Super Agents currently include required default tools such as edit_self, execute code, workspace search, task retrieval, and scheduling. I am unable to toggle those permissions off and they cause me extreme concern due to prompt injection possibility.

That creates a meaningful security concern for agents that process untrusted external content such as forwarded emails, form submissions, or imported text. A prompt-injection attack could potentially exploit those built-in capabilities even when the agent's instructions explicitly say not to.

It would be much safer if admins could disable individual default tools per agent, especially:

Edit self

Execute code

Search Workspace

Retrieve task lists

Create schedules

Persistent memory/write-to-do capabilities

For narrow-purpose agents, least-privilege controls should be enforced by the platform rather than relying primarily on prompt instructions.