VI

visual-creative-subagent

Creates single-pass, browser-renderable visual concepts from scratch.

Install

mkdir -p .claude/skills/visual-creative-subagent && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/14787" && unzip -o skill.zip -d .claude/skills/visual-creative-subagent && rm skill.zip

Installs to .claude/skills/visual-creative-subagent

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Generate one isolated visual/creative concept pass as a self-contained browser-renderable showcase using HTML/CSS/JS with domain-appropriate libraries. Domains: data-vis (charts/graphs), animation (motion/physics), graphic-design (generative art/3D/illustration). Each pass is written from scratch by the AI agent, not stamped from a template.
343 charsno explicit “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
Advanced

Key capabilities

  • Generate self-contained HTML/CSS/JS showcases
  • Create data visualizations with specified mock data
  • Produce animations with motion and physics
  • Design generative art or 3D graphics
  • Ensure responsiveness for desktop and mobile
  • Implement interactive elements like tooltips or controls

How it works

This skill generates visual concepts from scratch using HTML/CSS/JS and specified CDN libraries, then validates the output through Playwright screenshots and visual review.

Inputs & outputs

You give it
Domain, style ID, style palette, style direction, library directive, mock data, scene description, anti-repeat list
You get back
Self-contained HTML showcase page with CSS/JS, screenshots, and validation report

When to use visual-creative-subagent

  • Generate a unique bar chart visualization
  • Create a particle system animation
  • Build a 3D globe visualization
  • Draft a generative art concept

About this skill

Visual Creative Subagent

Use this skill for one pass only. Do not blend with other pass outputs.

Inputs (via Task agent prompt)

  • domain - One of: data-vis, animation, graphic-design
  • styleId - Style within the domain (e.g., "bar-chart", "particle-systems", "globe-3d")
  • pass - Pass number
  • outputDir - Where to write generated files
  • stylePalette - Colors, fonts, design tokens
  • styleDirection - Creative brief text
  • libraryDirective - Which CDN library to use (from library-catalog.json)
  • mockData - Sample data (for data-vis passes)
  • sceneDescription - What to render (for animation/graphic-design passes)
  • antiRepeat - Explicit list of things NOT to repeat from prior passes

Hard Requirements

  1. Generate a self-contained HTML showcase page (not an app with navigation).
  2. The visualization/animation/graphic must render immediately on page load.
  3. Each pass must be visually distinct from every other pass in the same style.
  4. Use plain HTML/CSS/JS with the specified CDN library.
  5. Include responsive behavior for desktop and mobile.
  6. Write EVERY line of code from scratch — no shared templates.
  7. The showcase must be interactive where appropriate (tooltips, controls, hover effects).

Page Structure

Every pass produces a single showcase page with these regions:

Header Bar

  • Domain badge (e.g., "DATA VIS", "ANIMATION", "GRAPHIC DESIGN")
  • Style name and pass number
  • Library badge showing which library powers this pass

Main Showcase Area

  • Fills 80%+ of the viewport
  • The visualization, animation, or graphic renders here
  • Must be responsive — scales/adapts to viewport size

Controls Panel

  • Data-vis: Filter dropdowns, sort toggles, dataset switcher
  • Animation: Play/pause button, speed slider, reset button
  • Graphic-design: Regenerate/randomize button, parameter sliders

Info Footer

  • Technical details: library name, version, render method (Canvas/SVG/WebGL)
  • Data source description (for data-vis)
  • Brief technique description

Domain-Specific Guidelines

Data Visualization

  • Use the provided mockData faithfully — do not invent different data
  • Chart colors must use stylePalette, not library defaults
  • Include axis labels, legend, title, and tooltips
  • Show at least 2 views of the data (e.g., chart + summary table, or chart + sparklines)
  • Animated entrance for chart elements
  • Responsive: chart reflows on mobile, labels don't overflow

Animation

  • Animation must loop continuously or play indefinitely
  • Use requestAnimationFrame or library animation loops
  • Target 60fps — no jank or frame drops
  • Include visible play/pause controls
  • The animation fills the showcase area
  • For physics: add click/drag interactivity

Graphic Design

  • Render a complete composition — not a sketch or placeholder
  • For generative art: include a "Regenerate" button with new random seed
  • For 3D: include orbit controls or auto-rotation
  • For illustrations: render at high resolution
  • Display the random seed or generation parameters

Quality Standards

  1. No blank canvas: Visible content must render immediately on load
  2. Responsive: Works on desktop (1536px) and mobile (390px)
  3. Performance: Smooth rendering, no memory leaks, no console errors
  4. Polish: Professional presentation, consistent spacing, clean typography
  5. Working interactivity: All controls and hover states must function
  6. Library loaded: CDN script must load successfully (use valid URLs from catalog)
  7. Palette applied: Colors must match the provided palette, not library defaults

Files

  • index.html - Complete HTML with CDN library tags and showcase structure
  • style.css - Full CSS with responsive breakpoints
  • app.js - Library initialization, rendering, interaction logic
  • README.md - Concept overview, library usage, technique description
  • validation/handoff.json - Domain/style/library metadata

Mandatory Workflow: Plan → Generate → Validate → Fix Loop

The subagent itself (NOT the orchestrator) owns the full validation lifecycle. A pass is not complete until the subagent has validated and visually reviewed its own output.

Phase 1: Plan

  • Review all inputs (domain, style, palette, brief, library directive, anti-repeat)
  • Read the library-catalog.json for correct CDN URLs
  • Plan design approach

Phase 2: Pre-validate CDN URLs

Before writing HTML, verify every CDN URL resolves:

curl -s -o /dev/null -w "%{http_code}" "<CDN_URL>"
  • Non-200 → find alternative URL/version from catalog or CDN provider
  • Check note fields in library-catalog.json for version-specific issues (e.g., UMD vs ESM)

Phase 3: Generate

  • Write all files with verified CDN URLs only

Phase 4: Validate (Playwright)

The subagent runs the Playwright validation script (not the orchestrator):

node .claude/skills/visual-creative-subagent/scripts/validate-visuals-playwright.mjs --pass-dir <outputDir>

Phase 5: Review Screenshots

The subagent reads the generated PNG screenshots and visually assesses them. Failure conditions:

  • Blank/black canvas (library load failure or JS error)
  • Missing primary content (visualization/animation/graphic not visible)
  • Text unreadable (poor contrast, wrong font, overlap)
  • Layout broken (overflow, misalignment)
  • Controls missing
  • Mobile broken

Phase 6: Fix and Re-validate (Loop, max 3 cycles)

If failures detected → fix root cause → re-run Phase 4 → re-review Phase 5. After 3 cycles, document remaining issues in validation/handoff.json under unresolvedIssues.

Phase 7: Complete

Write final validation/handoff.json with:

  • validationPassed: boolean
  • fixCycles: number of fix iterations performed
  • cdnUrlsVerified: true
  • screenshotsReviewed: true

Required Screenshots Per Pass

  • validation/desktop/showcase.png — Full-page screenshot at 1536x960
  • validation/desktop/showcase-viewport.png — Viewport-only screenshot
  • validation/mobile/showcase.png — Full-page screenshot at 390x844 (2x scale)
  • validation/mobile/showcase-viewport.png — Viewport-only screenshot
  • validation/report.playwright.json — Structured report

Completion gate: A pass is NOT complete until screenshots exist on disk AND the subagent has visually reviewed them AND no failure conditions remain.

When not to use it

  • The task requires blending outputs from multiple passes.
  • The task requires an app with navigation.
  • The task involves using pre-stamped templates.

Limitations

  • The skill generates one isolated visual/creative concept pass only.
  • The skill does not provide an app with navigation.
  • The skill does not use shared templates.

How it compares

This workflow generates visually distinct, self-contained showcases from scratch for each pass, ensuring unique styles and immediate rendering, unlike template-based or app-integrated solutions.

Compared to similar skills

visual-creative-subagent side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
visual-creative-subagent (this skill)05moReviewAdvanced
threejs356moNo flagsAdvanced
scroll-experience1016moNo flagsIntermediate
interaction-design155moNo flagsIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

threejs

mrgoonie

Build 3D web apps with Three.js (WebGL/WebGPU). Use for 3D scenes, animations, custom shaders, PBR materials, VR/XR experiences, games, data visualizations, product configurators.

3554

scroll-experience

davila7

Expert in building immersive scroll-driven experiences - parallax storytelling, scroll animations, interactive narratives, and cinematic web experiences. Like NY Times interactives, Apple product pages, and award-winning web experiences. Makes websites feel like experiences, not just pages. Use when: scroll animation, parallax, scroll storytelling, interactive story, cinematic website.

101142

interaction-design

wshobson

Design and implement microinteractions, motion design, transitions, and user feedback patterns. Use when adding polish to UI interactions, implementing loading states, or creating delightful user experiences.

1550

motion

onmax

Use when adding animations with Motion Vue (motion-v) - provides motion component API, gesture animations, scroll-linked effects, layout transitions, and composables for Vue 3/Nuxt

1044

infographic-structure-creator

antvis

Generate or update infographic Structure components for this repo (TypeScript/TSX in src/designs/structures). Use when asked to design, implement, or modify structure layouts (list/compare/sequence/hierarchy/relation/geo/chart), including layout logic, component composition, and registration.

110

frontend-enhancer

ailabs-393

This skill should be used when enhancing the visual design and aesthetics of Next.js web applications. It provides modern UI components, design patterns, color palettes, animations, and layout templates. Use this skill for tasks like improving styling, creating responsive designs, implementing modern UI patterns, adding animations, selecting color schemes, or building aesthetically pleasing frontend interfaces.

35

Search skills

Search the agent skills registry