Problem
In the sidebar Agents panel, each agent's model line ended with a bare thinking label, e.g. anthropic/claude-opus-4-8 • on. The • on / • off reads as if the model is on/off rather than its thinking/reasoning effort. The label was also the first thing truncation removed (google/gemini-3.1-pro-previe…), and a single on covered both adaptive and token-budget thinking.
Proposed solution
Give thinking state a self-describing visual language:
- A compact effort gauge (6 lossless cells,
▰▰▰▱▱▱) for effort levels, with a dim empty gauge for off, auto for adaptive, and ◉ <count> for a token budget.
- Split the ambiguous
on into adaptive vs. a token count at the source.
- Left-truncate model names so the informative tail (variant/version) survives.
- Show the gauge together with the written value on the focus card / details modal so the card teaches the gauge.
Scope
Sidebar rendering, a runtime thinking-label tweak, and shared toolcommon helpers (effort gauge, thinking description, token formatting, left-truncation). No config-schema changes.
Problem
In the sidebar Agents panel, each agent's model line ended with a bare thinking label, e.g.
anthropic/claude-opus-4-8 • on. The• on/• offreads as if the model is on/off rather than its thinking/reasoning effort. The label was also the first thing truncation removed (google/gemini-3.1-pro-previe…), and a singleoncovered both adaptive and token-budget thinking.Proposed solution
Give thinking state a self-describing visual language:
▰▰▰▱▱▱) for effort levels, with a dim empty gauge foroff,autofor adaptive, and◉ <count>for a token budget.onintoadaptivevs. a token count at the source.Scope
Sidebar rendering, a runtime thinking-label tweak, and shared
toolcommonhelpers (effort gauge, thinking description, token formatting, left-truncation). No config-schema changes.