The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the user to clone a code repository from an untrusted source: https://github.com/SHAdd0WTAka/Zen-Ai-Pentest.git. This repository is not associated with a verified or trusted organization.
[REMOTE_CODE_EXECUTION]: The installation guide requires the user to run pip install -r requirements.txt and npm install within the cloned untrusted repository, followed by executing the application using docker-compose or uvicorn. This process leads to the execution of code provided by an unverified remote source.
[COMMAND_EXECUTION]: The framework is designed to orchestrate and execute real security tools such as nmap, nuclei, sqlmap, and ffuf on local or remote targets. While this is the intended functionality of the framework, it grants the agent the ability to execute high-impact system commands.
[PRIVILEGE_ESCALATION]: The manual setup instructions include using sudo apt-get install to install security packages, which requires the user to grant administrative privileges to the installation process.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of target domains or URLs which are then passed to various security scanners.
Ingestion points: Target domain/URL input provided to create_scan and run_scan functions in SKILL.md.
Boundary markers: None identified in the provided code snippets to distinguish between instructions and data when invoking external tools.
Capability inventory: Subprocess execution of 72+ security tools including nmap, sqlmap, and nuclei via the agent system.
Sanitization: The skill references a SafetyValidator class that checks if a target is public and blocks private IPs/localhost, but doesn't show sanitization against command injection in tool arguments.

zen-ai-pentest-framework