Hey all, I can integrate open API spec actions into my custom GPT, but every time the action call is triggered, I need to click the “Validate” button for the UI prompt “Some info will be sent to , only do this for sites you trust”. Any way to avoid this dialogue button click altogether and just invoke the action every time?
This is human in the loop protection, I do not believe it can be disabled.
I think you can control that with the Consequential flag.
- If the
true, we treat the operation as “must always prompt the user for confirmation before running” and don’t show an “always allow” button (both are new features of GPTs designed to give users more control).
- If the
false, we show the “always allow button”.
- If the field isn’t present, we default all GET operations to
falseand all other operations to
After you install actions in a new GPT, go to “GPT’s privacy settings”. There the added action is listed, and one can toggle “Ask”, and “Always Allow” for the action. I selected “Always allow”, but it didn’t make a difference, the “Validate” button still showed up.
@seer-pint0t where could I put this flag, in open API schema of the action? I tried that and added it under operationId field but it was rejected/ignored.
There might be two separate aspects, one is “do you want us to send your information to this server?” and the other is “what we’re about to do could have an impact, so please explicitly confirm we should do it”
@seer-pint0t did this work for you? I added this to my spec(for a POST endpoint) and it still requires explicit approval each time(no always allow button appears).
However, the Action is not updated now, so you have to Delete and Add for it to work properly.
Yes in the schema under operationId field. Here is the link to the documentation
for me it worked. The first time I run the action after including this I got the “Always Allow” button and after clicking that I was not prompted again.