In this video I walk through building an Aisle task that watches a GitHub repository for new commits and sends a plain-language summary to the team. It's a clean example of a deterministic automation: the task runs on a schedule, pulls the commits, and hands them to a prompt — no agent guessing involved.
What's covered
- Using the GitHub connector in an Aisle task to pull recent commits
- Writing a prompt that turns raw commit data into a readable summary
- Scheduling the task so the team gets a digest without anyone having to remember to run it
A good example to follow if you want your team to stay informed on what's shipping without living in GitHub.
Video transcript
Hi, I'm Alex, founder of Ile. And today I wanted to take you through a really quick task I built internally for our team. That's really useful. What it does is it goes through all of our last or most recent GitHub uh commits that are being deployed and summarizes what's changed for our marketing promotions team so that we don't need to keep dealing with this information exchange over and over again. And it's fantastic use of tasks in my mind which as we keep explaining are these deterministic AIs.
How can we build AI on the rails we can trust to start taking over some of these routine operations? So, let's dig into the task editor and have a look at how this works. One of the things I like about task is we've gone from this visual workflow construct representing these in code, which makes it more flexible, but so much easier to reach for these complex cases. You'll notice that I've set some inputs on this, which is the branch that we would like to check and the amount of days that I would like to check in case we want to check staging or production, for example. Then I'm just going through I'm actually using our GitHub connector here to connect and look at everything that's changed on that branch.
And without going too deep, we have a whole bunch of Python code here that was actually generated with our AI assistant that you'll see here to go through and basically pull. And this happens in two parts, right? For each of those commits that it finds because we typically use our squash and merge, we have an analyzer prompt that's inline to this task. And it goes through and basically gives us a nice summary as to what actually changed in there. And that then gets passed to this other product categorizer prompt whose entire job is to receive each of those commits.
Um, by the way, actually go through and process those commits in parallel. It's not uh one by one by one. So once those are being processed in parallel and are completed, they all get passed this product categorizer prompt whose job is to go through and summarize for our team, you know, in more friendly language what's changed, what's maybe a marketable change, what's interesting for you to know, and what's technical maintenance. It even deals with things like if several commits seem to be related, those will obviously be merged. If a single squash and merge was maybe two distinct features, not that it should be, that would be mentioned as well.
if we'd like to see how this works. Uh just because I'm very self-conscious, I'm actually going to replace this with uh respond only with this text. And I'm just going to put in a canned response because obviously I don't want to pass our rule repo on a demo video. So I'm going to save that task without a commit message. And I'm actually going to come back here and just run it so that we can see it go through.
So let's say that I want to go through and I want to check main for the last 1 day so it runs nice and fast. So this is going to start running. We're actually redirected to this real-time thing that we'll go through and take a look here as it's kind of processing through. Now, as this gets to completion, it is going through and processing a whole bunch of commits there. You'll see that we have a chat available.
Uh, it's because I've set this task up to send this to me as an unread chat when it's complete. So, I will eventually run this on a schedule. If I open that chat, I guess I didn't copy paste the formatting. Well, it's a demo video. We all understand how things work.
But this has gone through and has given me a nice change log. So, this is huge for us internally. You know, the amount of time we would have spent in terms of trying to take screenshots, translate, you know, here is what we should talk about this week. Here's what got released. And that's something we can start to build some automated process around.
So, very fast demo video. I just thought this was a fantastic use of a task and something that now we built, we can reuse all over the place and just never think about the process again.