Agent Studio Slots infer with Resolver Strategies

I completely agree with your take ​@andrewwhitley - By any chance, has a ticket been raised for this issue? The Assistant should be resolving.

Another issue that could be causing the resolver to fail is that your resolver might be returning too much information instead of a list of names. Maybe you only need to return a list of names with associated IDs instead of the full API payload.

Thanks Kevin, returning the full payload could be part of the issue. I do need the id and possible other attributes depending on the slot, but I could reduce it to just those values; I will give that a try.

However Moveworks does normally recognize the correct value I am talking about as it calls it out specifically as the response, but asks for confirmation instead of inferring.

I don’t have a ticket for this specifically yet, I wanted to ask here first in case it was known or there were other workarounds like reducing the payload being returned.

Thanks for coming here first! I think the problem may be context bloat; it depends heavily on how much data your resolver returns, so either the AI isn't sure what to pick and needs your help, or it knows, but there’s so much data that it wants you to confirm, or it can’t figure it out at all.

You can also leverage the output mapper to prune the response to only the fields you need.

I leveraged the output mapper to include the 4 fields that I needed. I specified in the slot description to do your best to infer based upon the description slot, that happens first, and to compare against the specific attribute from the resolver strategy.

However I did not see any change in behavior. It still showed all options asking me to confirm the correct option for one slot. For the second resolver strategy slot, it was not able to guess or infer the possible option even though I had the exact value in my description slot.

Can you tell me which slots your plugin has and how they're set up? Are you creating any slot dependencies? 

Can you make sure you are collecting the slot and the context on the same step?

Here is the rough setup:

1. Decision Policy where I collect a description and a user
   1. Decision condition is True. It is setup like this as I was hoping collecting the description first would enable the slots to be inferred correctly.
2. Decision Policy where I collect my first resolver strategy slot (dynamic based upon api call)
   1. Decision condition is True as I need it to collect this slot before the next one
3. API call action where I also collect my second resolver strategy slot (dynamic based upon api call). It has a strategy mapping configured from the first slot resolver as it is dependent upon what was selected.
4. Decision Policy where I take different actions based upon what was selected.

When I removed the first decision policy and collected the description and user together with the first resolver strategy slot, it asked for the resolver strategy first and then the description after, so collecting them together is not working either. I need the description to help infer the second dependent slot resolver strategy as well. If that is not possible currently, I can let this go and just go with a generate structured value action after collecting the description.

​@Kevin Mok - I tried putting the description, application, and access_type all in the same action as required slots in the order I typed. However it asked for all 3 at the same time. I responded with just the description and it still asked me to confirm the application even though the name I provided was an exact match for both the application and type of access (the end user will never do this).

Hey ​@andrewwhitley - would you mind waiting for optional slots? That may solve the issue you are encountering.