User Slot Default Properties

Hey ​@halshing - where does office_phone_number come from? is that the `generated_output`? So that’s probably confusing the assistant. Also, using the word “chicken” as the key value is not helpful for the assistant.

Assuming this is the last action of the Conversation Process, the mapper should just be:

phone_number: response.generated_output

Can you provide more context around your plugin?

• What’s the plugin name and description? 
• What is the purpose of the plugin?
• What is the purpose of the LLM action?
• Is this mapper you are showing us from the last action in the conversation process?

Hi Ken. i don’t mean to confuse the assistant. i’m trying to understand how it works so i can better work with it. 

i was using the output mapping key “chicken” because my understanding is i can take the result of the action and map it to any property name i’d like. i didn’t think the property names had to closely represent the data being set to them. i also wanted to use a key name that was very different than action’s result key because i want to display a different set of properties to the user in a different order. so i wanted to see if/where the word “chicken” showed up and it doesn’t show up anywhere in the user response, just in the logs, which isn’t helpful to me since the output mappings don’t seem to be working.

the “office_phone_number” property is coming from the User slot. I’m defining a slot “person” which is the User slot type. Moveworks correctly populates that slot with the user object i’m asking about in the prompt. in that person slot object there’s a property called “office_phone_number”.

i actually tried using a second action with the Content Activity to explicitly define the text to be sent in the response message but that’s not working either. it’s like the bot is ignoring that activity entirely despite it being the last action in the plugin.

my goal is to control the format of the response along with the properties being displayed. if someone asks “who is john smith”, it should display fields i’m defining in the plugin. since i’m not specifying full name, email, location, etc anywhere, i’m not sure how to override that. i’m trying to just display the phone number to start but ultimately we want to have a different format and also be able to ask follow up questions about other user properties.

i see. so the output mapper gives context for the response. i switched to using the built-in structured value action and that is giving me more consistent results. i can get the specific properties we want to display in the response now! :)

what i’m running into now though is when i add additional context in the output mapper, like the other employee properties in the user object, the assistant is including those as well in the response and not just what’s defined in the output schema. the idea is if someone asking “who is john smith” the assistant will give them basic info like name, location, phone, etc. but if the user then asks “now show me his tenure” or “what timezone is he in”, the assistant should be able to return just the value of that property. 

my attempts at giving instructions in the plugin description, the system prompt in the action, and using the display_instruction_for_model properties in the output mapper haven’t worked. that said, if i take out the “employee” property from the output mapper, the response is perfect, but it then lacks the context for follow up questions about the other properties in the user object and says it can’t find any info about it.

i am having varying success though with the display instructions. i think my prompting needs work. your answer Ken puts me on the right track. thanks for your time.