image-pipeline

When to Use

Image optimization — compress, convert WebP/AVIF, resize, describe. Single-file atomic edits, batch processing, metadata extraction. NOT: visual regression (visual-testing) · LLM image description of UI screens (vision-analyze) · page screenshot capture (visual-testing).

Pipeline Stages

Input → [Validate] → [Transform] → [Optimize] → [Verify] → Output

Stage	Tools	Purpose
Validate	file, identify (ImageMagick)	MIME, dimensions, corruption check
Transform	magick convert, cwebp, avifenc	Resize, format convert, color space
Optimize	mozjpeg, oxipng, svgo	Lossless/lossy compression
Verify	compare (SSIM/PSNR), identify	Quality gates, regression detection

Hard Rules

• Match format: WebP (lossy photo) / AVIF (high crunch) / PNG (lossless) / SVG (vector)
• Preserve aspect ratio: no upscale beyond source
• Strip EXIF metadata unless describe needs it
• Responsive width units only; no hardcoded w/h px
• Quality gate: SSIM/PSNR >= 0.95 vs source; reject on regression

Output

IMG-PIPELINE:<file> STATUS:<ok|warn|error> FORMAT:<fmt> SIZE:<before->after> RATIO:<pct> CHANGES:<n>

Anti-Patterns

Upscaling beyond source | hardpx width on responsive | metadata not stripped | lossy on SVG | no SSIM gate | missing alt-text

Anti-Rationalization

Rationalization	Red Flag	Verification
"WebP sin fallback"	Solo WebP sin <picture> srcset	<picture> AVIF→WebP→PNG + srcset + fallback verify
"Describir imágenes con inventado"	alt inventado sin metadata	Strip EXIF unless needed; describe desde metadata + SSIM≥0.95
"Batch sin atomic edits"	Batch sobrescribe sin verify	Single-file atomic + compare SSIM/PSNR ≥0.95 gate

Red Flags

• Doing work without checking output format → STOP, re-read skill
• Second occurrence of same rationalization → force RED zone

Verification

• Output matches skill ## Output contract + file:line citaton
• cross-ref-check.ps1 → SKILL.md OK

Cross-Refs: visual-testing | vision-analyze | web-quality-audit | performance

Verification

• Output: response matches the ## Output contract format exactly
• token_budget: total tokens within frontmatter token_budget
• frontmatter: name, description, triggers, token_budget present and stable
• cross-refs: each referenced skill exists
• anti-patterns: none of the listed anti-patterns reintroduced

---

See reference.md for extended details, examples, and detailed patterns.