User Manual    [Previous]   [Next]   

Umple AI Skills

This page describes AI agent skills for working with Umple and the Umple CLI (Command line interface). This is in beta and will be enhanced.

A 'skill' is a technology released by Vercel that is gaining acceptance in 2026. The npm skills package is here.

Repository overview

The umple-skills repository (https://github.com/umple/umple-skills or initially here ) provides reusable AI agent skills, scripts, and guidance for working with Umple models and diagrams. This page gives a high-level overview and links to separate sections for each skill.

• Install the skills pack. Use npx skills add umple/umple-skills to add the skills to your agent environment.
• Skill definition (SKILL.md).
  • name declares the skill identifier used for routing and invocation.
  • description summarizes when the skill should be selected.
  • allowed-tools lists the executable commands the agent may run.
  • Prompt body (the markdown content) gives step-by-step guidance and output expectations.
• Scripts and references.
  • scripts/ holds runnable helpers (for example, scripts/main.ts) invoked by the skill.
  • references/ contains focused guidance (for example, diagram rules) that the prompt cites.

Skills follow a lazy-loading strategy: only the selected skill folder is loaded at runtime. Semantic routing uses the name and description fields to pick the best skill for a user request.

umple-diagram-generator

• Generates class diagrams and state machines from natural-language requirements by producing Umple models and rendering them to SVG. It uses guidance from the references/ folder and relies on the Umple CLI, Graphviz, and Bun.
• Output location. The skill invokes the Umple CLI through scripts/main.ts.
  • Scenario A: output path not specified. The script creates a folder named after the diagram plus a timestamp, and stores all generated files inside it.

    <output-dir>/
    └── <diagram-name>_<timestamp>/
        ├── model.ump
        ├── model.gv
        └── model.svg

  • Scenario B: output path specified. If you provide a full .svg path, the script writes only the SVG file to that exact location.

    <output-dir>/
    └── <diagram>.svg