Windows: Full conversation re-renders on each tool call, causing duplicate display and potential token waste

Description

On Windows 11, Claude Code's TUI re-renders the entire visible conversation history every time a tool call completes. This causes the same response blocks to appear multiple times on screen, making the output very hard to read.

Steps to reproduce

1. Run Claude Code on Windows 11 (tested on v2.1.89 and v2.1.90)
2. Ask a question that triggers multiple sequential tool calls (e.g., Glob + Read + Bash)
3. Observe the terminal output

Expected behavior

Each tool call should update in-place. The previous conversation turn should appear only once.

Actual behavior

The previous turn's full response is re-displayed from scratch every time a new tool call starts or completes. For a turn that triggers 5+ tool calls, the same response block appears 5+ times stacked vertically.

Example

A single turn with ~8 tool calls (Glob x2, Read x3, Bash x1, Read x4) produced output where the same assistant response block was repeated 5 times before the final answer appeared:

> user prompt

- [assistant response - full block]
  Simmering...

> user prompt           <- repeated

- [same assistant response - full block again]
  Simmering...

> user prompt           <- repeated

- [same assistant response - full block again]
  ... (repeats 5 times total)

Potential impact

• Readability: Very hard to follow what is happening when the same content repeats 4-5 times
• Token consumption: If the re-rendered conversation is being sent back to the model (not just a display issue), this could explain the excessive token usage many users have reported. Each tool call would re-send the full visible conversation, multiplying input tokens significantly.

Environment

• OS: Windows 11 Pro 10.0.26200
• Terminal: Windows Terminal
• Shell: bash (Git Bash)
• Claude Code: v2.1.90
• Model: Claude Opus 4.6 (1M context)
• Node: v24.14.0
• npm: 11.9.0

Notes

• PostToolUse hooks are configured but the re-rendering happens regardless of hook output
• The issue may be related to TUI in-place update not working correctly on Windows terminals
• Japanese text (multi-byte characters) in the output may contribute to rendering calculation issues