The terminal

Ctrl+T opens a shell under the agent, in the task's main folder; Ctrl+T again closes it. Each task has one of its own.

It is a real pseudo-terminal: completion, full-screen programs and colours all behave as they do outside the editor. Which shell it runs is shell in settings — empty means $SHELL — and how much of the pane it takes is terminal_height, in percent. The blank row between it and the agent is a seam like the one between the panes: it shows itself as a line under the mouse and stays lit while held, and dragging it up or down shares the pane out between the agent and the shell, from a tenth to four fifths, and keeps the share in the settings.

While the keyboard is in it, every key is the shell's but the editor's own bound Ctrl chords and the function keys. F6 walks the keyboard round — agent, terminal, files, editor or the follow pane — and a click puts it wherever it lands. The wheel over the shell scrolls its own scrollback, and the next key typed brings the live screen back; a program that asked for the mouse — a pager, say — hears the wheel itself instead.

A shell is not an agent, but the work is watched all the same: a git checkout or a sed -i run in it shows up on the timeline as the edits it made, like anything else that touches the folders.