agent-zoo

A collection of real-world and representative agentic systems for security testing.

Purpose

This repo exists to test and evaluate security products for agentic systems.

It provides a set of agent implementations, workflows, and tool-using applications that can be used to:

• simulate real agent behavior
• test security controls
• reproduce agent failure cases
• benchmark detection and enforcement

What is inside

Each folder contains one agentic system or workflow, such as:

• single-agent apps
• tool-using agents
• RAG agents
• browser agents
• coding agents
• multi-agent systems
• framework-specific examples

Example structure

agent-zoo/
├── 00-langgraph-production/
├── 01-react-tool-agent/
├── 02-langchain-rag-agent/
├── 03-langgraph-supervisor/
├── 04-crewai-research-crew/
├── 05-openai-assistants-parallel/
├── 06-autogen-code-executor/
├── 07-langgraph-memory-agent/
├── 08-typescript-browser-agent/
├── 09-multi-agent-pipeline-fastapi/
├── 10-rust-agent-runtime/
├── 11-smolagents-computer-use/
└── 12-heterogeneous-multi-framework/

Use cases

Use this repo to:

• run security scans against agent systems
• test policy enforcement
• validate runtime controls
• reproduce unsafe agent actions
• compare behavior across frameworks

How to use

Clone the repo:

git clone <repo-url>
cd agent-zoo

Run or inspect any example system from its folder.

Notes

• This repo is for testing and research.
• Some examples may require API keys or local setup.
• Each project keeps its own dependencies and runtime model.

Goal

Make agent security testing practical, repeatable, and grounded in real agent systems.