RightNow Agent can delegate independent work to child sessions and keep long-running commands active without blocking the conversation. Each subagent has its own context and reports a result to its parent.
The task tool starts a child session with a focused prompt, a short description, and an agent type. A task can block until the child finishes or run in the background and return an ID for later inspection.
| Agent type | Use |
|---|---|
| general-purpose | Full-capability work that may read, edit, and execute under its permissions. |
| explore | Read-only investigation, search, and shell inspection. |
| plan | Codebase exploration followed by a structured implementation plan, without file edits. |
Only the top-level session can create subagents. A child cannot create another child, so the maximum nesting depth is one.
| Shortcut | Pane | Contents |
|---|---|---|
| Ctrl+G | Tasks | Subagents, background commands, monitors, loops, statuses, elapsed time, and task IDs. |
| Ctrl+T | Todos | The current task's separate todo list. |
A command launched through run_terminal_command with background: true returns a task ID immediately. Press Ctrl+B to move an already-running foreground command into the background.
| Tool | Purpose |
|---|---|
| get_command_or_subagent_output | Read current output and status, optionally waiting up to a timeout. |
| wait_commands_or_subagents | Wait for any or all of up to 20 task IDs. |
| kill_command_or_subagent | Stop a command or subagent, or report that it already exited. |
The monitor tool runs a long-lived script and turns each new stdout or stderr line into a conversation notification. Filter noisy streams and use line-buffered pipes so events arrive promptly. A persistent monitor runs for the lifetime of the session until it is killed.
/loop 5m Check whether the test suite passes
/loop 2h Summarize new commitsIntervals use seconds, minutes, hours, or days. The minimum is 60 seconds. A loop fires immediately, creates a new turn on each interval, expires after 7 days, and is managed through the scheduler tools.
A persona is a behavioral layer for a subagent. It can shape tone, output format, task focus, model, reasoning effort, isolation, and declared input or output contracts without replacing the underlying agent type.
Open /personas to manage personas, or /config-agents to open the full Agents and Personas modal. Project personas live under .rightnow/personas/*.toml; user personas live under ~/.rightnow/personas/*.toml.