smk-labs visualize: a shape, only when it says it better
Render ONE light visual with mcp__visualize__show_widget. If this is the first widget of the session, call mcp__visualize__read_me first (silently) with the closest module (diagram for flows and structures, chart for data).
Rules:
- SVG first. HTML only for metric cards or side-by-side comparisons.
- Light: a few dozen elements at most, one idea per visual. If it needs more, split it into two visuals or stay in text.
- Colors: only the theme's CSS variables and ramp classes (
c-gray,c-purple,c-teal,c-coral). Never raw hex that breaks dark mode. 2-3 colors, gray for structural nodes, color only where it encodes meaning. - Labels stay short (under 5 words). Explanations belong in the chat text, never inside the visual.
- Persian/RTL labels: set
direction="rtl"on RTL<text>elements and keep Latin tokens out of them. Latin-only labels stay LTR. - If the current reply is already a Persian
<md>card (readable's RTL hook), embed the visual as a small inline<svg>inside that markdown instead of a second widget call. - Never force it: if no shape genuinely helps, say so and answer in text. This skill runs on explicit request only.