GITBUSTER-COIN/AgentGuardX

GitHub: GITBUSTER-COIN/AgentGuardX

Stars: 0 | Forks: 0

AgentGuardX

AgentGuardX Terminal Demo

AI Security Red-Team Toolkit

App -- Docs -- API

python license uv commitizen Ruff coverage build

## What is AgentGuardX? AgentGuardX is an open-source **AI Security Red-Team Toolkit** designed to help security researchers, developers, and organizations evaluate and harden LLM-powered applications and autonomous agents. It provides a modular testing framework for prompt injection, tool misuse, data exfiltration, unsafe agent behavior, policy bypasses, and common agentic AI security risks. ## Features | Module | Description | |---|---| | Attack Modules | Collection of practical red-team attack scenarios for LLM agents. | | Tool Analysis | Analyze tool integrations for risky configuration and misuse. | | Automated Scanning | Scan agents and workflows for common vulnerabilities. | | Reporting | Generate JSON and HTML security reports. | | Extensible Core | Add custom payloads, detectors, adapters, and policies. | ## Quick Start pip install -e . agentguardx scan \ --endpoint http://127.0.0.1:8000/agent \ --response-path answer \ --html report.html Run the mock target: uvicorn examples.mock_target:app --reload ## Project Structure AgentGuardX/ ├── assets/ # README banner and brand assets ├── agentguardx/ # Core Python package ├── payloads/ # Attack payload definitions ├── examples/ # Demo vulnerable targets ├── docs/ # Documentation ├── tests/ # Unit tests ├── .github/workflows/ # CI pipeline ├── CODE_OF_CONDUCT.md ├── CONTRIBUTING.md ├── SECURITY.md └── README.md ## Documentation - [Documentation](docs/) - [API Reference](docs/api.md) - [Threat Model](docs/threat-model.md) ## Ethical Use AgentGuardX is intended for authorized security research, defensive testing, and educational use only. Do not run tests against systems you do not own or have explicit permission to assess. ## Security To report a vulnerability, see [SECURITY.md](SECURITY.md). ## License Apache License 2.0. See [LICENSE](LICENSE).