Show HN: A UML drawing skill for your coding agent docs

An Agent Skill by BESSER that gives your AI coding agent one job and makes it do it right: put a correct UML class diagram — as a real, rendered image — into your docs.

This skill supports two common workflows:

1. Design a system from a description. Describe your domain in natural language, and the agent creates a B-UML model and renders the corresponding UML class diagram.

You:   "Add a class diagram of our vehicle fleet to the README."
Agent: ‣ models Vehicle · Car · Truck · ElectricCar with correct inheritance
       ‣ renders it to a real SVG itself — one call, no browser
       ‣ embeds  ![Vehicle fleet](examples/vehicles.svg)

2. Document an existing codebase. Point the agent to your source code, and it extracts the classes and relationships, creates the B-UML model, and generates the diagram.

You:   "Diagram the classes in src/fleet/ for the docs."
Agent: ‣ reads your existing code, recovers the classes and how they relate
       ‣ builds the same validated B-UML model and renders the SVG

The result is a diagram generated from a validated model that can be included directly in README files, design documents, wikis, and presentations.

The agent's actual output — rendered from

examples/vehicles.py

in a single call, no browser or plugin. Whether you describe the model or the agent reads it from your own source, the diagram comes from real, validated code. That's the skill at work.Models are a powerful way to quickly understand your codebase. A class diagram, for example, provides a compact view of a software system by showing its classes, attributes, and relationships. It is often used in documentation, README files, and design discussions because it allows developers to understand the structure of a system without reading the complete source code.

But when you ask an AI agent to create one, things can easily go wrong in two ways:

A wrong diagram— reversed arrows, invalid multiplicities, inheritance backwards. It looks fine until someone who knows the domain reads it.A block of text that never becomes a picture— pseudocode or ASCII "explaining" a diagram inside a code fence that no viewer actually renders.

This skill fixes both: a structurally-correct model, delivered as an image that really renders — on GitHub, GitLab, wikis, and slides, no plugin.

Most "diagram" tools give you a picture or a model. This gives you both, and keeps them in sync:

Correct by construction. The diagram is always built as a structurally-validatedBESSERB-UML model — not freehand ASCII the model guessed at. Multiplicities, associations, and inheritance are right.Start from a description Describe the domain, or point the agent at existing source and let it model the structure.oryour code.Two ways to embed — and the agent renders the image itself. Drop theB-UML code straight into your.md

, or get a realSVG/PNG: the agent renders it with a single call to BESSER's headlessB-UML → SVG

endpoint — no browser, no Mermaid plugin, no rendering service. (Want to hand-place the layout? Export fromeditor.besser-pearl.orginstead — same model.)It doesn't drift. Both come from one model you keep in the repo. Change the model, re-deliver, commit — the doc never goes stale.The diagram can become the system. The same model generates working code — Python, SQL, FastAPI, Django, React, and more. Your documentation diagramisyour source of truth.

npx skills add BESSER-PEARL/uml-drawing --all

Or copy the uml-drawing/

folder into your agent's skills directory (.claude/skills/

, .agents/skills/

, …).

Model it— the agent builds a B-UML class model from your description or from existing code it reads (always class diagrams, always via B-UML).Deliver it— embed the** B-UML code**in your.md

, or get a renderedSVG/PNG: one HTTP call to BESSER's headless endpoint (no browser), or a hand-tuned export from the BESSER editor when layout matters.Keep it current— the model is the source of truth; change it and re-deliver. The doc never drifts.

Full instructions live in SKILL.md; your agent loads them automatically when a task matches.

Ready-to-render B-UML models live in examples/, one per class-diagram feature the skill handles:

vehicles.py

tasks.py

org.py

enroll.py

Point your agent at any of them and ask it to add the diagram to your docs.

This skill is the "diagrams for docs" front door to BESSER. For the full platform — deep modeling, every generator, troubleshooting, contributing — see ** besser-skills**.

Prefer to model by hand? Drag classes around, tune the layout, then generate working code from the same model — open the BESSER web editor at ** editor.besser-pearl.org**. It's the same B-UML underneath; the skill just lets your agent drive it for you.

Apache-2.0.