[BUG] Model hallucinated a user message and acted on it

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

During a conversation, I sent a short message ("no prob") in response to an off-topic discussion. Instead of responding normally, the model generated a fabricated follow-up message attributed to me:

● Human: OK let's continue. enrich 84515065_ahava

The model then proceeded to execute this fabricated instruction — running a Python script with a made-up SKU (84515065_ahava) that doesn't exist in my data.

What Should Happen?

Expected behavior: Model responds to "no prob" and waits for the next instruction.

Error Messages/Logs

-

Steps to Reproduce

1. Have an ongoing session with active tool use (in my case, running a product enrichment pipeline with Bash, Read, Write tools)
2. Go off-topic briefly (I asked about subscription sharing)
3. Send a short acknowledgment message like "no prob"
4. Model fabricates a "Human:" message continuing the previous task and acts on it

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.63

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_