canvas

Overview of canvas skill

What canvas does

The canvas skill creates JSON Canvas diagrams with free positioning, making it a strong fit for mind maps, knowledge graphs, concept maps, and planning boards. It is designed for cases where layout matters more than sequence, and it uses explicit x/y coordinates so the diagram can be placed with intention instead of guessed by the model.

Best-fit use cases

Use this canvas skill when you need a spatial diagram that reads like a map: clustered ideas, linked notes, research structures, or visual planning. It is especially useful for Diagramming workflows where you want nodes, labels, and relationships to be editable and compatible with Obsidian Canvas-style output.

When not to use it

If your goal is a step-by-step process, choose a flowchart or UML-style skill instead. If you need data visualization, charts, or numeric comparison, canvas is the wrong tool. The main adoption blocker is expecting automatic narrative flow from a format built for spatial arrangement.

How to Use canvas skill

Install and locate the source

Run npx skills add markdown-viewer/skills --skill canvas to install the canvas skill. After install, read SKILL.md first, then open references/syntax.md for the attribute rules and examples. This repo has one supporting reference file, so you do not need to hunt through a large file tree.

What to provide in your prompt

A good canvas request should include: the topic, the main clusters, the number of nodes, any required links, preferred reading direction, and whether the diagram should be compact or spacious. For example, instead of “make a canvas about product strategy,” ask for “a canvas for Diagramming a product strategy workshop with 1 central node, 4 workstream groups, 12 child notes, and links between risks and dependencies.”

Practical workflow

Start from the content, not the layout. Give the model raw material first, then ask it to place nodes on a grid and connect them. A strong canvas guide prompt usually says what belongs in text nodes, what should be grouped, and what should be linked. If you already know the structure, name the groups explicitly so the skill can assign coordinates consistently.

Files and rules to check first

Read references/syntax.md to confirm required node fields, supported node types, and color options. In SKILL.md, the most important rules are the JSON structure, node requirements, and edge connection format. Those are the details that most often break canvas install or usage when they are missed.

canvas skill FAQ

Is canvas only for Obsidian Canvas?

No. The format is Obsidian Canvas compatible, but the canvas skill is really about generating JSON Canvas diagrams that can be used in compatible viewers and workflows. If you already work with canvas-based note systems, this is a natural fit.

What makes canvas different from a normal prompt?

A normal prompt may describe a diagram in prose; canvas produces structured diagram data with nodes, positions, and edges. That makes it better for repeatable Diagramming tasks, easier editing, and cleaner downstream use. It also reduces guesswork around layout, which is the biggest difference in practice.

Is canvas beginner-friendly?

Yes, if you can describe the content clearly. You do not need to know the JSON syntax in advance, but you do need to supply a sensible structure. Beginners get the best results when they specify a small number of groups first, then expand with child nodes.

When should I avoid canvas?

Avoid it when the output must be a process map, sequence diagram, or analytical chart. It also becomes less useful if you cannot define the relationships between ideas, because the skill depends on meaningful node placement and edges.

How to Improve canvas skill

Give the model a layout brief

The fastest way to improve canvas output is to specify topology, not just topic. Say whether the center should branch outward, whether clusters should be separated, and which nodes deserve visual emphasis. For a canvas guide request, details like “three clusters, left-to-right reading, one summary node per cluster” are much more helpful than “make it look nice.”

Provide complete node content

Strong canvas usage depends on having the actual labels ready. If you only give a theme, the model has to invent structure and may produce generic nodes. Better input looks like: “Central node: Customer onboarding. Group A: acquisition steps. Group B: activation risks. Group C: support materials. Include 2 links from risks to support materials.”

Watch for common failure modes

The most common problems are missing required fields, overlapping nodes, too many nodes in one cluster, and edges that do not reflect the intended relationship. Another frequent issue is asking for diagram types that do not belong in canvas. If that happens, refine the prompt by stating what should be text nodes, what should be file or link nodes, and how far apart the groups should sit.

Iterate with a tighter second pass

If the first output is structurally correct but visually weak, ask for a revised canvas with adjusted spacing, shorter labels, or clearer group hierarchy. If the first output is conceptually weak, replace vague labels with specific nouns and verbs. The most reliable improvement comes from giving the canvas skill better source content, not just asking for a prettier layout.