Track Agent Pane UX refinements

[vanzue]

Goal

Track the recent Agent Pane UX refinement work as one coherent experience initiative. The work improves interaction clarity, input efficiency, activity feedback, tool-call readability, and mouse usability without changing the underlying ACP architecture.

This is a retrospective umbrella for already-merged work plus the remaining in-flight refinement. Architecture-only decomposition remains tracked separately by #515.

Input and navigation

• Align selection styling across agent pane pickers #467 — align selection styling across Agent Pane pickers
• Add per-tab prompt input history #478 — add per-tab prompt history with draft and multiline preservation (closes Support reviewing prompt history with Up/Down #479)
• Add agent subcommand autocomplete #487 — add /agent <prefix> autocomplete and inline ghost text
• Fix multi-line user prompts collapsing to one line in agent pane transcript #493 — render multiline user prompts correctly in the transcript (closes Multi-line user prompts render as a single truncated line in agent pane transcript #492)
• Add per-tab agent pane move command #429 — add /move for per-tab Agent Pane positioning
• Unify inline notice styling #570 Notification polish and model selection UI polish
• Improve agent pane layout at compact heights #580 Better visual when height is limited

Visual polish

• Make agent pane title bars follow their theme (#348) #381 — make Agent Pane title bars, text, and logos follow the pane theme (closes Theme not respected by the Agent bar (bottom of terminal) #348)

Activity and progress feedback

• Improve tool call status styling #499 — replace raw tool-call status text with semantic markers and colors
• Refine agent pane activity indicators #501 — refine Thinking and active tool-call breathing indicators
• Pin TUI thinking indicator above input #508 — pin the Thinking indicator directly above the input box for the full in-flight turn

Historical context: #189 tracked the original missing Thinking indicator during permission/tool activity and was resolved by #434. The PRs above are subsequent UX refinements.

Mouse interaction and text selection

• Implement WTA chat mouse interactions and text selection #506 — add wheel scrolling, drag selection, word/line selection, and explicit copy behavior
• Scrolling in agent management not working as expected #513 — Agent Management mouse-wheel behavior (related implementation: Implement WTA chat mouse interactions and text selection #506)

Historical context: #128 requested broader mouse-friendly Agent Pane behavior but was closed as by-design before this implementation landed.

Tool-call readability

• Show tool-call/permission target and kind glyph, not just generic title #509 — show concrete tool targets and kind glyphs, split multi-statement commands, and establish clearer visual hierarchy and conditional spacing

Completion criteria

• Prompt history and multiline input render correctly.
• Agent selection supports keyboard completion.
• Thinking and tool execution have stable, non-jumping activity feedback.
• Chat supports first-class mouse scrolling, selection, and copy.
• Tool-call surfaces clearly communicate the operation and concrete target.