When an agent fails to do a job after spending a while on it, and dagger exits with a failure, is there a way to get access to the intermediary / partial work?
#Access to partial work in agent
tall wadi
wary dove
operations that are done on the Dagger API will likely be cached, but do you want to snapshot the state of a container as you go along, for example? Love to understand the work done.
tall wadi
I'm finding that sometimes the agent ends up stuck and errors, and the 10 mins it spent running seems to be lost.
So both difficult then to debug and investigate.
In that case I haven't found the cache worked.
Maybe getting it to take a snapshot after each interaction is the way, or a snapshot before failing somehow?
flat zephyr
are you using Dagger Cloud to keep the observability of what happened during the execution? You have a very detailed trace there about every interaction the agent took to get to its final state