Terminal UI¶

agent6 tui opens the full-screen interface on this repository; agent6 tui <id> opens one session. This page shows it screen by screen; the plain CLI is on Command line. Every image is from a recorded run; click to enlarge.

Hub¶

The new-work composer (n) picks the mode, the checkout (this one or a new worktree), the preset and the model; a second row picks the plan to run from and a spend cap in USD. A plan's Run this plan opens the same composer with the plan chosen. On a terminal 140 columns wide or more the table adds a run's branch; R opens the composer with /review <branch> for the selected run, and m merges it after a confirm that names the unmerged runs that changed some of the same paths. The composer accepts /parallel [SPEC --] TASK (grammar and examples). A session that starts waiting on an answer or ends is announced with a toast and a desktop notification (notify-send, when present).

Files¶

The Files screen opens from the hub and shows the working tree's files. A Markdown file opens rendered, and s shows its source.

Conversation¶

Opening a run lands on its conversation: the task, the model's reasoning and every tool call, following live. Esc or Ctrl+K returns to the hub, the TUI's session switcher. The composer takes the steer directives.

Ctrl+C copies the selection, or the whole transcript when nothing is selected
Ctrl+Z leaves the view: a run started with agent6 run --tui detaches after its current turn (agent6 attach reattaches); one opened with attach --tui or from the hub keeps running as it was
on a terminal narrower than a modal's button row the later buttons are off screen; the question modal's fields and Ctrl+S still answer

Run dashboard¶

Ctrl+D toggles the dashboard: task graph beside live reasoning, tool calls with results, event log and latest commit diff side by side. A session that dispatched children shows them in a table (id, model, status, usage), refreshed every second while the session lives, a child that appears later included; Enter on a row opens that child.

Findings¶

Open Findings from the hub's View menu (v) or the run dashboard's View menu (Ctrl+F).

Receipts¶

Open Receipt (e) from the Findings screen to inspect recorded evidence.

Code-review rule sources¶

See Code-review rules for discovery and limits.

Event log¶

Configuration¶

The config page shows every setting, its effective value, and the layer that set it: default, preset, global, repo, or flag (a --config FILE). Editing a role's model opens one provider/model field with live suggestions; unlisted routes are accepted.

In a named-entry form, a blank field keeps an existing value or uses the schema default.

Remove entry (x) confirms the selected entry's name and config layer before removing the whole entry, including its subtables; on the reviewer roster it picks a reviewer by position. It removes from the highest layer holding the entry, and accepts only global or repo config. Values from lower layers may become visible. The edit popup of an entry's row offers the same Add and Remove for its table.

Preset… (p) opens the agent6.preset setting: the rows show the config under the selected preset, and a save, an unset or an entry change writes [presets.<name>] in the chosen layer; Add preset… (P) creates a table and selects it; Remove preset removes the selected table and resets preset (Presets). With none picked, edits go to the global and repo config; entries are edited only then.

Keys¶

Watching a state machine¶

An agent state machine runs in the terminal like anything else: author the file, read its graph, watch it execute. Here code-fixer runs a fix loop: an agent state edits the repo to make a failing check pass, a tool state re-runs it, and the machine routes on the result until the check is green or the attempt budget is spent.