My setup:
I'm using actions-on-google to create a custom action for google assistant, every intent is a fallback in dialogflow that is then redirected to my server where I check the intent with a custom nlp engine and then respond to it, using the actions-on-google package.
The issue:
But when an intent is suicidal, for example: "I want to kill myself", google stop the action and give a response instead of letting my custom action handle it. Is there a way to avoid this?, it also happens with certain words like "Save" in that case google stop the action and ask me if I want to save an event or whatnot.
Can I override this behavior?
You're running into a "feature" that Google calls "no-match yielding". In some cases, if you're using a Fallback Intent in Dialogflow, and the Google Assistant itself can handle the question - it will do so.
To avoid this you can use regular Intents that match the @sys.any System Entity.
However, if all you're doing for everything is sending it to your NLU, then you don't need to be using Dialogflow at all. You can just use the Actions on Google v2 to send all TEXT Intents to your NLU for processing.