[BUG] Internal session title generation response leaks into conversation history display

Problem / Background

When viewing conversation history in Claude Code, the history entry shows internal auto-generated text that should not be visible to users:

Okay, let's tackle this! Here's a title and summary based on your message: **Tit...

This appears to be the beginning of a response Claude generates internally when creating a session title/summary — but this internal metadata is being exposed as the conversation history label or preview text.

Steps to Reproduce

1. Start a new conversation in Claude Code
2. Send a message
3. Open the conversation history / session list
4. Observe: the history entry shows Okay, let's tackle this! Here's a title and summary based on your message: **Tit... instead of a clean session title

Expected Behavior

Conversation history should display either:

• A clean, human-readable session title derived from the first user message
• A summary of the conversation topic

Actual Behavior

The raw internal response text used for title generation is shown verbatim and truncated in the history display, e.g.:

"Okay, let's tackle this! Here's a title and summary based on your message: **Tit..."

Acceptance Criteria

• [ ] Internal title/summary generation responses are not exposed in the history UI
• [ ] Conversation history shows a clean title or the first user message as a fallback
• [ ] Truncation is handled gracefully with proper ellipsis if the title is long

Technical Considerations

• The session title generation appears to use a Claude API call internally; the raw response text is being stored or displayed directly
• The fix likely involves parsing the generated title/summary out of the response before storing it, rather than storing the full raw response
• Alternatively, a dedicated structured output format (e.g., JSON) could be used for title generation to avoid this issue

Environment

• Platform: macOS (darwin 25.1.0)
• Shell: zsh