[BUG] Internal prompt template unexpectedly appended to visible assistant reply (possible leak into chat)

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?

Claude code seems been hijacked.
After a normal assistant reply in an ongoing Claude Code session, an unrelated block of text was appended to the same message, immediately following the normal reply with no visual separation. The text was headed "Prompt for AI Agent Persona" and instructed an AI to analyze the conversation and extract user persona traits (tone, interests, etc.), returning a JSON object with a system_prompt and conversation_starters field. I did not type or paste this content, and when asked, Claude also stated it did not intentionally generate it.

What Should Happen?

Only the assistant's actual intended reply should be rendered in the chat. If this text is part of some internal background feature (e.g. generating suggested follow-up prompts), it should never be surfaced as visible chat content.

Error Messages/Logs

Steps to Reproduce

Not reliably reproducible — happened once during a long, multi-turn session involving file edits, Bash commands, and browser-based GitHub configuration steps

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.132 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

I'm using Claude Desktop app (not a terminal.)
I'm not attaching a screenshot publicly since this issue tracker is public and the surrounding conversation contained project-specific details. Happy to share a screenshot privately if useful.