leash

README

LeaSH — LLM Execution Audited SHell

A policy-enforced shell execution engine for LLMs and agents.

Getting started

Install

git clone https://github.com/bornholm/leash.git
cd leash
make build          # produces ./leash

Or download a pre-built binary from the Releases page.

Run

# Interactive REPL
./leash --policy policies/default.yaml repl

# One-shot execution
./leash --policy policies/default.yaml exec --exec 'echo hello | tr a-z A-Z'

# MCP server (stdio, for Claude Desktop and other MCP clients)
./leash --policy policies/default.yaml mcp stdio

Use as an MCP tool

Add to your MCP client configuration (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "leash": {
      "command": "/path/to/leash",
      "args": ["--policy", "/path/to/policy.yaml", "mcp", "stdio"]
    }
  }
}

Documentation

• CLI reference — all commands and flags
• Policy files — control what the engine is allowed to do
• Skills — Go — register custom Go functions as shell commands
• Skills — Tengo — write skills as Tengo scripts (no compilation)
• Skills — Shell — write skills as POSIX shell scripts
• MCP server — expose LeaSH as an MCP tool set
• Go library — embed LeaSH in your own application
• Security model — how isolation and enforcement work

License

GPL-3.0