After Hermes Agent uses the terminal tool to start background commands, the output will not always be flashed in the current answer, but will be managed through the process tool. The process given in the official tool documentation is: return session_id from background start, and then view and control it with list, poll, wait, log, kill, and write.
The right way to see it
- List background processes: Have Hermes invoke
process(action="list"). - See if it's done: Use
poll. - Wait ends: Use
wait. - See the full output: Use
log. - Terminate when stuck: Use
kill.
Why is it designed this way?
Long tests, builds, and server starts that keep occupying the current tool call will prevent the agent from continuing inference. The background process mechanism allows it to get session_id and then view the logs on demand. This is suitable for running tests, starting dev servers, and handling long tasks.
If you want to interact with the CLI tool, also check if you need PTY mode. Normal background processes are suitable for non-interactive commands; Tools that need to be entered for Hermes to use process write or PTY.
