Fragile detection of curlens's own summarization calls

[cnighut]

Problem

Currently, curlens detects its own cursor agent -p summarization calls by checking if the chat name is "New Agent" (for chats created after Jan 2026). This is fragile because:

1. Cursor could change the default name for -p chats at any time
2. If they make it meaningful (e.g., based on prompt content), our filter would break
3. We'd start summarizing our own summarization requests, causing circular/redundant entries

Current Workaround

• Skip chats named "New Agent" created after Jan 1, 2026
• Additionally filter out meta-chats containing summarization prompt phrases

Ideal Solution

If Cursor CLI adds a --name flag for -p mode:

cursor agent -p --name "curlens-internal" --model grok "Summarize..."

We could then reliably filter by a unique name we control.

Action Items

• Monitor Cursor CLI updates for naming options
• Consider alternative detection methods (e.g., workspace path, prompt fingerprinting)
• Add more robust meta-chat detection phrases as fallback

Related

• Cursor CLI docs: https://cursor.com/cli