I prefer a clear CLI operation to a prompt trying to anticipate every move an agent makes. Reusing work should reduce how much I need to explain. When the solution requires memorizing a manual for every task, maintenance has merely moved somewhere else and now consumes context too.
Mistral introduced Spaces in March 2026 as a CLI for people and agents, covering scaffolding, configuration, and staging deployments. That shared interface is what interests me, because an operation I can understand at a terminal can also give an agent a predictable way to execute part of a job.
Take a command that prepares a project structure and reports what it created. The skill explains when to use it and how to check the result. If the request needs a different structure, the agent should recognize the difference. Forcing every task to fit the available command is a remarkably laborious way to save effort.
Parameters need to explain what they select. If one identifies an environment, I want that meaning to be legible. When a prerequisite is missing, the tool should name it. Then I can choose the next step without treating a vague message as success, and so can an agent calling the same interface.
The response also needs to describe the change made. Getting back just "done" means I have to hunt down the result before continuing. I do not want a giant report; I want enough information to connect the operation with the state it left behind. It is a small, concrete thing to demand of a CLI.
My rule is to keep documentation close to the command and have the skill refer to it. Copying the entire contract into several prompts multiplies maintenance when the tool changes. Before I call a skill useful, I test two requests: one the command handles and one that should make the agent leave it.