ipa-lab/hackingBuddyGPT

What it solves

It provides a standardized framework for security researchers and penetration testers to build and test LLM-driven agents that can autonomously perform security testing, such as privilege escalation and web API pentesting, without having to rebuild the basic connectivity and logging infrastructure from scratch.

How it works

The framework uses a loop-based execution model where LLMs are connected to targets via connectors (SSH, local shell, or WinRM-style psexec). It supports two main execution styles: native tool-calling agents that maintain a chat history and use function calling, and simple-text command strategies that use Mako templates to generate single commands. It integrates with litellm to support a wide variety of LLM providers and uses OpenTelemetry/GenAI JSONL traces for structured logging.

Who it’s for

Ethical hackers, security researchers, and penetration testers who want to automate security testing using LLMs or autonomous agents.

Highlights

• Diverse Use-Cases: Includes pre-built agents for Linux and Windows privilege escalation, web testing, Web API testing, and Active Directory assumed-breach pentesting.
• Unified Run Limits: Ability to cap runs by rounds, tokens, cost, and wall-clock duration.
• Provider Agnostic: Supports any provider reachable via litellm, including OpenAI, Anthropic, and local models via Ollama.
• Ground-Truth Verification: Includes success detection for privilege escalation that verifies actual command output rather than relying on LLM claims.
• Benchmarking Tools: Ships with a Docker-fleet launcher to run regression tests against multiple vulnerable targets simultaneously.