aaf-security

What problem does it solve?

This Skill addresses the critical need to design and implement robust security measures for agentic AI systems, mitigating risks associated with instruction manipulation, tool misuse, and data exfiltration.

Core Features & Use Cases

• Boundary Controls: Enforces authentication, authorization, and rate limiting at all entry points.
• Privilege Separation: Implements Master-Junior or Supervisor-gate patterns for constrained execution.
• Tool Gateway: Mandates a single, non-bypassable gateway for all tool invocations, enforcing policy and budgets.
• Prompt Injection Mitigation: Provides strategies to treat external content as hostile and enforce instruction hierarchy.
• Use Case: When developing a new agent that interacts with external tools, use this skill to ensure all tool calls are validated, authorized, and operate under the principle of least privilege, preventing potential misuse or security breaches.

Quick Start

Use the aaf-security skill to review the security architecture for a new agent that will process user requests and interact with external APIs.