/Rag workflow vectorizing full natural language sentence

[mmwoolmer]

Via Claude, I asked the following question: What are the top 10 complaints related to "Irresponsible behavior by customer support"? I expected the Rag workflow to vectorize and search on the text in quotes, however this is what it shows:
"cleaned_question": "What are the top 10 complaints related to "Irresponsible behavior by customer support"?"

Another example is /rag What are the top 10 complaints related to "Identity Theft" and the "cleaned_question": "What are the top 10 complaints related to "Identity Theft"?",

The top returning result: Similarity: 0.794 "IDENTITY THEFT" I would expect this to be closer to 100% match.

I ran a test and just used /rag Identity theft and received a perfect match Similarity: 1.000 (Perfect Match)
"IDENTITY THEFT"

To Reproduce
Steps to reproduce the behavior:

1. Go to Claude front end and enter a natural language sentence for /rag as shown above

Expected behavior
The workflow should be able to take the specific prompt input and separate it out from the full sentence.

Desktop (please complete the following information):

• OS: Windows 11
• Browser N/A (Claude front-end)
• Version [Sonnet 4 is the model)

Additional context
Happy to be contacted on this