The software factory
A team's lines together make up its software factory. Each line is a route that work takes through stations, and each station says which agent works there, what it is told, what it may use and where the work goes next. Spotlight runs the factory. What happens at each station, and in what order, is for the team to decide.
Why each team shapes its own
Teams build software in different ways. One triages every request before anyone plans it; another writes a failing test before any code. One lets agents run whatever they need in their container; another wants to approve each command. In the docs' view, a single fixed pipeline would suit one of them and get in the way of the rest.
So the parts of the factory are configuration:
- a line's stations, and where each passes its work
- each station's prompt, the coding tool its agent runs (its harness, such as Claude Code) and the model
- the skills a station's agents can read, and the tools they can call to reach a team's own APIs
- the rules that move a task when a check fails or a pull request is merged
- what agents are allowed to read, edit and run
Most of these are set on a line, in
spotlight_line. Permissions are set in a
harness profile, and skills and
tools are defined as skills and
integrations that a station names.
A new workspace starts with two lines, Delivery and Review. They are one arrangement of these parts, included so the factory works before the team has configured anything. A team changes them once it has seen them run, or replaces them.
Why the factory is code
A station's prompt and permissions decide what an agent does in a team's
repositories. A change to them is the kind a team wants reviewed before it
happens and traceable afterwards, like a change to the code the agents work on.
A Terraform configuration in a repository gives both: terraform plan shows
what a pull request will change in Spotlight, and the repository's history is
the history of the factory.
That is also why the web app shows the factory but doesn't change it. Why configuration lives in Terraform covers the choice in full, with the alternative that was turned down and what the choice costs.
How a factory improves
A factory improves through the same kind of change. In the docs' view, the places where agents struggle show what to change. An agent that keeps asking the same question points at a prompt that leaves something open. Work that Code Review keeps sending back points at a station missing earlier in the line. A team makes the change, applies it, and the next runs show whether it helped.
Make the factory yours puts a workspace's factory into a repository and makes a first change to it. Lines, stations and tasks explains how a task moves through a line once the factory is running.