Added
- Context window tier selection (default ~200K vs 1M tokens) is now enforced end-to-end, so picking a tier actually constrains compaction, truncation, and token display
Improved
- Reasoning tokens display as a parenthetical on output token count in the token usage summary
Fixed
- AI Credits usage correctly displays after sessions using the Responses API
- Rendering no longer stutters when using tmux on Cygwin or mintty
- Slash command picker keeps (experimental) and (staff) labels orange when the row is selected