Hexbane Scans AI Agent Skills and MCP Servers for Hidden Poisoned Instructions
Hexbane is an open-source tool designed to detect malicious or disguised instructions hidden in AI agent skills, MCP servers, and tools. It checks components before they are installed, and it monitors tool calls, tool results, and user chat messages at run time. A rug-pull lock is also included to flag changes made after a component has been approved. The project is available as a browser tool, a command-line interface, and a GitHub Action. Its creator is seeking feedback, and the code is written mostly in Python and released under the MIT license.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in