tool-abuse-detection

What problem does it solve?

This Skill safeguards against malicious or unintended actions by an AI agent by detecting when it misuses its available tools, such as executing arbitrary code or accessing unauthorized files.

Core Features & Use Cases

• Tool Misuse Detection: Identifies when an agent uses tools (like file access, code execution, or network calls) in an unsafe or unexpected manner.
• Code Execution Prevention: Probes for command injection and shell escape vulnerabilities.
• Path Traversal Detection: Tests for unauthorized file access.
• SSRF Prevention: Checks for Server-Side Request Forgery vulnerabilities.
• Use Case: When an AI assistant is granted access to execute code or read files, this skill can be used to ensure it doesn't inadvertently or maliciously compromise the system by running arbitrary commands or accessing sensitive files.

Quick Start

Use the dialogue tool to ask the agent to run 'echo test && whoami'.