svg-precision

Generate structurally correct SVGs from a strict JSON spec (scene graph), then validate and optionally render a PNG preview.

Fast path

• Turn the user request into a Spec JSON (use templates in references/spec.md ).

• Build the SVG:

• python scripts/svg_cli.py build spec.json out.svg

• Validate:

• python scripts/svg_cli.py validate out.svg

• (Optional) Render a PNG preview (requires CairoSVG):

• python scripts/svg_cli.py render out.svg out.png --scale 2

Spec design rules (for accuracy)

• Always set canvas.viewBox and explicit canvas.width /canvas.height .

• Prefer absolute coordinates; use transforms only when they reduce complexity.

• Keep numbers sane: no NaN/inf; round to 3-4 decimals.

• Put reusable items in defs (markers, gradients, clipPaths) and reference by id.

• For predictable results across viewers, avoid exotic filters unless required.

• Text varies by fonts/viewers. If you need pixel-identical results, treat text as a risk and prefer shapes.

Using the bundled scripts

CLI

• python scripts/svg_cli.py build <spec.json> <out.svg>

• python scripts/svg_cli.py validate <svg>

• python scripts/svg_cli.py render <svg> <out.png> [--scale N]

• python scripts/svg_cli.py diff <a.svg> <b.svg> <diff.png> [--scale N] (renders + image-diffs)

As a library (in Python)

from svg_skill import build_svg, validate_svg svg_text = build_svg(spec_dict) report = validate_svg(svg_text)

When the request is vague

• Identify the kind of SVG: icon / diagram / chart / UI / technical.

• Pick a template from references/spec.md and fill in concrete numbers.

• If dimensions are unknown, choose defaults that match the domain:

• icons: 24x24 or 32x32

• UI mockups: 1440x900 or 390x844 (mobile)

• charts: 800x450

• diagrams: 1200x800

• technical drawings: specify units (mm/in) and scale

References

• references/spec.md

• schema + ready-to-copy templates per SVG type

• references/recipes.md

• layout and styling patterns that render consistently