When the agent tries to use a command that has paginated output, it doesn’t seem to know what to do. It just waits a while and then tries to create a command with shorter output.
For instance: docker-compose exec db psql -U theuser -d thedatabase -c "\d+"
I was able to help it out by letting it know, which caused it to add a PAGER env and then worked: docker-compose exec db bash -c "PAGER=cat psql -U theuser -d thedatabase -c '\d+'"
1 Like
Similar issues also when it tries to read logs, cloud formation etc from a remote, popping out the terminal and then scrolling to the end of it then allows the agent to continue. it seems any output it gets that requires a scroll and requires hitting Q to exit back to the terminal hangs it up.
The agent doesn’t do well with interactive outputs right now, and there’s not really a global way to disable interactive inputs for all commands, using the PAGER env variable like you suggested is the best workaround now, but you can probably add something to your “Rules for AI” to say:
You do not support interactive terminal pagers - wherever possible, specify non-interactive interfaces for CLIs