메뉴
Classify untagged media against Arcanea canon using Gemini Vision
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
The Claw Fleet — 5 AI engines (Media, Forge, Herald, Scout, Scribe) with 33 skills, cross-claw events, and fleet orchestration for creative operations.
Visual web dashboard for managing AI agent pipelines. Inbox with approval cards, live agent monitoring, social scheduling, and publish pipeline.
Complete AI-powered media processing pipeline for creative world-building. Scans, classifies, processes, scores, and uploads media with fantasy/mythology-aware tagging.
Auto-generate platform-optimized image variants and AI-written captions for Instagram, LinkedIn, X, YouTube, and TikTok from any source image.
AI-powered aesthetic quality scoring for images. 5-dimension evaluation: Canon Alignment, Design Compliance, Emotional Impact, Technical Fit, Uniqueness. Scores 0-100 with automatic tier assignment.
Detect and reject duplicate media files by content hash
| name | media-classify |
| description | Classify untagged media against Arcanea canon using Gemini Vision |
| trigger | pipeline |
| depends_on | ["media-scan"] |
| sandbox | true |
| inputs | [{"name":"batch_size","type":"int","description":"Max images to classify per cycle","default":20},{"name":"model","type":"str","description":"Gemini model for vision classification","source":"config.yaml#arcanea_claw.classify.model"}] |
| outputs | [{"name":"classified_count","type":"int","description":"Number of assets classified this cycle"}] |
| dependencies | {"python":["google-generativeai","supabase","Pillow"]} |
| tables | ["asset_metadata (UPDATE guardian, element, gate, tags, content_type, status)"] |
Queries asset_metadata for rows with status='new' or guardian IS NULL,
sends each image to Gemini Vision with a canon-aware prompt, and updates the
row with the identified Guardian, Element, Gate, tags, and content type.
Processes at most 20 images per cycle to stay within API rate limits.
Step 2 of 8 — runs after media-scan, feeds media-process and taste-score.
