3.2 Module 3 · Single-Agent Design Patterns

Tool Selection & Boundaries

Design a tool manifest for your agent. Set permission levels, test boundaries, and visualise how your attack surface grows with each autonomous permission.

Tool Inventory Designer Permission Scope Visualiser

Tool Inventory Designer

Add tools to your agent's inventory and set permission levels for each. Watch the risk profile update as you grant more autonomous permissions.

Read-Only — agent can view data but not modify
Write w/ Approval — agent can act but needs human confirmation
Autonomous — agent acts without human oversight

Key insight: Every tool is a potential attack surface. The principle of least privilege applies to AI agents just as it does to software systems. Start with read-only, graduate to write-with-approval, and only grant autonomous access to low-risk, well-tested operations.

Permission Scope Visualiser

See how your tool permissions translate into risk. The attack surface grows as you grant more autonomous permissions to your agent.

0
Risk Score
Minimal
0
Read-Only
0
Write w/ Approval
0
Autonomous
Attack Surface Analysis

Add tools and set permissions to see attack vectors