Because Trailblaze runs inside an external process, we don't get signals on whether an app has crashed. With info from #54 regarding the active process ID, we could have the LLM "figure it out" like a human would based on what the foreground process is.
Ideally we could send chunks of the logcat messages to the LLM to give it more context. This is a general approach question though. Because we run inside a separate process, on-device. The instrumentation process can't access the logs of another application while running on-device.
It could be possible with "host mode" to view logs at some point, but that's currently out of scope as we are investing in on-device.
Because Trailblaze runs inside an external process, we don't get signals on whether an app has crashed. With info from #54 regarding the active process ID, we could have the LLM "figure it out" like a human would based on what the foreground process is.
Ideally we could send chunks of the logcat messages to the LLM to give it more context. This is a general approach question though. Because we run inside a separate process, on-device. The instrumentation process can't access the logs of another application while running on-device.
It could be possible with "host mode" to view logs at some point, but that's currently out of scope as we are investing in on-device.