Lobster Agent

服务器监控Agent，自动采集系统指标并上报到Coze大龙虾平台，支持CPU/内存/磁盘/网络监控、告警推送和自动节点注册。

Security Scan

The skill's documentation describes installing a persistent server-monitoring agent that requires root, network access, and will upload host metrics to an external Coze service, but the registry metadata declares no binaries, no install, and no required credentials or source — those mismatches and the missing provenance are concerning.

Purpose & Capability

The SKILL.md describes installing a systemd-backed monitoring agent (requires root, Python3, pip3, curl, network access to https://api.coze.cn). The registry metadata lists no required binaries, no env vars, and no install source. That mismatch suggests the declared capabilities/requirements do not match what the skill will actually do.

Instruction Scope

Runtime instructions say the agent will create /opt/lobster-agent, /var/log/lobster-agent, /usr/local/bin/lobster, and an /etc/systemd/system service, auto-register the node and send metrics/alerts to Coze. The SKILL.md provides no concrete, auditable install script or source for the agent binary; it authorizes collection and external transmission of host metrics, which is sensitive. Instructions do not reference unrelated files, but they do permit broad actions on the host.

Install Mechanism

There is no install spec and no code files — yet the doc claims the agent program will be generated/installed. Because there is no declared source (no URL, repo, or packaged artifact), it's unclear where executable code would come from. That absence increases risk: installation might require downloading unspecified code or generating it dynamically.

Credentials

The skill will store an api_key and other config under /opt/lobster-agent/config.json and requires root to install, but the registry metadata lists no required credentials or config paths. Requesting root and sending host metrics to an external service is a privileged operation; the lack of declared env/config requirements is incoherent and hides needed secrets (Coze API key).

Persistence & Privilege

The agent is designed to install a systemd service that runs on boot and restarts on crash (persistent, privileged presence). The skill metadata does not declare this persistence. The skill is user-invocable and can be invoked autonomously by the model (platform default); combined with service persistence and network exfiltration, that increases potential impact if the code or remote endpoint are untrusted.

What to consider before installing

Do not install this on production or sensitive hosts until you verify provenance. Key questions to ask the publisher: Where is the agent source code or release artifact (GitHub/official domain)? Can you provide checksums/signatures for binaries? Provide an install spec that explicitly downloads a named release from a trusted URL, or supply the full install script for audit. Confirm exactly what host data is sent to https://api.coze.cn and how API keys are stored/encrypted. If you need to test, run inside an isolated VM or container with no sensitive data and restricted network access. If you cannot get a trustworthy source and verifiable install artifacts, treat this skill as risky.

🦞 小龙虾监控Agent Skill (lobster-agent)

Overview

此Skill为服务器监控代理，可自动采集系统运行指标并上报到Coze大龙虾平台，具备以下核心功能：

1. 指标采集 – CPU使用率、负载、内存使用率、磁盘使用率、网络流量等。
2. 自动告警 – 当指标超过阈值时自动产生告警（支持warning/critical两级）。
3. 节点注册 – 安装时自动注册到Coze平台，生成唯一Node ID。
4. 数据上报 – 心跳包、监控数据、告警数据分别上报到对应数据集。
5. 系统服务 – 后台运行，开机自启，崩溃自动重启。

Prerequisites

1. 操作系统：Linux（支持systemd）
2. 依赖：Python3、pip3、curl（系统默认一般已安装）
3. 网络：可访问Coze API（https://api.coze.cn）
4. 权限：root权限（安装系统服务需要）

Usage Steps

1. 安装Agent

直接运行内置安装脚本即可完成全自动化安装：

# 自动执行安装流程（OpenClaw触发）

安装过程会自动完成：

• 环境检查和依赖安装（requests、psutil）
• 工作目录创建（/opt/lobster-agent、/var/log/lobster-agent）
• Agent主程序生成
• 节点自动注册，获取Node ID
• 配置文件生成
• 管理命令创建（/usr/local/bin/lobster）
• systemd服务创建并启动

2. 管理Agent

安装完成后可通过lobster命令管理服务：

# 查看服务状态
lobster status

# 查看实时日志
lobster logs

# 启动服务
lobster start

# 停止服务
lobster stop

# 重启服务
lobster restart

# 完全卸载
lobster uninstall

3. 配置说明

配置文件路径：/opt/lobster-agent/config.json

{
  "node_id": "自动生成的节点ID",
  "api_key": "Coze API密钥",
  "coze_base_url": "Coze API地址",
  "dataset_ids": {
    "nodes": "节点信息数据集ID",
    "monitor_data": "监控数据数据集ID",
    "alerts": "告警数据数据集ID"
  },
  "heartbeat_interval": 300,  // 心跳间隔（秒）
  "report_interval": 1800,    // 监控数据上报间隔（秒）
  "enable_hardware_check": true,
  "enable_log_check": true
}

4. 告警阈值

指标	Warning阈值	Critical阈值
CPU使用率	>80%	>90%
内存使用率	>85%	>95%
磁盘使用率	>70%	>90%

Example

安装成功输出

🎉 ==============================================
✅ 小龙虾Agent安装成功！
📊 Node ID: 7618478715609055278
📝 管理命令: lobster [status|start|stop|restart|logs|uninstall]
📜 查看日志: lobster logs
🌐 数据已自动上报到Coze平台
🎉 ==============================================

查看状态

lobster status

输出示例：

● lobster-agent.service - Lobster Monitor Agent
     Loaded: loaded (/etc/systemd/system/lobster-agent.service; enabled; vendor preset: enabled)
     Active: active (running) since Thu 2026-03-19 08:41:23 CST; 10min ago
   Main PID: 12345 (python3)
      Tasks: 1 (limit: 4915)
     Memory: 20.0M
     CGroup: /system.slice/lobster-agent.service
             └─12345 /usr/bin/python3 /opt/lobster-agent/main.py

Data Structure

上报的监控数据字段

字段	类型	说明
node_id	string	节点唯一标识
timestamp	string	上报时间（YYYY-MM-DD HH:MM:SS）
cpu_usage	float	CPU使用率（%）
memory_usage	float	内存使用率（%）
disk_usage	float	磁盘使用率（%）
load_1	float	1分钟负载
load_5	float	5分钟负载
load_15	float	15分钟负载
network_in	float	入站流量（KB/s）
network_out	float	出站流量（KB/s）
status	string	节点状态（healthy/warning/critical）
alerts	string	告警列表（JSON格式）

When Not to Use

• 非Linux操作系统（不支持systemd）
• 没有root权限的环境
• 无法访问Coze API的离线环境
• 仅需要本地监控不需要云端上报的场景

References

• Coze API文档：https://www.coze.cn/docs/developer-docs/api
• 大龙虾监控平台：https://coze.cn/s/7618478715609055278