I've been very interested in software factories lately but every time I've read about them, I see folks talking alot about high level concepts and the idea of a software factory but never actually walk you through how to set one up. Or if they do walk you through how to set one up, it's always overly complicated and requires a ton of up-front investment to get things running. In response to this I built rowbly, a CLI that tries to make it as easy as possible to understand and setup a software factory.
Rowbly at it's core works like a kanban board and each column can contain an optional prompt that instructs an agent what to do when a story is within this column. Once you have some stories on your board you run a supervisor skill that monitors the board for changes and spawns agents to work on stories within each column.
The default configuration ships with 5 columns: backlog, refinement, development, review and done. Once a story is moved into refinement, the supervisor automatically dispatches an agent to refine the story based on the instructions in that column. When refinement is finished the ticket is automatically pushed into the development column when another agent is spawned, then into review and finally into done when the PR is merged.
While this is a default configuration, things can get much more exotic based on how things are setup. Each column prompt will define where the story goes after it's finished in that column, so you could wire up a very sophisticated setup, review loops, security scans, testing, etc.
From a data perspective I tried to keep things as simple as possible, all data is stored in plain markdown and json files in the folder where you initialize rowbly.
If you're interested you can run it directly with npx using: `npx rowbly` or install globally with `npm install -g rowbly`. When you run it for the first time, it will walk you through adding stages, prompts and companion skills but all of this can be modified to your liking.
Github Repo: https://github.com/sgolovine/rowbly/