This topic is about improving how Assistants handle actions that are impossible by design. Currently, an Assistant may claim it performed an action (for example, deleting an automation) even though the underlying tool does not support that operation. When the tool silently fails or the feature doesn’t exist, the user receives no clear error and is left thinking the Assistant actually succeeded.
This creates confusion, frustration, and inconsistent behavior across threads.
I’m proposing a change to how the Assistant handles “impossible actions” so users get accurate feedback instead of silent failures or misleading confirmations.