Currently we do not have a fixed way of doing validation on how the result from a prompt ended up. We should have a fixed abstracted way of doing this, so people knows where to look for this.
This means that we could also add retries into the prompting part instead of the whole agent, where that specific query can be retried until it fits the criteria.
Later on you could have validation agents here as well, that does more complex validation if the understanding was correct, rather than the data structure.
Create an interface for this
Create an JSON version of the interface for this
For now keep it simple with one method checking the output and responding TRUE when working and FALSE or a string to add to the next try when not working.
Active
1.0
Code
Automatically closed - issue fixed for 2 weeks with no activity.