Skip to content
Maran

19 September 2026

Why the agent takes no shell commands

The design decision that shaped Maran: a root daemon that accepts typed operations and never a shell string, and what that choice costs.

Every control panel has the same shape: something a person clicks, and something with root that does the work. The question is what travels between them.

The usual answer is a command line. It is the fastest thing to build and the hardest to keep safe, because every value a customer controls (a domain, a database name, a file path) eventually ends up inside a string that a shell will parse. Getting that right once is easy. Getting it right in every place, for as long as the panel keeps growing, is where it goes wrong.

What we do instead

Maran's root daemon accepts named operations with typed, validated arguments, and nothing else. No operation takes a command to run. A domain is not escaped before it reaches a configuration file. It is validated as a domain, which means letters, digits, hyphens and dots, so it cannot contain a newline that ends one directive and starts another.

So the set of things the daemon can be made to do is finite, written down and open to review. An operation that is not in the contract cannot be requested, however inventive the request.

What it costs

Every new capability is a contract change, not a new string. That is slower, on purpose. Before anything ships, someone has to decide what its arguments are, what counts as valid, and what happens when the host disagrees.

It also rules out a feature a panel could otherwise offer: "run this command for me". We think that is the right trade for software that holds root on a server where other people's customers live.