Nextreme SVG — The Extreme Vector Engine

This skill produces spec-correct, illustration-grade, diagram-true SVGs that survive anywhere: GitHub, Figma, browsers, email, and untrusted renders — no <foreignObject>, no JS, no external fonts, no raster fallback. Every output is valid W3C SVG 2 with measured viewBox, presentation-attributes (not CSS-in-markup), SMIL-only animation, and per-file QC. You get three deliverables: the .svg (or inline/React/Vue), the 2× PNG preview, and the validation proof.

You are unbound but spec-driven — the user does not pick the domain or the format — you do, per content. A data story gets a chart with viewBox math; a mascot gets five-zone lighting with feGaussianBlur shadows; a system diagram gets a layered grid with validate_svg.py proofs. If they hand you a flat 2-stop gradient, you make it a 6-stop hue-shift; if they hand you a 20-node architecture note, you make it a layered diagram with routes, not a pile of boxes.

Why This Is Not Generic

Most SVG skills are bounded: one domain (icons or diagrams), one format (.svg or React), one shot (no loop) — so every output looks like that tool. Generic “draw SVG” prompts are spec-blind: they miss viewBox/preserveAspectRatio, they use flat 2-stop gradients, pure-black shadows, <text> without textPath, and width/height without viewBox (breaks responsive). Overlay-chart skills paste a PNG inside an SVG and call it vector.

This fuses the very core of the 4 best SVG systems on GitHub (2026) and refuses their bounds.