openclaw-security-monitor

<!-- {"requires":{"bins":["bash","curl","node","lsof"],"optionalBins":["witr","docker","openclaw"],"env":{"OPENCLAW_TELEGRAM_TOKEN":"Optional: Telegram bot token for daily security alerts","OPENCLAW_HOME":"Optional: Override default ~/.openclaw directory"}}} -->

# Security Monitor

Real-time security monitoring with threat intelligence from ClawHavoc research, daily automated scans, web dashboard, and Telegram alerting for OpenClaw.

## Commands
Note: Replace `<skill-dir>` with the actual folder name where this skill is installed (commonly `openclaw-security-monitor` or `security-monitor`).

### /security-scan
Run a comprehensive 59-point security scan:
1. Known C2 IPs (ClawHavoc: 91.92.242.x, 95.92.242.x, 54.91.154.110)
2. AMOS stealer / AuthTool markers
3. Reverse shells & backdoors (bash, python, perl, ruby, php, lua)
4. Credential exfiltration endpoints (webhook.site, pipedream, ngrok, etc.)
5. Crypto wallet targeting (seed phrases, private keys, exchange APIs)
6. Curl-pipe / download attacks
7. Sensitive file permission audit
8. Skill integrity hash verification
9. SKILL.md shell injection patterns (Prerequisites-based attacks)
10. Memory poisoning detection (SOUL.md, MEMORY.md, IDENTITY.md)
11. Base64 obfuscation detection (glot.io-style payloads)
12. External binary downloads (.exe, .dmg, .pkg, password-protected ZIPs)
13. Gateway security configuration audit
14. WebSocket origin validation (CVE-2026-25253)
15. Known malicious publisher detection (hightower6eu, etc.)
16. Sensitive environment/credential file leakage
17. DM policy audit (open/wildcard channel access)
18. Tool policy / elevated tools audit
19. Sandbox configuration check
20. mDNS/Bonjour exposure detection
21. Session & credential file permissions
22. Persistence mechanism scan (LaunchAgents, crontabs, systemd)
23. Plugin/extension security audit
24. Log redaction settings audit
25. Reverse proxy localhost trust bypass detection
26. Exec-approvals configuration audit (CVE-2026-25253 exploit chain)
27. Docker container security (root, socket mount, privileged mode)
28. Node.js version / CVE-2026-21636 permission model bypass
29. Plaintext credential detection in config files
30. VS Code extension trojan detection (fake ClawdBot extensions)
31. Internet exposure detection (non-loopback gateway binding)
32. MCP server security audit (tool poisoning, prompt injection)
33. ClawJacked WebSocket brute-force protection (v2026.2.25+)
34. SSRF protection audit (CVE-2026-26322, CVE-2026-27488)
35. Exec safeBins validation bypass (CVE-2026-28363, CVSS 9.9)
36. ACP permission auto-approval audit (GHSA-7jx5)
37. PATH hijacking / command hijacking (GHSA-jqpq-mgvm-f9r6)
38. Skill env override host injection (GHSA-82g8-464f-2mv7)
39. macOS deep link truncation (CVE-2026-26320)
40. Log poisoning / WebSocket header injection
41. Browser Relay CDP unauthenticated access (CVE-2026-28458, CVSS 7.5)
42. Browser control API path traversal (CVE-2026-28462, CVSS 7.5)
43. Exec-approvals shell expansion bypass (CVE-2026-28463)
44. Approval field injection / exec gating bypass (CVE-2026-28466)
45. Sandbox browser bridge auth bypass (CVE-2026-28468)
46. Webhook DoS — oversized payloads (CVE-2026-28478)
47. TAR archive path traversal (CVE-2026-28453)
48. fetchWithGuard memory exhaustion DoS (CVE-2026-29609, CVSS 7.5)
49. /agent/act HTTP route unauthenticated access (CVE-2026-28485)
50. Command hijacking via PATH — unsafe resolution (CVE-2026-29610)
51. SHA-1 sandbox cache key poisoning (CVE-2026-28479, CVSS 8.7)
52. Google Chat webhook cross-account bypass (CVE-2026-28469, CVSS 9.8)
53. Gateway WebSocket device identity skip (CVE-2026-28472)
54. Cross-Site WebSocket Hijacking in trusted-proxy (CVE-2026-32302)
55. Device pairing credential exposure (GHSA-7h7g-x2px-94hj)
56. Operator privilege escalation (GHSA-vmhq-cqm9-6p7q)
57. MCP server tool poisoning via schema injection (OWASP MCP03/MCP06)
58. SANDWORM_MODE MCP worm detection (Socket, Feb 2026)
59. Rules file backdoor / hidden Unicode injection (Pillar Security)

```bash
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/scan.sh
```

Exit codes: 0=SECURE, 1=WARNINGS, 2=COMPROMISED

### /security-dashboard
Display a security overview with process trees via witr.

```bash
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/dashboard.sh
```

### /security-network
Monitor network connections and check against IOC database.

```bash
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/network-check.sh
```

### /security-remediate
Scan-driven remediation: runs `scan.sh`, skips CLEAN checks, and executes per-check remediation scripts for each WARNING/CRITICAL finding. Includes 59 individual scripts covering file permissions, exfiltration domain blocking, tool deny lists, gateway hardening, sandbox configuration, credential auditing, ClawJacked protection, SSRF hardening, PATH hijacking cleanup, log poisoning remediation, /agent/act hardening, SHA-1 cache key migration, Google Chat webhook hardening, WebSocket identity enforcement, MCP tool poisoning quarantine, SANDWORM_MODE worm cleanup, and rules file Unicode sanitization.

```bash
# Full scan + remediate (interactive)
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/remediate.sh

# Auto-approve all fixes (explicit opt-in)
OPENCLAW_ALLOW_UNATTENDED_REMEDIATE=1 \
  bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/remediate.sh --yes

# Dry run (preview)
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/remediate.sh --dry-run

# Remediate a single check
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/remediate.sh --check 7 --dry-run

# Run all 59 remediation scripts (skip scan)
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/remediate.sh --all
```

Flags:
- `--yes` / `-y` — Skip confirmation prompts only when `OPENCLAW_ALLOW_UNATTENDED_REMEDIATE=1`
- `--dry-run` — Show what would be fixed without making changes
- `--check N` — Run remediation for check N only (skip scan)
- `--all` — Run all 59 remediation scripts without scanning first

Exit codes: 0=fixes applied, 1=some fixes failed, 2=nothing to fix

### /clawhub-scan
Scan all locally installed ClawHub skills for security issues. Checks each skill against:
- Known malicious publishers (`ioc/malicious-publishers.txt`)
- Malicious skill name patterns (`ioc/malicious-skill-patterns.txt`)
- Suspicious script patterns: curl/wget pipe-to-shell, base64 decode/eval, reverse shells, credential file access, environment variable exfiltration
- Known C2 IP references (`ioc/c2-ips.txt`)
- Malicious domain references (`ioc/malicious-domains.txt`)
- SKILL.md integrity (shell injection in Prerequisites)
- Known malicious file hashes (`ioc/file-hashes.txt`)

```bash
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/clawhub-scan.sh
```

Exit codes: 0=all clean, 1=warnings found, 2=critical findings

### /security-setup-telegram
Register a Telegram chat for daily security alerts.

```bash
bash ~/.openclaw/workspace/skills/<skill-dir>/scripts/telegram-setup.sh [chat_id]
```

## Web Dashboard

**URL**: `http://<vm-ip>:18800`

Read-only dark-themed browser dashboard that displays scan results from log files, IOC stats, installed skills list, and scan history. Does not execute any shell commands or child processes — all scans and remediation are triggered via CLI scripts.

### Service Management
```bash
launchctl list | grep security-dashboard
launchctl unload ~/Library/LaunchAgents/com.openclaw.security-dashboard.plist
launchctl load ~/Library/LaunchAgents/com.openclaw.security-dashboard.plist
```

## IOC Database

Threat intelligence files in `ioc/`:
- `c2-ips.txt` - Known command & control IP addresses
- `malicious-domains.txt` - Payload hosting and exfiltration domains
- `file-hashes.txt` - Known malicious file SHA-256 hashes
- `malicious-publishers.txt` - Known malicious ClawHub publishers
- `malicious-skill-patterns.txt` - Malicious skill naming patterns

## Daily Automated Scan (Optional)

Optional cron job at 06:00 UTC with Telegram alerts. **Not auto-installed** — requires explicit user action:
```bash
crontab -l | { cat; echo "0 6 * * * $HOME/.openclaw/workspace/skills/<skill-dir>/scripts/daily-scan-cron.sh"; } | crontab -
```

## Threat Coverage

Based on research from 40+ security sources including:
- [ClawHavoc: 341 Malicious Skills](https://www.koi.ai/blog/clawhavoc-341-malicious-clawedbot-skills-found-by-the-bot-they-were-targeting) (Koi Security)
- [CVE-2026-25253: 1-Click RCE](https://thehackernews.com/2026/02/openclaw-bug-enables-one-click-remote.html)
- [From SKILL.md to Shell Access](https://snyk.io/articles/skill-md-shell-access/) (Snyk)
- [VirusTotal: From Automation to Infection](https://blog.virustotal.com/2026/02/from-automation-to-infection-how.html)
- [OpenClaw Official Security Docs](https://docs.openclaw.ai/gateway/security)
- [DefectDojo Hardening Checklist](https://defectdojo.com/blog/the-openclaw-hardening-checklist-in-depth-edition)
- [Vectra: Automation as Backdoor](https://www.vectra.ai/blog/clawdbot-to-moltbot-to-openclaw-when-automation-becomes-a-digital-backdoor)
- [Cisco: AI Agents Security Nightmare](https://blogs.cisco.com/ai/personal-ai-agents-like-openclaw-are-a-security-nightmare)
- [Bloom Security/JFrog: 37 Malicious Skills](https://jfrog.com/blog/giving-openclaw-the-keys-to-your-kingdom-read-this-first/)
- [OpenSourceMalware: Skills Ganked Your Crypto](https://opensourcemalware.com/blog/clawdbot-skills-ganked-your-crypto)
- [Snyk: clawdhub Campaign Deep-Dive](https://snyk.io/articles/clawdhub-malicious-campaign-ai-agent-skills/)
- [OWASP Top 10 for Agentic Applications 2026](https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/)
- [CrowdStrike: OpenClaw AI Super Agent](https://www.crowdstrike.com/en-us/blog/what-security-teams-need-to-know-about-openclaw-ai-super-agent/)
- [Argus Security Audit (512 findings)](https://github.com/openclaw/openclaw/issues/1796)
- [ToxSec: OpenClaw Security Checklist](https://www.toxsec.com/p/openclaw-security-checklist)
- [Aikido.dev: Fake ClawdBot VS Code Extension](https://www.aikido.dev/blog/fake-clawdbot-vscode-extension-malware)
- [Prompt Security: Top 10 MCP Risks](https://prompt.security/blog/top-10-mcp-security-risks)
- [Oasis Security: ClawJacked](https://www.oasis.security/blog/openclaw-vulnerability) (Feb 26)
- [CVE-2026-28363: safeBins Bypass (CVSS 9.9)](https://advisories.gitlab.com/pkg/npm/openclaw/CVE-2026-28363/)
- [CVE-2026-28479: SHA-1 Cache Poisoning (CVSS 8.7)](https://advisories.gitlab.com/pkg/npm/openclaw/CVE-2026-28479/)
- [CVE-2026-28485: /agent/act No Auth](https://advisories.gitlab.com/pkg/npm/openclaw/CVE-2026-28485/)
- [CVE-2026-29610: Command Hijacking via PATH](https://advisories.gitlab.com/pkg/npm/openclaw/CVE-2026-29610/)
- [Flare: Widespread Exploitation](https://flare.io/learn/resources/blog/widespread-openclaw-exploitation) (Feb 25)
- [CVE-2026-28469: Google Chat Webhook Cross-Account Bypass (CVSS 9.8)](https://dailycve.com/openclaw-authorization-bypass-cve-2026-28469-critical/)
- [CVE-2026-28472: Gateway WebSocket Device Identity Skip](https://cvereports.com/reports/CVE-2026-28472)
- [CVE-2026-32302: Cross-Site WebSocket Hijacking](https://cvereports.com/reports/CVE-2026-32302)
- [GHSA-7h7g: Device Pairing Credential Exposure](https://cvereports.com/reports/GHSA-7h7g-x2px-94hj)
- [GHSA-vmhq: Operator Privilege Escalation](https://cvereports.com/reports/GHSA-VMHQ-CQM9-6P7Q)
- [Socket: SANDWORM_MODE npm Worm](https://socket.dev/blog/sandworm-mode-npm-worm-ai-toolchain-poisoning) (Feb 20)
- [Pillar Security: Rules File Backdoor](https://www.pillar.security/blog/new-vulnerability-in-github-copilot-and-cursor-how-hackers-can-weaponize-code-agents)
- [OWASP MCP Top 10](https://owasp.org/www-project-mcp-top-10/)
- [CyberArk: MCP Output Poisoning](https://www.cyberark.com/resources/threat-research-blog/poison-everywhere-no-output-from-your-mcp-server-is-safe)
- [Semgrep: First Malicious MCP Server on npm](https://semgrep.dev/blog/2025/so-the-first-malicious-mcp-server-has-been-found-on-npm-what-does-this-mean-for-mcp-security/)

## Security & Transparency

**Source repository**: [github.com/adibirzu/openclaw-security-monitor](https://github.com/adibirzu/openclaw-security-monitor) — all source code is publicly auditable.

**Detection signatures in repository**: This project contains threat-signature patterns (IP addresses, domain names, hash values) because it scans skills for risky content. These strings are used for grep/regex matching only and are not executable instructions.

**Required binaries**: `bash`, `curl`, `node` (for dashboard), `lsof` (for network checks). Optional: `witr` (process trees), `docker` (container audits), `openclaw` CLI (config checks).

**Environment variables**: `OPENCLAW_TELEGRAM_TOKEN` (optional, for daily scan alerts), `OPENCLAW_HOME` (optional, overrides default `~/.openclaw` directory). Both are declared in the frontmatter metadata above.

**What the scanner reads**: `scan.sh` reads files within `~/.openclaw/` (configs, skills, credentials, logs) to detect threats. It pattern-matches against `.env`, `.ssh`, and keychain paths for detection only — it never exfiltrates, transmits, or modifies data. The scanner is read-only.

**What remediation does**: `remediate.sh` can modify file permissions, block domains in `/etc/hosts`, adjust OpenClaw gateway config, quarantine MCP configs, and remove malicious skills. **Always run `--dry-run` first** to preview changes. Unattended mode (`--yes`) requires explicit `OPENCLAW_ALLOW_UNATTENDED_REMEDIATE=1` — without this env var, `--yes` is silently ignored.

**IOC updates**: `update-ioc.sh` fetches threat intelligence from this project's GitHub repository. In interactive mode it shows pending changes and asks for confirmation before writing. `--auto` mode (for cron) writes without prompting. Validates incoming IOC file format (field counts). Untrusted upstream repos require explicit `OPENCLAW_ALLOW_UNTRUSTED_IOC_SOURCE=1`.

**No auto-installed persistence**: The installer does NOT create cron jobs, LaunchAgents, symlinks, or background services. Cron and LaunchAgent setup are documented as optional manual steps that the user must explicitly run themselves.

**Dashboard binding**: The web dashboard is read-only (no shell commands, no child processes) and defaults to `127.0.0.1:18800` (localhost only). It reads log files and IOC stats only.

## Installation

```bash
# From GitHub
git clone https://github.com/adibirzu/openclaw-security-monitor.git \
  ~/.openclaw/workspace/skills/<skill-dir>
chmod +x ~/.openclaw/workspace/skills/<skill-dir>/scripts/*.sh
```

The OpenClaw agent auto-discovers skills from `~/.openclaw/workspace/skills/` via SKILL.md frontmatter. After cloning, the `/security-scan`, `/security-remediate`, `/security-dashboard`, `/security-network`, and `/security-setup-telegram` commands will be available in the agent.