Reads a Figma .fig file offline and writes text artifacts an AI can implement a screen from:
hierarchy, size, position, auto-layout, fills, strokes, effects, fonts and text, with component
overrides already applied.
| Tool | Version |
|---|---|
| Node.js | 20 or newer — tested on 24 |
openfig-core |
^0.4.1, pure JS, no native build |
# Install the dependencies once (node_modules/ is not committed)
npm install --prefix Scripts/Design/FigmaWithout
node_modules/, both commands stop with the exactnpm installline to run.
kings fig-parse <file.fig> [flags] # catalog, or the frames the flags select
kings fig-check [file.fig] [--update] # regression check against expected countsBoth run through node --max-old-space-size=8192 — parsing a real file peaks near 1 GB of heap.
fig-parse flags:
| Flag | What it does |
|---|---|
| (no filter) | Writes only the catalog (index.md + thumbnail.png) |
--frame <text> |
Extracts the top-level frame whose name contains <text> |
--id <sid:lid> |
Extracts by exact id — the reliable selector |
--all-frames |
Extracts every top-level container (use with --page) |
--page <text> |
Limits to pages whose name contains <text> |
--out <folder> |
Output root — default $KINGS_FIG_DIR |
--stdout |
Prints to the screen instead of writing to disk |
--json |
With --stdout, prints JSON instead of the tree |
--images used|all|none |
Which images to copy — default used |
--svg <id|name> |
Converts that vector node to SVG in icons/ — repeatable, opt-in |
--all |
In the catalog, also lists loose nodes |
--include-hidden |
Includes nodes with visible:false |
--no-expand |
Doesn't expand instances — debugging only |
--depth <n> |
Maximum tree depth — default 40 |
--max-nodes <n> |
Node cap per frame — default 20000 |
fig-check:
| Call | What it does |
|---|---|
kings fig-check |
Checks every .fig in $KINGS_FIG_DIR/figmas/ |
kings fig-check <file.fig> |
Checks only that file |
kings fig-check <file.fig> --update |
Rewrites its expected counts — for an intentional change |
Exit codes: 0 success · 1 error, mismatch or missing baseline.
KINGS_FIG_DIR (in ~/.kingsScripts/config.env) defaults to ~/Documents/FigExtracts.
| Path | What it is |
|---|---|
$KINGS_FIG_DIR/<fig-name>_<YYYY-MM-DD_HHMM>/ |
One new folder per run — -2, -3 suffix in the same minute |
…/index.md |
Catalog: pages and their top-level containers with id and size, links to what was extracted |
…/frames/<page>/<frame>--<id>.tree.txt |
The main read — one line per node |
…/frames/<page>/<frame>--<id>.json |
Same data, for values the tree hides (font postscript, gradient stops) |
…/images/<hash>.<ext> |
Rasters cited in the tree as fill:image(<hash>) |
…/icons/<name>--<id>.svg |
Only with --svg |
…/thumbnail.png, …/meta.json |
Thumbnail embedded in the .fig, and extraction metadata |
$KINGS_FIG_DIR/figmas/ |
The .fig files fig-check runs on by default |
$KINGS_FIG_DIR/fixtures/<name>.expected.json |
Expected counts per file — outside the repo, so private designs never get committed |
Result lines go to stdout; progress, warnings and errors to stderr.
- The typical failure is silent. A format change or a regression keeps parsing and returns
less — default component text instead of the screen's, for example. That's why
fig-checkcompares counts (nodes, expanded instances, applied overrides), not just "it ran". --depthdefaults to 40 on purpose — nested components easily pass 25 levels, and a depth of 8 cut three quarters of a real screen.- Frame names repeat a lot (several identical names on one page is common): use
--idfromindex.mdwhenever more than one frame matches. --svgis opt-in and never runs on a normal extraction — vector nodes come out asvectorPath: "unavailable", which is expected. Gradients andINSIDE/OUTSIDEstrokes are approximated, and the SVG says so in a comment.--images allcopies every raster in the file — more than 100 MB on a real one. The defaultusedcopies only what the extracted frames cite.