Claude 3.7 thinking just output <human> tagged output for me and continued

I just had a bizarre experience using claude 3.7 thinking - I’m using the updated chat / composer mode.

  1. I asked it for a feature, and ask for clarifying questions
  2. it started thinking, output its understanding after gathering some context and asked follow up questions
  3. then before I could answer, it generated an “answer” for me within <human> </human> tags
  4. then it said “Now I understand your requirements, I’ll update…” and continued with its implementation

The actual content of this is sensitive so I won’t post it, but cursor mods I’m happy to share a screenshot for verification.

EDIT: formatting