcat listeners, tunneling tools)
• Exposed credentials (world-readable .env files, SSH keys)

ClawHub Supply Chain Monitoring

The clawhub_sync.py script protects against malicious or compromised skills:

• Fetches the full ClawHub catalog (400+ skills)
• Compares installed skill versions against latest
• Checks against a blocklist of known-bad skills
• Flags skills not found on ClawHub (potential forks or custom builds)
• Detects version drift across multiple agents
• Pushes results to the dashboard (if configured)

# One-time sync
python3 scripts/clawhub_sync.py

# JSON output (for automation)
python3 scripts/clawhub_sync.py --json

# With dashboard push
python3 scripts/clawhub_sync.py --push

Automatic Scan Schedule

Crusty Security automatically configures recurring scans when your OpenClaw agent first uses the skill. No manual setup needed. The agent sets up:

Frequency	What runs	Requires Dashboard?
Every 5 min	Dashboard heartbeat (agent status)	✅
Daily 3am	Incremental workspace scan + agent integrity check	❌
Weekly Sunday 3am	Full workspace scan + host audit + all skills audit + security report	❌
Every 12 hours	ClawHub skill inventory sync	✅
Monthly 1st	Deep host security audit	❌

You can adjust schedules by asking your agent to modify the cron jobs.

Requirements

• OS: Linux (tested on Debian/Ubuntu, works in Docker)
• Python: 3.8+
• ClamAV: Installed via install_clamav.sh or manually
• Disk: ~300MB for ClamAV signatures

Raspberry Pi / Low Memory

• <2GB RAM: Runs in on-demand mode (no ClamAV daemon)
• <1GB RAM: Use skill auditing + agent monitoring (lightweight shell/Python scripts)

Environment Variables

Variable	Default	Required	Description
CRUSTY_API_KEY	—	No	Dashboard API key (from crustysecurity.com)
CRUSTY_DASHBOARD_URL	—	No	Dashboard URL
CLAWGUARD_QUARANTINE	/tmp/clawguard_quarantine	No	Quarantine directory
CLAWGUARD_LOG_DIR	/tmp/clawguard_logs	No	Scan log directory
CLAWGUARD_MAX_FILE_SIZE	200M	No	Max file size for scanning
CLAWGUARD_WORKSPACE	/data/workspace	No	Agent workspace path

File Structure

clawguard/
├── SKILL.md              # Agent instructions (OpenClaw reads this)
├── README.md             # Human documentation (you're reading it)
├── LICENSE.txt           # MIT License
├── CHANGELOG.md          # Version history
├── scripts/
│   ├── install_clamav.sh     # ClamAV installer
│   ├── scan_file.sh          # File/directory scanner
│   ├── audit_skill.sh        # Skill static analysis
│   ├── host_audit.sh         # Host security audit
│   ├── monitor_agent.sh      # Agent behavior monitoring
│   ├── generate_report.sh    # Security report generator
│   ├── clawhub_sync.py       # ClawHub catalog sync
│   └── dashboard.sh          # Dashboard integration library
└── references/
    ├── setup.md              # Detailed setup guide
    ├── threat-patterns.md    # Threat pattern database
    └── remediation.md        # Incident response procedures

Offline Mode

Crusty Security works fully offline with reduced capability:

Feature	Offline	Online
ClamAV file scanning	✅ (local signatures)	✅ (fresh signatures)
Skill auditing	✅ (static analysis)	✅
Host auditing	✅	✅
Agent monitoring	✅	✅
ClawHub sync	❌	✅

Contributing

Issues and PRs welcome at github.com/silentcool/crusty-security.

License

MIT — see LICENSE.txt.

Links

• 🌐 Dashboard: crustysecurity.com
• 📦 ClawHub: clawhub.com (search "crusty-security")
• 🐙 GitHub: github.com/silentcool/crusty-security
• 🦀 Built by: Black Matter VC