Feature request: nudge/interrupt action to check queued user messages during processing

Problem

When Claude Code is executing a chain of tool calls (e.g., monitoring a long-running process with repeated bash commands), user messages typed during processing are queued but not acknowledged until the current tool batch completes. The system-reminder tags ("IMPORTANT: After completing your current task, you MUST address the user's message") do appear, but only after the current tool finishes — which can be many seconds or minutes later during intensive monitoring loops.

Current workarounds

• Escape — cancels the entire response, typed text stays. Closest to a "nudge" but destructive (loses in-progress work)
• Ctrl+C — hard interrupt, even more destructive
• Ctrl+B — backgrounds the task, but doesn't make the model read the queued message

None of these are ideal. Users want to interject ("what's going on?", "stop that", "try this instead") without killing the entire response chain.

Proposed solution

A new action like app:nudge or chat:interruptAndRead that:

1. Pauses the current tool chain (doesn't cancel it)
2. Forces the model to read any queued user messages
3. Lets the model decide whether to address the message and resume, or change course

This could be bound to a keybinding (e.g., Ctrl+N) via ~/.claude/keybindings.json.

Context

This is especially painful during infrastructure work where Claude monitors long-running processes (model loading, quantization, builds) with repeated status checks. The user can see what's happening and wants to redirect, but has to wait or use destructive interrupts.

Environment

• Claude Code 2.1.101, Windows 11 + WSL2
• Opus 4.6 (1M context)