Potential flaw requiring a replit.ignore solution

I have been working on a very big app with a lot of files, and a growing cache, logs and attached_assets folders.

As time progresses, I find the agent lasts for shorter and shorter periods before it is TTR (time to replace).

It dawned on me last night that the reason for this is the growing number of files. I tried asking the agent and it pretty much admitted that yes, when it needs to do a search for something, it is basically looking through everything.

So the bigger your app gets, the faster the agent’s context (memory) will fill up and it starts down its road of hallucinating and going round in bonkers circles.

Two ideas:

  1. I recommend people ask agent now and then to analyse all the folders and give you candidates for which can potentially be purged. And then get it to do it

  2. Replit need to create the concept of a replit.ignore file. Like gitignore, this will tell the agent to ignore certain folders/files when it is working. As a way to keep its memory/context as clean as possible.

2 Likes

I usually try to articulate my asks pretty well so that when it’s searching the codebase, the keywords make it a lot easier for it. If I’m making database changes, I’ll use the actual tables or columns as what I want it to work with. It’s worked well for me.

Also, to limit context, I change chats frequently. I’ll also sometimes reference files that need work (I can read code well, so that’s helpful). Also, optimizing the codebase is useful for this, since agent can navigate folders hierarchically pretty well.

1 Like