Interrupt signals (Escape and Ctrl+C) show feedback but don't stop agent execution

Environment

• Platform (select one):
• [x] Anthropic API claudemax 20x
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other:
• Claude CLI version: latest 51. This started from 48 or 49
• Operating System: macOS
• Terminal: ghosty + tmux -

Bug Description

When attempting to interrupt a running agent by pressing either the Escape key or Ctrl+C, red feedback text appears in the terminal indicating the interrupt was registered, but the agent continues executing its task without actually stopping.

Steps to Reproduce

1. Start Claude Code and begin a task that involves an agent
2. While the agent is running, press either Escape or Ctrl+C
3. Observe the red feedback text appear in the terminal
4. Notice that the agent continues running despite the feedback

Expected Behaviour

When pressing interrupt keys (Escape or Ctrl+C), the agent should immediately stop execution after showing the interrupt feedback.

Actual Behaviour

Red interrupt feedback text appears in the terminal for both Escape and Ctrl+C, but the agent ignores the interrupt signal and continues executing its current job to completion.

Additional Context

• Both Escape and Ctrl+C exhibit the same behaviour
• The interrupt feedback suggests the keypress is being detected
• According to the docs, Ctrl+C should be the proper interrupt key
• This creates confusion as the visual feedback suggests the interrupt worked when it didn't
• The agent completes its entire task despite multiple interrupt attempts