Carousel
Agent permissions carousel
An 11-page visual framework for scoping agent tools and keeping destructive actions behind human approval.
- Who it is for
- Teams giving an AI agent tools that can change data or send messages.
- How to use it
- Separate what a tool can do from what the agent may do.
- Format
- Carousel, PDF, 11 pages.
What is inside
- Tool capability versus permission
- Least-privilege boundaries for consequential actions
- A visual prompt-ignored permission test
Goes with the article The admin access trap: prompt injection is only half the problem.
Browse the research library