Architecture¶

rup-flow is a skill plus a small command line. The skill tells an AI coding agent what to do and what never to do; the command line does the work that must be exact: creating structure, keeping an index in sync, drawing diagrams and checking the result. There is no service, no account and no dependency to install beyond Node.js 18.

This section explains how the pieces fit, in the order a newcomer needs them.

The shape in one picture¶

A person asks an AI coding agent; the agent reads the rup-flow skill, runs its CLI and writes the project files; the CLI draws diagrams through a PlantUML server on the same machine

The person never talks to rup-flow directly. They talk to an agent. The agent reads the rules of the skill, runs the CLI for the mechanical steps and writes the documents itself, because understanding a business is the part that needs a language model. Each side does what it is good at, and the boundary between them is a plain command line that any agent with a shell can use.

Principles¶

Principle	What it means in practice
One portable core	The skill follows the open Agent Skills format. Anything specific to one agent lives in a small adapter and never in the skill
The agent analyses, the CLI does the boilerplate	Structure, indexes, rendering and checks are deterministic code; judgement stays with the agent
Honesty over completeness	The rules forbid inventing content, and the CLI refuses what an edition does not include instead of faking it
Nothing to install	Node.js 18 or newer is enough; Docker only draws diagrams
Two editions, one code base	The free CLI loads an extension when it finds one; there is no fork