Run a one-off agent query from the CLI

agents & sessions

// what it does

openclaw agent runs the assistant once from the terminal — --message is the prompt, --thinking sets reasoning depth (for example high), and --deliver sends the reply back out to a channel instead of only printing it. --to targets a contact; --agent picks which agent answers.

// shell

$ openclaw agent --message "summarize today's incidents" --thinking high
$ openclaw agent --agent ops --message "status" --deliver --to +15555550123

// gotcha

Without --deliver the reply only prints locally; with it, the response is actually sent to --to — double-check the target before delivering to a real contact.

// resources