返回目录

安全与治理插件

hermes-security-audit

axelfreeman/hermes-security-audit

🔒 12-method security audit for Hermes Agent — virus scan, rootkit detection, SSH brute force protection

查看仓库项目主页

Stars: 1
Forks: 0
Issues: 1
更新: 3 天前

PROJECT TOPICS

项目标签

ai-agents audit deepseek-harness dsh-plugin linux security

INSTALL REFERENCE

安装参考

未验证

dsh plugin --profile web add github:axelfreeman/hermes-security-audit

在 DSH 中安装

该命令指向仓库当前默认分支；尚无绑定当前 commit 的完整验证结果。

PROJECT README

README

🔒 Hermes Agent Security Audit — AI Agent Antivirus & Vulnerability Scanner

⭐ Star this repo to help more people audit their AI agents for malware.

12 detection methods. 5 minutes to run. Open source.
Check your Hermes agent for malware, rootkits, crypto miners, and SSH brute force attacks.

What is this?

A security audit skill for Hermes Agent — an open-source AI assistant by Nous Research. If you run Hermes Agent on your own Linux server, this skill scans it for viruses, rootkits, backdoors, and other threats.

Think of it as antivirus for AI agents. Just like you scan your laptop, you should scan the server where your AI agent lives.

Why you need an AI agent security audit

Hermes Agent runs on Linux servers with root access, Docker containers, API keys, and open network ports. That's a lot of attack surface. This Hermes agent vulnerability scanner checks:

Threat	Detection Method
🦠 Viruses & malware	ClamAV — industry-standard antivirus
👻 Rootkits & backdoors	rkhunter + chkrootkit — dual rootkit detection
⛏️ Crypto miners	Process audit — signature + behavioral analysis
🔑 Exposed credentials	File scan — API keys, tokens, private keys in plain sight
🚪 Open doors	Port audit — unnecessary services (CUPS, SMTP, printing)
🤖 SSH brute force	Log analysis — 11,000+ failed attempts found and blocked
🐳 Docker escapes	Container audit — privileged mode, exposed ports
📅 Hidden cron jobs	Cron audit — all users, all schedules
📁 SUID/SGID exploits	Privilege escalation vector detection
🌐 Outbound connections	Network connection audit — who's calling home

Quick Start

# 1. Install Hermes Agent (if not already)
# https://github.com/nous-research/hermes-agent

# 2. Load the security skill
hermes skill load hermes-security-audit

# 3. Run the audit
# Say to your Hermes agent: "run a security audit on my server"

Or just copy the SKILL.md into ~/.hermes/profiles/your_profile/skills/.

12 Security Checks — Full Breakdown

1. Open Ports (TCP/UDP)

Detects hidden services listening on unexpected ports. Finding: CUPS print service on port 631, SMTP on port 25, Zabbix agent on 10050 — all unnecessary, all removed.

2. Process Audit

Scans for crypto miners (xmrig, stratum), suspicious process names mimicking legitimate ones, and orphaned services. Finding: orphaned Xray SOCKS5 proxy from previous testing — killed.

3. SSH Brute Force Detection

Counts failed authentication attempts and identifies attacker IPs. Finding: 11,099 failed SSH attempts across three servers. Top attacker IPs blocked, fail2ban installed.

4. SUID/SGID Files

Finds binaries with elevated privileges — potential privilege escalation vectors. Finding: 0 dangerous SUID files found.

5. Cron Job Audit

Lists crontabs for all system users. Finding: legitimate outreach scripts and certbot renewals — no hidden miners.

6. Docker Container Audit

Checks all containers for privileged mode, host network access, and unexpected images. Finding: only known containers running (amnezia-awg2, open-seo).

7. chkrootkit

Signature-based rootkit scanner. Finding: false positives for basename, date, dirname — expected on modern kernels.

8. rkhunter

Behavioral rootkit and backdoor detection. Finding: standard warnings for snap users, hidden config files — all benign.

9. ClamAV

Full antivirus scan of /tmp, /opt, and /root. Finding: 0 infected files.

10. Lynis Security Score

Comprehensive hardening audit with numerical score. Finding: initial score 65/100 — improved by disabling unnecessary services.

11. Exposed Credentials

Scans home directories for API keys, tokens, private keys stored in plain text. Finding: Amnezia VPN private keys moved to protected storage.

12. Network Connections

Checks active outbound connections for suspicious destinations. Finding: only legitimate Hermes gateway and Chrome browser connections.

Real Results from Production Server

Server: 195.133.93.243 (Hetzner, Ubuntu 26.04)
Profile: Hermes Agent marketing profile + 2 additional gateways

Before audit:
  - PermitRootLogin: yes (password auth allowed)
  - Open ports: 22, 25, 443, 631, 10050, 18960, 20808
  - SSH brute force: 3,194 failed attempts
  - VPN private keys: exposed in /root/
  - Lynis hardening index: 65/100

After audit + fixes:
  - PermitRootLogin: prohibit-password (key-only auth)
  - Open ports: 22, 443, 18960 (memmy-agent)
  - SSH brute force: fail2ban active, 8 attacker IPs blocked
  - VPN keys: moved to ~/.hermes/profiles/ (600 permissions)
  - Lynis hardening index: 72/100 (+7 points)
  - CUPS, SMTP, Xray SOCKS5: all removed

When to Run This Audit

✅ After installing Hermes Agent on a new server
✅ Weekly as a cron job (0 0 * * 0)
✅ After installing new tools or Docker containers
✅ When you notice unusual CPU/memory usage
✅ Before exposing your agent to external platforms (Telegram, Discord)

Who Is This For?

Hermes Agent users running self-hosted instances
AI agent developers who want to secure their infrastructure
Linux server admins — works on any Linux server, not just Hermes
Security-conscious teams building on open-source AI tools

Need this done for you?

The free half of this work is in this repo. If you want it run against your own market — segment picked, contacts sourced and verified over SMTP before the first send, touches written, replies worked until meetings are booked:

License

MIT — use, modify, share freely. Part of Axel Freeman's open-source marketing and security toolkit.

Last updated: August 2026.

CLASSIFICATION EVIDENCE

分类依据

项目类型插件

功能分类安全与治理

规则置信度高

系统优先读取 GitHub Topics，再与站内分类词典和词根规则比对。当前命中： audit、security。