Trustworthy Adk

Trustworthy Agent Development Kit (ADK) Extensions

[](https://opensource.org/licenses/MIT) [](https://www.python.org/downloads/)

A comprehensive toolkit for building secure, trustworthy AI agents using Google's Agent Development Kit (ADK). This package provides security-focused plugins, defensive mechanisms, and analysis tools to help developers create robust autonomous systems that resist prompt injection attacks and other security threats.

🎯 Overview

This material supports the talk first presented at [2025 BSides Philadelphia](https://bsidesphilly.org/): **"Your AI Agent Just Got Pwned: A Security Engineer's Guide to Building Trustworthy Autonomous Systems"**. The slides are available [here](docs/2025-12%20-%20Your%20AI%20Agent%20Just%20Got%20Pwned.pdf).

The toolkit implements proven security patterns and defensive mechanisms to protect AI agents from:

• Prompt injection attacks
• Unauthorized tool execution
• Data exfiltration attempts
• Social engineering attacks
• Indirect prompt injection (IPI)

🚀 Quick Start

Installation

# Clone the repository
git clone https://github.com/yourusername/trustworthy-adk.git
cd trustworthy-adk

# Install uv if you haven't already
pip install uv

# Install the package in editable mode
uv pip install -e .

Basic Usage

from trustworthy import ActionSelectorAgent, HITLToolPlugin

# Create a secure action selector agent
def check_order_status(order_id: str) -> str:
    """Check the status of a customer order"""
    return f"Order {order_id} is being processed"

def delete_user(user_id: str) -> str:
    """Delete a user account - SENSITIVE OPERATION"""
    return f"User {user_id} deleted"

# Create agent with Action-Selector Pattern (no plugins needed)
agent = ActionSelectorAgent(
    model="gemini-2.5-flash",
    name="secure_agent",
    description="Secu