Background Agent is painfully slow and limited

I configured and tried to run a background agent. It started to behave as expected. However, I never got it to run to completion, and usually not much further than build state.

Most time, what happens is either:

  1. Build fails on some deb package install or network request error (I use a Dockerfile)
  2. The instance starts, but the Agent is simply unable to use any tool. The tools simply timeout 90% of the time. (stuck on “Running…” then, when I ask, it says there was no output)

This is actually weird because if I run the exact same commands I ask in the chat, into the terminal, they work real fine.

I would prefer that my background job is throttled until there is a real capacity available, or not run at all. Here I find myself waiting ages for a build and then the tools timeout… big waste of time.

You might find this feedback a bit negative… don’t! If I am frustrated it’s because I see so much potential in this tool. The execution and UI seems so good.

It just does not work for me, and I will retry until it does, so good luck team, ramp up the infra before too many people get a wrong first impression!

Cheers

Aurelien

2 Likes

Does this have anything to do with your . env?

Environment variables don’t get passed to the Background Agent, which might be causing your failures.

The task was “Run the ls command line and then wait for my prompt”.

It got immediately stuck trying to run ls

Running ls myself in the terminal worked fine.

Here is a prefix that I use, trying to increase the chances it work:

Start by waiting that the instances cools down by running `sleep 60 && echo done`.
If the output is `done`, then continue, otherwise there might be an error, 
so just stop the work immediately. Then start the work: 

Not sure if it helps fwiw

Edit: it did not help. in the last run it passed the sleep and then got stuck on the next command…

yes i captured same here as well

1 Like