media-classify

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)"]

media-classify

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.

Pipeline Position

Step 2 of 8 — runs after media-scan, feeds media-process and taste-score.