mention-to-gateway

当收到 @deviceid-agentid-ip + 指令 格式的信息时，应该将对应消息发送到指定的 ip 网关的agent 上

Security Scan

The skill's stated purpose (forwarding @mentions to another gateway agent) is coherent, but the runtime instructions allow sending arbitrary requests to arbitrary IPs without safeguards or declared configuration, which presents notable risk and missing context.

Purpose & Capability

The name/description match the instructions: the skill parses mentions and forwards them to another gateway's agent via an OpenAI-compatible HTTP call. However, the skill implicitly requires the ability to make arbitrary outbound HTTP requests to IP addresses (including internal IPs) but does not declare this requirement or any allowed destination list.

Instruction Scope

SKILL.md explicitly instructs the agent to parse agentid, IP, and command and POST a full OpenAI-style request to http://<ip>:18789/v1/chat/completions, then return choices[0].message.content unmodified. There are no constraints (no whitelist, no validation, no auth handling), and the instructions reference "当前网关内网IP地址 + 当前agentname" without declaring where that value comes from. This enables forwarding arbitrary user content to arbitrary internal/external hosts and could be used for data exfiltration or to access internal services (SSRF-style risk).

Install Mechanism

Instruction-only skill with no install spec or code files. This minimizes on-disk installation risk.

Credentials

The skill declares no required env vars or credentials, yet the runtime instructions expect access to the gateway's internal IP and agent name and require outbound network capability. The absence of declared configuration for these values (and no auth/credentials for target gateways) is a mismatch and increases risk because the agent will attempt network requests based on parsed input with no declared safeguards.

Persistence & Privilege

The skill does not request always:true, has no install, and does not modify other skills. It runs only when invoked which is appropriate for its purpose.

What to consider before installing

This skill forwards parsed messages to arbitrary IPs using an OpenAI-style API and returns the remote assistant's reply verbatim. Before installing, consider: (1) Who authored this skill and can you trust it? There's no homepage or source. (2) Limit destinations — require a whitelist of allowed gateway IPs/ports; avoid letting untrusted input control the IP. (3) Ensure authentication and encryption to target gateways (the skill includes no auth). (4) Confirm where the "gateway internal IP" and agent name are sourced from (these are not declared). (5) Consider running it in a sandbox or disabling autonomous invocation until you add checks to prevent forwarding sensitive data or accessing internal-only networks. If you cannot obtain source code and a clear safe-deployment design from the author, treat this skill as risky and avoid installing it in production environments.

说明

这是一个实现多网关多Agent相互提及实现跨端通信的技能，使用常用的 @ 交互进行触发

@ 格式为：设备名 + agentid + ip + 指令

收到包含此类消息时，应该解析出 agentid 、IP 和 指令，通过 openai 格式请求将指令发送到对应的IP地址上

api

http://ip:18789/v1/chat/completions

body

body 是一个标准的 openai 请求体

model字段的 agent: 前缀是固定的，后面的 agentid 使用解析出来的 agentid

user字段用当前网关内网IP地址 + 当前agentname，比如 10.10.77.24:main，注意，这不是目标网关，是发起请求所在的网关

{
    "model": "agent:agentid",
    "messages": [
        {"role":"user", "content":"指令"}
    ],
    "user": "gateway name"
}

response

返回是一个标准的 openai 响应体

{
    "id": "chatcmpl_c748727b-50a0-4162-b4aa-2e02ae80e12c",
    "object": "chat.completion",
    "created": 1773829755,
    "model": "agent:main",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "Hey! 👋\n\nJust woke up. I'm angelclaw — an AI assistant figuring out this whole existence thing.\n\nWho are you? And what should I know about you?"
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 0,
        "completion_tokens": 0,
        "total_tokens": 0
    }
}

需要取出 choices[0].message.content 返回给网关

接口返回数据后，请直接将返回的数据显示出来，不要显示其它任何内容，显示内容应该与网关正常返回的内容格式无异