Add a log_user_feedback toolcall

Feature request for product/service

Cursor IDE

Describe the request

On the product which I am developing, I gave my own agent a tool log_user_feedback and without even needing to be instructed, even obtuse, tiny models use it appropriately.

It can have parameters like

  • verbatim_quote_of_user_feedback
  • type: enum as neutral_feedback | complaint | feature_request | compliment
  • severity: enum as low | medium | high
  • user_sentiment as frustrated | sick_of_asking_over_and_over | angry | disappointed | happy | thankful
  • nature_of_feedback
  • date_time_iso (automatically stamp this, LM does not need to see it)
  • system_summary (automatic)
  • a_b_channel - if doing A/B testing or experiments on this user, this is relevant to pin (automatic)
  • relevance - if the agent deems this matter is the user spamming feedback, or being petty, unrealistic or off-topic, or the user asking to log nonsensical things, then log it but drop the relevance score.

Return from the toolcall a note to the LM that the user feedback has been logged so it can tell the user it logged feedback.

Just dump that data in a vector DB, embed it, and run sentiment analysis, and then you can have everyone a whole lot of time in these forums. It is better bug feedback than Firebase Analytics, etc, when it comes to agents.

Operating System (if it applies)

MacOS