GitHub Auto Reply Agent for issues and discussions

[ekzhu]

An long running agent or multi agent workflow that monitors new discussions and issues on a GitHub repo, and automatically generate a response. It supports posting the response once user edits and approves it.

Interface can be a Teams bot, or something more general.

Try this on the AutoGen repo.

Components:

• Basic connection to GitHub, authorization
• UI: Teams Bot
• Memory: search index over existing replies
• Learning: learn user edits, preferences, and triaging routines over time, filter learning sources by users

Discussions and issues moved to https://github.com/microsoft/gh-gitgen

[gagb]

What if we first focused on an AutoGen powered agentic CLI that complements the gh CLI?
https://cli.github.com

bot issue 1234 draft-reply
bot pr 43 draft-reply
bot issue 999 find-duplicates

The bot instance (including its memory) will be tied to the repo.

[ekzhu]

https://cli.github.com/

This is lovely.

It supports creating your own extension, which is just an executable script.

In its basic form, the agent can be a script that runs through the open github issues that the users have not responded to and suggest a response.

[gagb]

Extension sounds right. Here's a template of gh autogen extension that calls the AutoGen hello world python example. We can replace it with a GH agent with access to repo name and GH API key and equip it with simple tools that access specific issues and PRs. And can generate well thought responses. We could use multi-agent reflection to ensure drafted responses fit maintainer's expectations. We can add tool calls that update maintainers expectations.

Structure

• Bash Entry Point: gh-autogen – Calls the Python script.
• Python Script: autogen.py – Contains the specified Python code.
• Metadata: gh-autogen.yml – Describes the extension.

Code

• gh-autogen:

  #!/bin/bash
  python3 "$(dirname "$0")/autogen.py"

• autogen.py:

  import asyncio
  from autogen_agentchat.agents import AssistantAgent
  from autogen_ext.models.openai import OpenAIChatCompletionClient
  
  async def main() -> None:
      agent = AssistantAgent("assistant", OpenAIChatCompletionClient(model="gpt-4o"))
      print(await agent.run(task="Say 'Hello World from gpt 4o!'"))
  
  asyncio.run(main())

• gh-autogen.yml:

  name: "gh-autogen"
  description: "A Python-powered GH CLI extension to run AutoGen tasks."

Usage

1. Install:

   gh extension install .

2. Run:

   gh autogen

   Output:

   Hello World from GPT-4o!
   

[gagb]

Ive made significant progress. Update soon.

[gagb]

I've started the development of a light weight sample code here: https://github.com/microsoft/gh-gitgen

[ekzhu]

Alright lets use https://github.com/microsoft/gh-gitgen for further discussions and issues.