<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Agent Wiz is a Python CLI tool for developers, researchers, and security teams to analyze and visualize AI agentic workflows. It addresses the complexity of modern LLM orchestration by extracting agent-tool interactions from code, mapping relationships, and performing automated threat assessments to enhance system visibility and security.

How It Works

The tool uses Abstract Syntax Tree (AST)-based static parsing to extract agent/tool definitions and interconnections from source code, generating interactive workflow graphs. For security, it applies established threat modeling frameworks, prioritizing MAESTRO, to systematically identify potential vulnerabilities. This approach offers a framework-agnostic method for understanding and securing complex AI systems.

Quick Start & Requirements

Installation: pip install repello-agent-wiz. Prerequisite: Set OpenAI API key (export OPENAI_API_KEY=sk-...). Core CLI usage:

1. Extract workflows: agent-wiz extract --framework <framework> --directory <path> --output <file.json>
2. Visualize graphs: agent-wiz visualize --input <file.json> --open
3. Analyze threats: agent-wiz analyze --input <file.json> See Agent Wiz GitHub Issues and PyPI.

Highlighted Details

• Supports numerous frameworks: AutoGen, CrewAI, LangGraph, LlamaIndex, etc.
• Extracts agents, tools, transitions, data flows, and group agent types.
• Generates interactive D3-based HTML visualizations of agent connections.
• Automates threat assessment using MAESTRO (Mission, Assets, Entrypoints, Security Controls, Threats, Risks, Operations).

Maintenance & Community

Provides CHANGELOG.md for version history and CONTRIBUTING.md for contributors. Specific community channels (Discord/Slack) are not detailed in the README.

Licensing & Compatibility

Licensed under the Apache 2.0 License. This is permissive, allowing commercial use and integration into closed-source projects without significant copyleft restrictions.

Limitations & Caveats

Additional threat modeling methodologies (STRIDE, PASTA, LINDDUN) are under development. Agent simulation-based threat exploration is also a planned future feature, indicating active evolution.