ad-asset-pipeline

Use when running the full ad asset pipeline — classify, center-crop, deduplicate, generate platform variants, and sync to Cloudinary. Orchestrates all 5 pipeline skills in sequence with progress reporting and resume capability.

Model	Source
sonnet	pack: ad-assets

Overview

Full ad asset pipeline orchestrator. Runs all 5 pipeline skills in sequence. Accepts a Cloudinary folder or local directory as input, processes every qualifying image through classification, smart cropping, deduplication, variant generation, and Cloudinary sync. Supports batch mode, per-asset error recovery, and resume from a prior run’s sync status.

Full Reference

ad-asset-pipeline

Mandatory Announcement — FIRST OUTPUT before anything else:

┏━ ⚡ ad-asset-pipeline ━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Pipeline: [source] → [N] assets → [platform(s)] ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛

Overview

Pipeline

[Cloudinary folder / local directory]
           ↓
  1. classify-ad-assets
     → AssetClassification[] (eligible assets only)
           ↓
  2. face-center-crop
     → Crop gravity assigned (g_face | g_auto)
           ↓
  3. deduplicate-assets
     → Deduplicated variant list (3-layer check)
           ↓
  4. generate-ad-variants
     → Variant manifest (all platform/placement URLs)
           ↓
  5. sync-to-cloudinary
     → sync-status.json + Cloudinary public IDs

Invocation

Full pipeline on a Cloudinary folder:

node pipeline.js --folder ads/source-images --platforms meta,pinterest,google

Resume a prior run (skip already-synced):

node pipeline.js --folder ads/source-images --resume

Single asset:

node pipeline.js --asset ads/source-images/photo-001 --platforms meta

Dry run (classify + report, no uploads):

node pipeline.js --folder ads/source-images --dry-run

Progress Reporting

Show a status table per asset, updated in real time:

photo-001   classify ✓  crop ✓  dedup ✓  variants ✓  sync ✓
photo-002   classify ✓  crop ✓  dedup ✓  variants ✓  sync ●
photo-003   classify ✓  crop ✓  dedup ✗  (duplicate of photo-001)
photo-004   classify ✗  (stage 1: dimension_too_small)
photo-005   classify ✓  crop ✓  dedup ✓  variants ✓  sync ✗  (failed)

[████████████░░░░] 3/5 complete

Error Handling

Individual asset failures do not halt the pipeline — the error is captured and processing continues
A summary of failures is reported at the end with actionable next steps
Retry failed assets by re-running with --resume — deduplicate-assets skips synced variants, sync-to-cloudinary retries failed ones

Final Summary

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Pipeline complete

photo-001   ✓  63 variants synced
photo-002   ✓  58 variants synced
photo-003   ✗  duplicate — skipped
photo-004   ✗  dimension_too_small
photo-005   ◐  2 variants failed

3 assets processed · 121 variants synced · 2 failed

▸ Re-run with --resume to retry failed variants

Pipeline Routing Table

User Says	Route To
”Classify this image / folder”	`classify-ad-assets`
”Crop for ads / face-center”	`face-center-crop`
”Remove duplicates”	`deduplicate-assets`
”Generate ad sizes / variants”	`generate-ad-variants`
”Upload to Cloudinary / sync”	`sync-to-cloudinary`
”Full pipeline / start to finish”	This skill — run all stages
”Resume / retry failed”	This skill — with `--resume`

Configuration

All pipeline config lives in a single pipeline.config.ts (or JSON) at project root. Each section maps to the corresponding skill’s config interface. See individual skill docs for per-stage config options.

export const pipelineConfig: PipelineConfig = {
  classify: {
    rules: { minWidth: 600, minHeight: 314, ... },
    gemini: { domainContext: 'Your domain here', ... },
    weights: DEFAULT_WEIGHTS,
    thresholds: DEFAULT_ELIGIBILITY,
  },
  variants: {
    cloudName: process.env.CLOUDINARY_CLOUD_NAME!,
    google: { allowBeforeAfter: false },
  },
  sync: {
    cloudName: process.env.CLOUDINARY_CLOUD_NAME!,
    concurrency: 5,
    maxRetries: 3,
    statusFilePath: './sync-status.json',
  },
  concurrency: 3,
  dryRun: false,
  resume: false,
  platforms: ['meta', 'pinterest', 'google'],
};