[Bug] Agent spams no-op echo probe commands to flush shell output

Bug Description
Title: Agent emits bursts of no-op echo probe commands to flush laggy shell output What happens: When the Bash tool's output is slow or buffered, the model spams dozens of throwaway commands like echo s1, echo s2, … echo s40 (sometimes sleep N; echo x) between real commands, apparently to coax buffered output to flush. Each renders in the terminal as: ● Bash(echo s37) ⎿ s37 ● Bash(echo s38) ⎿ s38 Dozens of these per turn bury the actual work and clutter the screen. They serve no purpose for the user. When it started: Began after the Claude Opus 4.8 upgrade. Was not happening on prior model versions. Scope: Happens across multiple/all projects, independent of project config. Expected: Run the real command once and wait for its result; tolerate shell latency. If waiting on async work, use a single backgrounded command with a real completion signal — never a burst of no-op echoes. Workaround in use: Added an explicit "never fire no-op probe commands" rule to global CLAUDE.md, but the model should not need this instruction — the default behavior regressed. Environment: Claude Code, model claude-opus-4-8 (1M context), Linux.

Environment Info

• Platform: linux
• Terminal: vte-based
• Version: 2.1.156
• Feedback ID: 4fc28aae-97fb-44bf-8b80-c9e14ac353a0

Errors

[]