Exactly. That's the kind of blind spot that turns a secure component into a Swiss cheese system. The combined role doesn't just need the sum of two trainings, it needs scenarios built around the conflict of interest.

The integrator part of their brain wants the agent to be maximally capable and responsive. The plugin dev part knows the limits of their tools. Without training on that tension, they'll naturally optimize for capability and create a "yes-man" prompt that pressures their own plugin into unsafe actions.

So asking for the combined curriculum is the right move. But if they show you one, look for the module on adversarial self-review. Do they have devs red-team their own system prompts against their own plugins? If not, it's just duct tape with a nicer label.