Plug-and-play terminal security layer for LLM agents. Drop-in gatekeeper that prevents dangerous shell commands. Works with OpenAI, Claude, Gemini & more.

SecureShell

A plug-and-play security layer for LLMs and Agents that prevents dangerous command execution.

SecureShell acts as "sudo for LLMs" - a drop-in zero-trust gatekeeper that evaluates every shell command before execution. It blocks hallucinated commands, prevents platform mismatches (e.g., Unix commands on Windows), and helps agents learn from mistakes.

Why SecureShell?

LLM agents with shell access can hallucinate dangerous commands like rm -rf / or dd if=/dev/zero. SecureShell solves this by:

• Zero-Trust Gatekeeper - Every command treated as untrusted until validated by independent gatekeeper
• Platform-Aware - Automatically blocks Unix commands on Windows (and vice versa)
• Risk Classification - GREEN/YELLOW/RED tiers with automatic handling
• Agent Learning - Clear feedback helps agents self-correct
• Drop-in Integration - Plug into LangChain, LangGraph, MCP, or use standalone
• Multi-LLM Support - Works with any LLM provider

Quick Start

TypeScript

npm install secureshell-ts

import { SecureShell, OpenAIProvider } from 'secureshell-ts';

const shell = new SecureShell({
    provider: new OpenAIProvider({
        apiKey: process.env.OPENAI_API_KEY,
        model: 'gpt-4.1-mini'
    }),
    template: 'development'
});

const result = await shell.execute(
    'ls -la',
    'L