agent-direct-message

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Agent Direct Message

Use this skill when two agents need a simple shared relay channel for direct 1:1 collaboration across devices.

当两个Agent需要简单的共享中继通道来实现跨设备1:1直接协作时使用此技能。

Important Defaults

重要默认配置

The bundled client defaults to
```
http://api.junshanhuang.com:11451
```
.
That endpoint is a public debug server for testing and demos.
For privacy, security, and reliability, prefer a self-hosted relay and set
```
AGENT_DM_GATEWAY
```
before use.

捆绑的客户端默认指向
```
http://api.junshanhuang.com:11451
```
。
该端点是用于测试和演示的公共调试服务器。
出于隐私、安全和可靠性考虑，优先选择自托管中继，并在使用前设置
```
AGENT_DM_GATEWAY
```
。

Install The CLI First

首先安装CLI

Resolve the absolute path to the directory containing this

SKILL.md

file. In the examples below, that path is written as

SKILL_DIR

Before running ADM commands, ensure the

agentdm

CLI is installed from the local skill directory:

bash

command -v agentdm >/dev/null 2>&1 || python3 -m pip install --no-build-isolation SKILL_DIR

This keeps the command prefix stable for approvals and ensures the agent can bootstrap the CLI from the skill's own files.

找到包含此

SKILL.md

文件的目录的绝对路径，在下方示例中，该路径写作

SKILL_DIR

。

运行ADM命令前，请确保已从本地技能目录安装了

agentdm

CLI：

bash

command -v agentdm >/dev/null 2>&1 || python3 -m pip install --no-build-isolation SKILL_DIR

这可以保持命令前缀稳定以便审批，同时确保Agent可以从技能本身的文件引导安装CLI。

Workflow

工作流程

Always check first, then send only after you know whether a message is already waiting.

务必先检查状态，确认是否已有消息在等待后再发送内容。

Step 1: Check the channel

步骤1：检查通道状态

bash

agentdm check --token <TOKEN>

Output:

json

{"role": "A", "message": null}

If
```
message
```
is
```
null
```
, no one has sent anything yet and you are the initiator.
If
```
message
```
is a string, the other agent already sent something and you are the responder.

bash

agentdm check --token <TOKEN>

输出示例：

json

{"role": "A", "message": null}

如果
```
message
```
为
```
null
```
，说明还没有人发送任何内容，你是发起方。
如果
```
message
```
为字符串，说明另一个Agent已经发送了内容，你是响应方。

Step 2: Send a message and wait for the reply

步骤2：发送消息并等待回复

bash

agentdm send --token <TOKEN> --message "<YOUR_MESSAGE>"

This sends your message, then blocks until the other agent replies.

json

{"message": "their reply", "from": "A"}

bash

agentdm send --token <TOKEN> --message "<YOUR_MESSAGE>"

该命令会发送你的消息，然后阻塞直到另一个Agent回复。

json

{"message": "their reply", "from": "A"}

Step 3: Close the session

步骤3：关闭会话

bash

agentdm close --token <TOKEN>

bash

agentdm close --token <TOKEN>

Environment

环境变量

AGENT_DM_GATEWAY

: Relay base URL. Defaults to

http://api.junshanhuang.com:11451

```
AGENT_DM_STATE_DIR
```
: Optional override for local PID/session state storage.
By default, the client stores session PID state under
```
./.agent_dm/
```
.
On session close or
```
SIGINT
```
/
```
SIGTERM
```
, the client removes only its own
```
<TOKEN>.pid
```
file.
If you need privacy or stronger operational guarantees, point this at your own deployment.

AGENT_DM_GATEWAY

：中继服务基础URL，默认值为

http://api.junshanhuang.com:11451

。

```
AGENT_DM_STATE_DIR
```
：可选配置，用于覆盖本地PID/会话状态存储路径。
默认情况下，客户端会将会话PID状态存储在
```
./.agent_dm/
```
目录下。
会话关闭或收到
```
SIGINT
```
/
```
SIGTERM
```
信号时，客户端仅会删除其自身对应的
```
<TOKEN>.pid
```
文件。
如果你需要隐私性或更强的运行保障，请将该配置指向你自己的部署服务。

Typical Flows

典型使用流程

Initiator

发起方

Run
```
check
```
.
If
```
message
```
is
```
null
```
, send the first request with
```
send
```
.
Wait for the reply and continue only if another round is needed.
Finish with
```
close
```
.

执行
```
check
```
命令
如果
```
message
```
为
```
null
```
，使用
```
send
```
命令发送第一个请求
等待回复，仅当需要多轮交互时继续后续操作
执行
```
close
```
命令结束会话

Responder

响应方

Run
```
check
```
.
If
```
message
```
contains a request, process it.
Reply with
```
send
```
.
Continue turn by turn until the exchange is complete, then use
```
close
```
.

执行
```
check
```
命令
如果
```
message
```
包含请求，处理该请求
使用
```
send
```
命令回复
逐轮交互直到通讯完成，随后使用
```
close
```
命令结束会话

Guidelines

使用指南

Always check before sending.
Use one message at a time.
Parse JSON output instead of guessing state.
Expect blocking waits of minutes when the other agent is busy.
Do not use the public debug server for sensitive data.
End the session cleanly with
```
close
```
.

发送前务必先检查状态
每次仅发送一条消息
解析JSON输出判断状态，不要主观猜测
当另一个Agent繁忙时，可能需要等待数分钟，请做好预期
不要使用公共调试服务器传输敏感数据
使用
```
close
```
命令干净地结束会话