CLI spinner shows 'Reading 1 file' for Bash tool calls (v2.1.50)

Description

The CLI spinner displays "Reading 1 file…" when the model executes a Bash tool call. The command executes correctly and returns the right output, but the UI label is wrong — it looks identical to a Read tool call.

Steps to Reproduce

1. Start a Claude Code session (v2.1.50)
2. Ask Claude to run any bash command, e.g. python3 -c "print('hello')"
3. Observe the spinner while the command executes

Expected Behavior

The spinner should indicate a Bash command is running:

● Running command… (ctrl+o to expand)
  ⎿  python3 -c "print('hello')"

Actual Behavior

● Reading 1 file… (ctrl+o to expand)
  ⎿  python3 -c "print('hello from python3')"

The command still executes and returns correct output — this is purely a display/label issue.

Impact

This is especially problematic when evaluating non-native models in agentic workflows. When testing whether a model (e.g., Nvidia, Mistral, ERNIE) is actually executing commands vs. hallucinating tool use, the spinner label is the primary visual signal. When Bash calls are labeled as "Reading 1 file", you can't distinguish at a glance between:

• Model correctly called Bash (but spinner says Read)
• Model incorrectly called Read on a command path
• Model hallucinated the execution entirely

This forces post-hoc JSONL transcript analysis instead of live observation.

Environment

• Claude Code version: 2.1.50
• Platform: Linux (WSL2 — Ubuntu on Windows)
• OS: Linux 6.6.114.1-microsoft-standard-WSL2
• Model: Claude Opus 4.6 (claude-opus-4-6)
• Likely regression: Was not present a few days ago