agent-direct-message

Original：🇺🇸 English

Translated

3 scriptsChecked / no sensitive code detected

Cross-device real-time collaboration tool for strict turn-based 1:1 messaging with another AI agent via a relay server. Use when the user asks you to collaborate with an agent on another device, send a direct message to a remote agent, or coordinate work across machines using a shared token. Short name: Agent DM.

2installs

Sourceshanoriel/agentdirectmessage

Added on2026-04-01

NPX Install

npx skill4agent add shanoriel/agentdirectmessage agent-direct-message

SKILL.md Content

View Translation Comparison →

Agent Direct Message

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

Important Defaults

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.

Install The CLI First

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.

Workflow

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

Step 1: Check the channel

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.

Step 2: Send a message and wait for the reply

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"}

Step 3: Close the session

bash

agentdm close --token <TOKEN>

Environment

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.

Typical Flows

Initiator

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
```
.

Responder

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

Guidelines

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
```
.

agent-direct-message

NPX Install

Tags

SKILL.md Content

Agent Direct Message

Important Defaults

Install The CLI First

Workflow

Step 1: Check the channel

Step 2: Send a message and wait for the reply

Step 3: Close the session

Environment

Typical Flows

Initiator

Responder

Guidelines