Generate Component Doc Figma

generate-component-doc-figma — Figma component → Markdown docs

Produce a complete documentation page for one component or component set: overview, anatomy tree, design tokens, variants/states matrix, typography, accessibility, content guidelines, and (optionally) design-code parity and YAML frontmatter. This skill collects structured data from the file via use_figma, then a deterministic Node converter (scripts/generate-doc.mjs) emits the Markdown — the same collected JSON always produces identical Markdown. Do not freehand the doc from prose.

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).
• The collect script runs anywhere (use_figma); the converter is Node and runs in a terminal (node scripts/generate-doc.mjs ...). Same split as export-tokens-figma. If you can't run Node, you can't produce the doc — say so rather than freehanding it.
• Markdown section layout + the cleanVariantName rule → [references/doc-template.md](references/doc-template.md). The converter already implements every rule in that file; treat it as the spec, not a checklist for hand-assembly.
• Reading/writing annotations as standalone specs → use the annotations-figma skill.
• Exporting the whole token system (not just one component's tokens) → use export-tokens-figma.

Workflow

1. Identify the target. Get the component or component-set node ID from the current selection or a

URL/ID the user provides. Component sets (with variants) produce the richest docs. Set NODE_ID in the collect script. Also note the file URL (for the figma: link / frontmatter).

1. Collect data → save JSON. Run [scripts/collect-component-data.js](scripts/collect-component-data.js)

via use_figma (skillNames: "generate-component-doc-figma"). It returns the anatomy tree, per-variant colors (with bound token id + name), typography (font family, numeric weight + name, size, line height, letter spacing), spacing tokens, component property definitions (variants/booleans/text props), the description, and annotations (with category names). Save the returned JSON to a file (e.g. collected.json).

1. Generate the Markdown (deterministic). Run the Node converter in a terminal:

``bash node scripts/generate-doc.mjs collected.json [--code-info codeInfo.json] [--out docs/components/Button.md] [--frontmatter] [--file-url ] ` It ports the source generator's section logic exactly: Overview (+ When to / When NOT to Use parsed from the description), Component Anatomy, Variants (matrix + icon mapping + configurable props), Token Specification (color + spacing tables), Typography, Content Guidelines, Accessibility, Design Annotations, and — with --code-info — Implementation, Design-Code Parity, and Changelog. cleanVariantName is applied automatically (Type=Image, Size=12 → Image / 12`). Same input → identical output.

1. Optional code parity. To add Implementation / Parity / [View Source] & [Storybook] links,

read the component source and write a codeInfo.json (props, importStatement, sourceFiles, baseComponent, changelog, …), then pass --code-info codeInfo.json.

1. Optional frontmatter. Pass --frontmatter to prepend YAML frontmatter (title, status, version,

tags, figma URL, lastUpdated).

1. Review & deliver. Read the emitted Markdown, sanity-check it, then write/return it. Report the

path and a one-line summary (variant count, token count).

Notes

• Font loading: if you add a step that writes text into Figma (e.g. stamping the doc back onto the

canvas), await figma.loadFontAsync(...) first. Pure documentation generation reads only.

• The collect script uses MAX_DEPTH to keep payloads small for deep trees — raise it only if the

anatomy looks truncated.

• Token names come from boundVariables; a color/spacing with no bound variable renders as — in the

token tables (a hardcoded value to replace with a token).

• The converter is dependency-free (Node 18+, ESM). It exits non-zero with a clear message if the JSON

is missing or malformed.

• These collect tools require the Figma Desktop app (the Plugin API isn't available in the browser).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: southleft
• Source: southleft/skills-for-figma
• License: MIT

Install and usage instructions live in the source repository linked above.