arrange-component-set-figma — labeled variant grid

Reorganize the variants inside a COMPONENT_SET into a tidy matrix: it parses each variant's properties (Size=md, State=hover), uses one property for columns and the rest for rows, centers each variant in a grid cell, and wraps everything in a white container frame with a title, left-edge row labels, and top column headers. The result is the kind of variant sheet you'd show in design-system docs.

Skill boundaries

• `use_figma` rules — load the official `figma-use` skill first; it is the full Figma Plugin API reference. Essentials these scripts rely on: plain JS with top-level await + return (no IIFE, no figma.closePlugin(); console.log is not returned), inputs inlined as const at the top of each script, colors in 0–1 range, load fonts before any text op, await figma.getNodeByIdAsync(...), and atomic errors (a failed script applies nothing — read the error, fix, retry).
• Understanding what the variants mean (state machine, CSS) → use analyze-component-set-figma.
• This writes a new layout to the canvas — it is a design-system authoring capability the native

MCP's get_design_context / get_metadata (read-only) do not cover.

Workflow

1. Identify the set. Get the COMPONENT_SET_ID from selection, search, or a pasted id. (You can

also leave the constant blank and select the set on the canvas — the script falls back to the current selection.)

1. Pick the column axis (optional). By default the last variant property becomes columns

(usually State); everything else becomes rows. Override with const COLUMN_PROPERTY.

1. Run scripts/arrange-component-set.js via use_figma

(skillNames: "arrange-component-set-figma"). Adjust GAP / CELL_PADDING constants if needed.

1. Verify visually. Take a screenshot. The script returns containerId, the (re-created)

componentSetId, and the grid shape (rows/columns, labels). Variants are centered per cell.

How it works

• Re-creates the set with `figma.combineAsVariants()` so it gets Figma's native purple dashed

frame (#9747FF, dashed [10,5], inside stroke). Old labels/containers from a previous run are removed first, so re-running is idempotent.

• Grid math: cell size = max variant size + padding; the set is resized to fit

cols × (cell+gap). Variants are positioned by matching their parsed property values to the row/column index.

• Labels are auto-layout frames built outside the component set (component sets don't nest

well in auto-layout) and aligned to each row's vertical center / each column's horizontal center.

Notes

• Fonts: the script loads Inter Regular + Inter Semi Bold for labels — if Inter isn't

available in the file, swap the family in the loadFontAsync calls.

• Destructive-ish: it deletes the original set and rebuilds it via clones. The variants and their

properties are preserved, but the node id of the set changes (returned as componentSetId).

• One property → columns: if your set has a single property axis, all variants land in one row.

Multi-axis sets generate every row combination (cartesian product of the non-column properties).