[eric-engberg]

What

In plain mode, manual separators placed next to each other between visible widgets all render again. Two spaces give a two-space gap, ' ' '•' ' ' gives • , and two pipes give two pipes. The status line and the TUI preview both change.

Separators on either side of a widget that renders empty still collapse, now a run at a time: a run with a visible separator keeps the boundary, and a run of spaces is replaced by the separators after the empty widget. With a single separator on each side, nothing changes.

One case that looks different: deleting the widget between two separators leaves both in the line, and both now show, as they do in the editor's list. Before, the preview showed one.

Why

renderStatusLine scans back from each manual separator to decide whether it renders. Since #525, an earlier separator met on that scan either owns the boundary (a visible one: the current separator is dropped) or is replaced (a spacing one: it is popped from the output). That rule is for separators on either side of a widget that renders empty, where one of them has to go. But the scan applied it whether or not a widget lay between the two, so separators placed side by side were deduplicated too. Each #525 test with two separators puts an empty widget between them, so none covers adjacent ones with every widget visible.

The TUI can build these lines: Space cycles a separator through |, -, , and space, and k copies the selected one.

Piped render with a scratch HOME, colorLevel: 0, custom-text widgets, Bun 1.4.2:

Line	main	this branch	before #525 (ef521ee9^)
A ' ' ' ' B	A B	A B	A B
C '|' '|' D	C | D	C | | D	C | | D
E ' ' '•' ' ' F	E•F	E • F	E • F

Calling renderStatusLine directly gives the same results for [A, ' ', '|', B] (main: A | B, now A | B) and [A, '|', ' ', B] (main: A | B, now A | B).

How

• In the backward scan, a separator met before any empty widget is part of the current separator's run. The scan steps over it to the content before the run, so every separator in the run renders.
• Across an empty widget, fix(renderer): preserve separators after hidden widgets #525's rule now applies per run. A run that contains a visible separator keeps the boundary, and the run after the empty widget is dropped: [A, ' ', '•', ' ', hidden, '|', C] renders A • C (main: A•C). A run of spaces is replaced by the run after the empty widget, and that whole run renders. Only that run's first separator removes the spaces it replaces, so the rest of its run is kept: [A, ' ', ' ', hidden, ' ', '|', C] renders A | C (main: A | C).
• With Inherit Separator Colors on, each separator in a run takes the colors of the widget before the run, which is what fix(renderer): preserve separators after hidden widgets #525 set up for single separators. Before fix(renderer): preserve separators after hidden widgets #525, only the first separator in a run inherited.
• Six new tests in renderer-separator-collapse.test.ts: four cases of adjacent separators between visible widgets, plus the two cases above around an empty widget. fix(renderer): preserve separators after hidden widgets #525's tests are unchanged and pass.

Left as is:

• Powerline mode drops manual separators ("manual separators disabled"). That doesn't change.
• docs/USAGE.md describes collapsing only around widgets that render empty, which matches this behaviour, so the docs are unchanged.

Demo

Line 1 is Model, ' ' '•' ' ', Context Length, |, Git Branch, |, Git Changes. The demo opens it, turns the | after Context Length into a space with Space, then copies it twice with k for a three-space gap. Before, the preview shows Claude•Ctx and a single space, and k leaves it unchanged. After, it shows Claude • Ctx and the gap widens with each copy. Powerline mode ignores manual separators, so its two recordings look the same; they are there as a control.

Powerline

Before:

After:

Plain

Before:

After:

Testing

• On main as of [low] perf(claude-status): hold a lock while the status page fetch is in flight #635 and [medium] perf(git-review): key the PR cache through the cached runGit #639 (62a371b): merges cleanly, bun run lint is clean, and bun test gives 3709 pass, 0 fail. The built CLI under Bun 1.4.4 and Node 26.13.0 matches main in plain and Powerline mode with the baseline configs, and the TUI opens and exits cleanly.
• bun test: 3174 pass, 0 fail (178 files); main: 3168 pass, 0 fail. The 6 new tests all fail on main. bun run lint is clean.
• renderer-separator-collapse.test.ts under vitest on Node 26.10.0: 25 pass (main: 19 pass).
• Built CLI, Bun 1.4.2 and Node 26.10.0: with the baseline demo configs, both runtimes agree and match main in plain and Powerline mode. With this demo's configs, plain differs from main as intended, Powerline matches main, and both runtimes agree. The TUI opens and exits cleanly under both.