[FEATURE] Visually distinguish assistant prose from tool-call output in the transcript

Problem / use case

In a long session, the assistant's actual conversational replies — analysis, findings, summaries, recommendations, answers to direct questions — are visually indistinguishable from tool-call invocations and tool-result blocks. When scrolling back to find "what did Claude actually tell me," it's all one uniform monospace stream. The signal (Claude's prose) is buried in the noise (tool I/O).

Request

A subtle but consistent visual treatment that sets the assistant's conversational text apart from tool-call / tool-result output — e.g. a theme-aware accent, a left gutter bar, a faint background tint, or similar. Ideally opt-in via settings or theme. The goal is scannability when scrolling, not heavy decoration.

Why hooks don't solve this

Hooks fire on lifecycle events (PreToolUse, PostToolUse, SessionStart, etc.) and can block actions or inject context, but they have no access to how assistant messages are rendered in the TUI. There is no seam for a user to restyle assistant prose — this has to come from Claude Code itself.

Related

• #34269 — "Add visual distinction for user messages to improve scrollback navigation" — same scrollback-navigation motivation, for user messages. This request is the sibling: assistant prose vs tool output.
• #56316 — "Light-mode UX: past user messages are hard to distinguish from assistant output" — overlapping theme.

Neither covers distinguishing assistant conversational text from tool-call output specifically.