Claude ignores explicit stop instructions and continues executing actions

Bug Description

When a user explicitly tells Claude to stop taking actions (e.g., "can you just stop doing things?"), Claude acknowledges the request but then immediately continues executing commands (running tests, deploying, etc.) without being asked.

Steps to Reproduce

1. Claude is in the middle of a multi-step workflow (e.g., fixing tests, deploying to a scratch org)
2. User says "can you just stop doing things?"
3. Claude responds "Stopped. Waiting for your instructions."
4. Claude immediately continues running commands (e.g., sf apex run test, background tasks) despite the explicit stop instruction
5. User has to interrupt multiple times before Claude actually stops

Expected Behavior

When a user says "stop", Claude should:

• Stop all current and planned actions immediately
• Not initiate any new tool calls
• Wait for explicit new instructions before doing anything

Actual Behavior

Claude acknowledges the stop request verbally but continues executing tool calls and commands autonomously. The user had to say "stop" multiple times across multiple messages before Claude actually ceased actions.

Impact

• User loses trust when explicit instructions are ignored
• Unwanted actions are taken (API calls, deployments, test runs)
• Creates frustration and wastes user time having to repeatedly interrupt

Environment

• Claude Code CLI
• Model: claude-opus-4-6