Status line

The footer has two rows aligned with the composer prompt. By default the first shows the model with its reasoning level (gpt-5.5 xhigh), and the second shows the working directory and Git branch. /statusline opens an inline setup view to change what each row shows.

Configure it

Row tabs in the header show which footer row is being edited; Tab switches rows. Fields already on that row are listed first with their position, then every other field under "Available", tagged when the other row owns it. Each field shows the value it renders right now, or why it renders nothing (no open PR, not on a Git branch). Branch and pull-request lookups only run once a row uses the field, so until then they read "shown once added", then "checking…". The focused field's description sits above the key hints.

Type to filter, use Up/Down to navigate, Space to add or remove a field, and Left/Right to change its position. Ctrl-P switches between dim and theme colors. Enter saves; Escape cancels. Adding a field assigned to the other row moves it to the current row.

The live footer below previews the draft: an accent bar marks the row being edited, the focused field's value is bold and underlined, and row 2 shows the draft even where it would normally show background-shell or agent activity or hide on short terminals. Enter keeps the layout only after a successful save; Escape or a failed save restores the saved layout.

Appearance

Each value's leaf (model, final folder, branch) is one step brighter than its context (reasoning level, parent folders), and separators recede further. Turn theme colors on to give the leaves subtle semantic accents that follow /theme; context stays muted.

Narrow rows shed detail before clipping: paths collapse from the middle (~/…/cluster/build-cli, down to ~/…/build-cli), then fields drop in the order input tokens, output tokens, context window, used tokens, task progress, pull request, project name, Git branch, context used, context left, reasoning level. Anything that still overflows ends in an ellipsis. The plan mode indicator retains priority on narrow terminals.

Available fields

Available IDs are model-with-reasoning, current-dir, model, reasoning, project-name, git-branch, pull-request, context-remaining, context-used, context-window-size, used-tokens, total-input-tokens, total-output-tokens, and task-progress.

Git branch discovery runs only when needed by the footer or setup preview, outside rendering, and refreshes after turns and shell commands. Project names reuse startup repository discovery, falling back to the nearest known project instruction root.

Pull request

The optional pull-request field shows PR #123 for an open GitHub PR. When gh reports the PR's web URL, the field is underlined and, in terminals that support OSC 8 hyperlinks, clicking it opens the PR. Only http(s) URLs link; a clipped field, a PR without a URL, and TERM=linux or dumb stay plain.

It reuses the installed GitHub CLI (gh) and its existing login: discovery tries gh pr view first, then PRs associated with HEAD in the upstream repository and fork. It never prompts for authentication. Missing gh, missing login, network errors, detached HEAD, or no open PR leave the field hidden.

PR lookup runs in the background only when selected in either row or preview, with an eight-second total timeout. Results (including no PR) are cached for 60 seconds per directory/branch; turns, shell commands, and picker activity refresh expired entries. Branch or directory changes invalidate the cache immediately. A newly created or closed PR on the same branch appears on the next refresh after the cache expires.

Usage and progress

Usage fields read the token counts the gateway reports with each model request and reset when /clear, a model switch, or an account change starts a new conversation. Input counts include cached tokens for every provider.

The context fields measure the latest request's input plus its completion against the usable context: the auto-compaction threshold or the model's window, whichever is smaller (the window alone with auto-compaction off). They are hidden until the first reply, and again after compaction until the next one.

The context window comes from /v1/models, looked up once per model in the background, only while a row or the setup preview shows a context field. It is hidden when the server doesn't report one; a failed lookup retries after 30 seconds.

Counts use at most three significant digits (950, 1.23K, 12.3K, 123K).

Saved settings

Settings are saved together in ~/.cluster/config.toml:

[status_line]
items = ["model-with-reasoning"]
secondary_items = ["current-dir", "git-branch", "pull-request"]
use_theme_colors = false

Omit items for defaults; items = [] hides first-row context. Omit secondary_items for the default working directory and Git branch, minus anything already on the first row; secondary_items = [] leaves the second row blank. The second row uses the existing bottom padding; it hides below 12 terminal rows and during transient footer hints. Plan mode stays on row one. Unknown IDs or duplicates within or across rows are configuration errors. Missing runtime values are omitted, including a branch in a detached-HEAD checkout.

Next steps

• Interactive sessions — the full slash-command reference.
• Config basics — other settings stored in config.toml.