Quote:
Originally Posted by 2MuchMark
N o i c e ... question: Are you just feeding the agent a generated ‘agents.md’ + the ticket, or do you have it dynamically pulling code context each time as well?
|
The AGENTS.md is generated by my context tool that I'm working on, I run it against a git repo and it loads up the git history and source code and just starts plowing through it and looks for as much context it can pull out to describe how the project is designed, key components, patterns, dependencies, etc. This is generated once and reused for all tickets.
Then I just use Copilot inside VSCode to start a new agent and I paste that ticket into the chat, where it automatically identifies the AGENTS.md file as a context file, then my ticket as the user prompt, that sets up the hierarchy to make it all work.
Then after it finishes I review the changes and request stuff in the chat and it will continue making changes until I decide it's good enough, then I close that chat and repeat the process with the next ticket.
Quote:
Originally Posted by Mindi
I read that and I was like wait... what did he just say?
This is eye opening. I could apply that to my process. I could build 50 with the same pattern.
That's a little overkill. I like it.  |
Yeah it's a pretty manual process on my end still but I'm just trying to figure things out and find a flow that I like, and this is what has come of it so far. I figure you can take the idea and apply it to your setup and gain noticeable improvements. Can't wait to hear more about how it changed things for you.
I'll share my tool here on GFY when I've got it ready for public usage.