Agentic-creator-os Visual Intelligence System (VIS)
Agentic visual asset management — scan, audit, and manage images across your site with AI-powered quality enforcement
install
source · Clone the upstream repo
git clone https://github.com/frankxai/agentic-creator-os
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/frankxai/agentic-creator-os "$T" && mkdir -p ~/.claude/skills && cp -r "$T/.claude/skills/vis" ~/.claude/skills/frankxai-agentic-creator-os-visual-intelligence-system-vis && rm -rf "$T"
manifest:
.claude/skills/vis/SKILL.mdsource content
Visual Intelligence System (VIS)
What VIS Does
VIS is the visual asset management layer for ACOS. It ensures every image across your site is:
- Tagged — auto-detected mood, theme, color palette, placement suitability
- Audited — placeholders, duplicates, orphans, and oversized images detected
- Mapped — every page linked to its images with coverage status
- Brand-aligned — validated against your brand visual DNA
Core Workflow
1. Always Query Registry First
Before generating ANY new image, query the visual registry:
# Find dark, atmospheric music images suitable for homepage jq '[.[] | select(.tags | contains(["music"])) | select(.mood == "atmospheric")]' data/visual-registry.json
If a suitable image exists, USE IT. Only generate new images when nothing fits.
2. Scan Images
Rebuild the registry after adding new images:
node scripts/scan-visual-registry.mjs # Full rebuild node scripts/scan-visual-registry.mjs --diff # Only new images node scripts/scan-visual-registry.mjs --report # Health summary
3. Audit Visual Health
Run automated quality checks:
node scripts/audit-visual-health.mjs # Full audit with score node scripts/audit-visual-health.mjs --json # JSON for n8n/Slack
Health Score Interpretation:
- 90-100: EXCELLENT — no action needed
- 70-89: GOOD — minor issues to address
- 50-69: NEEDS ATTENTION — several issues
- 0-49: CRITICAL — immediate action required
4. Check Page Coverage
Query the sitemap-image map:
# Find pages that need images jq '[.[] | select(.status == "needs-images")]' data/sitemap-image-map.json # Find pages using placeholders jq '[.[] | select(.status == "placeholder")]' data/sitemap-image-map.json
Key Data Files
| File | Purpose | Update Frequency |
|---|---|---|
| 408+ images tagged & indexed | After every image addition |
| Brand visual standards | Rarely (brand changes) |
| Page-to-image coverage | After page/image changes |
Brand Visual DNA
The brand DNA file (
data/brand-visual-dna.json) defines:
- Palette — primary background (#0a0a0b), accent colors with usage rules
- Image Standards — hero (16:9, min 1600px), OG (1200x630), thumbnails (16:10)
- Mood Profiles — atmospheric, branded, technical, artistic, cinematic
- Quality Council — 3-perspective review before any visual ships:
- Brand Guardian — Does it match visual DNA? Dark theme, accent harmony?
- Conversion Optimizer — Does it drive action? Mood-setting vs distracting?
- Accessibility Auditor — Sufficient contrast? Meaningful alt text?
Generation Rules
When generating new images:
- Use
with model_tier "pro" for heroesmcp__nanobanana__generate_image - Always use dark background (#0a0a0b) — images must blend with site theme
- Aspect ratio 16:9 for heroes, 1200x630 for OG images
- Run through council review before deploying
- Auto-register in visual-registry.json after generation
- Include negative_prompt: "text, watermark, logo, bright background, cartoon, flat design"
Common Issues & Fixes
| Issue | Detection | Fix |
|---|---|---|
| Placeholder SVG | in frontmatter | Generate unique hero image |
| Duplicate hero | Same image in header AND body | Remove body reference |
| Oversized image | >2MB in registry | Optimize with sharp or re-generate at lower resolution |
| Orphaned image | In registry but no code reference | Investigate — may be dynamically loaded |
| Missing alt text | Empty on | Add descriptive alt text |
Integration Points
- visual-creation skill — Generation pipeline with quality gates
- brand-rules skill — Brand Guardian enforcement
- design-system skill — Color domain mapping and token architecture
- nanobanana MCP — Image generation (Gemini 3 Pro/Flash)
- n8n — Automated weekly audits via webhook