Forem: Lunran

Turning GitHub Copilot CLI into an AI Agent via ACP

Lunran — Sun, 22 Mar 2026 11:11:48 +0000

I wanted an AI agent that could check the news, emails, and my calendar. While OpenClaw is currently popular, dealing with its security and permission requirements can be a hassle.

By operating the familiar GitHub Copilot CLI through scheduled execution scripts or Discord, I was able to achieve what I wanted. Receiving execution results periodically is convenient, but being able to ask follow-up questions or give instructions for deeper dives directly is even better.

GitHub Repository: Lunran/acp-client

Step 1: Launch GitHub Copilot CLI as an ACP Server

First, start the GitHub Copilot CLI as an ACP server.

copilot --acp --port 8100

Running it within a sandbox can improve safety.

Reference: https://zenn.dev/lunran/scraps/5105de92cb9687

docker ps -a --format '{{.Names}}' | grep -q "^copilot-acp-container\$" || \
docker start -ai copilot-acp-container || \
docker run -it \
  --name copilot-acp-container \
  -p 8100:8100 \
  -v $(pwd):/workspace \
  -v ./.copilot:/home/agent/.copilot \
  -e GITHUB_TOKEN=$GITHUB_TOKEN \
  copilot-sandbox \
  copilot --acp --port 8100 --autopilot --yolo --model gpt-5-mini

The ACP server seems to expect coding tools as clients, but by following the protocol, you can connect it to any tool.

Copilot CLI ACP server - GitHub Docs

Step 2: Configure Discord and Start the ACP Client

Next, configure the Discord settings and start the ACP client.

git clone https://github.com/Lunran/acp-client.git
cd acp-client
cp .env.example .env

Enter your Discord settings in the .env file.

uv sync
uv run python main.py

Note: Although a model is specified when starting the ACP server, for some reason the default model (Claude Sonnet 4.6) often ends up being used. Therefore, the client includes a process to explicitly set the model during startup.

How to Set Up a Sandbox Environment for GitHub Copilot CLI on Linux

Lunran — Sun, 22 Mar 2026 09:54:40 +0000

When running AI coding agents like GitHub Copilot CLI, isolation is key for security and keeping your host environment clean. While Docker Desktop provides a native Sandbox feature for Mac and Windows, Linux users often face different challenges.

In this post, I’ll share how to set up a robust sandbox environment for GitHub Copilot CLI on Linux, based on my recent experiments.

The Challenge

Official Docker Sandboxes are currently not supported on Linux.

To solve this, I used the following template:
henrybravo/docker-sandbox-run-copilot

Troubleshooting the Setup on Linux

When using the repository above on a Linux machine, you might encounter a few hurdles:

1. Quickstart Limitations

The "Option 1: Using Docker Sandbox" recommended in many guides is not applicable because the docker sandbox command is missing on Linux.

2. CRLF Line Ending Issues

If you try "Option 2: Using Docker Run (Standalone)", you might encounter errors because the entrypoint.sh file in some repositories is saved with Windows-style line endings (CRLF). Linux environments require LF.

3. The Solution: Local Build

To get everything working correctly:

Clone the repository.
Fix the line endings of entrypoint.sh (e.g., using dos2unix or your editor).
Follow Option 3: Build Locally.

Persisting Session Data

By default, a sandbox environment is ephemeral. This means you lose your session information, such as planning results (plan.md) or research logs, once the container stops.

To fix this, you need to mount the state directory from your host to the container.

How to mount the session state:

Add a volume mount to your docker run command for the following path:
/home/agent/.local/state/.copilot

An example of complete commands:

docker ps -a --format '{{.Names}}' | grep -q "^copilot-container$" && \
docker start -ai copilot-container || \
docker run -it \
    --name copilot-container \
    -v $(pwd):/workspace \
    -v ./.copilot:/home/agent/.copilot \
    -e GITHUB_TOKEN=$GITHUB_TOKEN \
    copilot-sandbox \
    copilot --autopilot --yolo --model gpt-5.4

Why not use the official Docker Sandbox?
Even on Mac, the current implementation of Docker Sandboxes does not support volume mounts or port forwarding. By using a standard Docker container approach on Linux (or Mac), you actually gain more flexibility in managing your data.

Conclusion

Setting up a sandbox for GitHub Copilot CLI on Linux requires a bit of manual tweaking—specifically handling line endings and choosing a local build over the automated sandbox commands. However, the ability to mount volumes and to use port forwarding currently makes this "manual" approach superior in some use cases.

References

Originally posted on Zenn.

Fixed Multilingual CLIP demo

Lunran — Sun, 23 Oct 2022 16:03:10 +0000

Fixed demo （Google Colaboratory）
https://drive.google.com/file/d/1q04sOt2rrrhlFzXOs0-WDYk-MRMj1qSJ/view?usp=sharing
Original repo
https://github.com/FreddeFrallan/Multilingual-CLIP
Changes
- Changed text, vision encoder
- Used pytorch model
- Supported GPU (Converted tokenizer output)

Translator Voicebot in VRChat 2

Lunran — Sat, 18 Jun 2022 03:00:02 +0000

What's this?

Voicebot that implemented speech-to-speech translation in VRChat
- https://youtu.be/DVAsHKrHV0Q

System

ASR, Translation: googletrans
- https://pypi.org/project/googletrans/
TTS: gtts
- https://pypi.org/project/gTTS/
Sound Pipeline
- application -> speaker -> dialog.py -> mic -> application
- Direct loop also exists.. (application -> speaker -> mic -> application)
(TBD) Release the scripts in GitHub
- Need to clean the codes and write documents..

Prerequisites

Ubuntu 18.04.5 LTS
Voice chat application (VRChat for example)

Setup

$ sudo apt install portaudio19-dev pavucontrol python-xlib scrot $ git clone https://github.com/Lunran/sts.git $ cd sts $ python3 -m venv venv $ . venv/bin/activate $ pip install --upgrade pip $ pip install -r requirements.txt

Usage

Run VRChat
- Open "Settings", "Audio Volume"
- Mute "UI/Menu", "World" and "Avatars"
Run the script
- refer to "$ python dialog.py -h" for details

Conversational AI in VRChat 2

Lunran — Sun, 08 May 2022 02:10:09 +0000

What's this?

Introduction of a voicebot that implemented conversational AI in VRChat
- https://youtu.be/j9L51pASeiQ

System

ASR: googletrans
- https://pypi.org/project/googletrans/
Text Generation: Rinna (1.3B-parameter Japanese GPT model)
- https://huggingface.co/rinna/japanese-gpt-1b
TTS: tyc-talksoft
- https://shirowanisan.com/tyc-talksoft
Sound Pipeline
- application -> speaker -> dialog.py -> mic -> application
- Direct loop also exists.. (application -> speaker -> mic -> application)
(TBD) Release the scripts in GitHub
- Need to clean the codes and write documents..

Prerequisites

Ubuntu 18.04.5 LTS
Voice chat application (VRChat for example)

Setup

Usage

Run VRChat
- Open "Settings", "Audio Volume"
- Mute "UI/Menu", "World" and "Avatars"
Run the script
- refer to "$ python dialog.py -h" for details

Conversational AI in VRChat 1

Lunran — Sun, 17 Apr 2022 09:44:34 +0000

What's this?

Introduction of a voicebot that implemented conversational AI in VRChat
- https://youtu.be/8u22YDqhY6Y

System

ASR: googletrans
- https://pypi.org/project/googletrans/
Text Generation: ILYS-aoba-chatbot (seq2seq model pretrained with Japanese dialog data)
- https://github.com/cl-tohoku/ILYS-aoba-chatbot
TTS: tyc-talksoft
- https://shirowanisan.com/tyc-talksoft
Sound Pipeline
- application -> speaker -> dialog.py -> mic -> application
- Direct loop also exists.. (application -> speaker -> mic -> application)
(TBD) Release the scripts in GitHub
- Need to clean the codes and write documents..

Prerequisites

Ubuntu 18.04.5 LTS
Voice chat application (VRChat for example)

Setup

Usage

Run VRChat
- Open "Settings", "Audio Volume"
- Mute "UI/Menu", "World" and "Avatars"
Run the script
- refer to "$ python dialog.py -h" for details

(Failed attemped) How to get Maixduino connected to AWS IoT

Lunran — Tue, 22 Dec 2020 10:26:41 +0000

(Failed attemped) How to get Maixduino connected to AWS IoT

Steps

AWS IoT Preparation

Create Resource
- https://docs.aws.amazon.com/iot/latest/developerguide/create-iot-resources.html
Display MQTT message
- https://docs.aws.amazon.com/iot/latest/developerguide/view-mqtt-messages.html

Preparing Maixduino

Smoke test
- Carry out the setup procedure
- https://maixpy.sipeed.com/en/get_started/power_on.html
Upgrade firmware
- ussl support was merged into master in 2020/1
  - https://github.com/sipeed/MaixPy/issues/105
- Build master and flash
  - https://github.com/sipeed/MaixPy/blob/develop/build.md
  - In the middle of the procedure, enable the ussl module with the following command
    - $ python3 project.py menuconfig
Place a script to enable WiFi
- https://github.com/sipeed/MaixPy_scripts/blob/master/network/network_esp32.py
Send MQTT message over SSL
- ref. https://qiita.com/sat0ken/items/ac18b0d14e58e0eae124
- (Caution) The signature of ussl.wrap_socket is different from the one of micropython official
  - Probably not compatible with root certificates (where is the source code?)
  - https://github.com/micropython/micropython/issues/2781
  - https://docs.micropython.org/en/latest/library/ussl.html
- -> Run-time error
  - Is it possible without root certificate..?
    - https://docs.aws.amazon.com/iot/latest/developerguide/protocols.html

import socket
import ussl as ssl
import json
from network_esp32 import wifi

SSID = "(set yours)"
PASW = "(set yours)"

KEY_PATH = '(set yours)'
CERT_PATH = '(set yours)'
END_POINT = '(set yours)'

def enable_esp32():

    if wifi.isconnected() == False:
        for i in range(5):
            try:
                # Running within 3 seconds of power-up can cause an SD load error
                # wifi.reset(is_hard=False)
                wifi.reset(is_hard=True)
                print('try AT connect wifi...')
                wifi.connect(SSID, PASW)
                if wifi.isconnected():
                    break
            except Exception as e:
                print(e)
    print('network state:', wifi.isconnected(), wifi.ifconfig())

def send():

    s = socket.socket()

    ai = socket.getaddrinfo(END_POINT, 8883)
    print("Address infos:", ai)
    addr = ai[0][-1]

    with open(KEY_PATH, 'rb') as f1:
        key = f1.read()
    with open(CERT_PATH, 'rb') as f2:
        cert = f2.read()
    s = ssl.wrap_socket(s, key=key, cert=cert)

    s.connect(addr)
    print('Connected to Server')
    temp = 100
    json_data = {'state':{'reported':{'Temperature': temp}}}
    send_data=json.dumps(json_data)

    s.write(b"POST /things/ESP32/shadow HTTP/1.1\r\n")
    s.write(b"Host: " + END_POINT + "\r\n")
    s.write(b"Content-Length: %d\r\n" % len(send_data))
    s.write(b"Connection: close\r\n")
    s.write(b"\r\n")
    s.write(send_data)
    print('Send Data to Server')

    recv_data = s.recv(4096)
    print(str(recv_data, 'utf8'))
    s.close()

if __name__ == '__main__':
    enable_esp32()
    send()