broker-recognize

broker-recognize: Photo → Product Info

Extracts structured product information from user-uploaded photos for C2C resale listing purposes.

Prerequisite: ANTHROPIC_API_KEY

This skill uses the Anthropic API (Claude Vision) for image analysis. Your current LLM model may not support multimodal vision, so the skill calls the Vision API directly.

export ANTHROPIC_API_KEY="sk-ant-..."

If the key is missing, the skill will output an error instructing the user to set it.

Workflow

1. User provides 1-3 photos of the item
2. Analyze using vision: brand, product name, model, category, condition (A-E), material, color, size, features, flaws, estimated original retail
3. Present structured card to user for confirmation/correction
4. Save product info to conversation context for next skill in pipeline

Condition Grade Scale

Grade	Label	Criteria
A	Like New	No visible wear, original packaging if applicable
B	Excellent	Minor signs of use, no significant flaws
C	Good	Visible wear, minor flaws, fully functional
D	Fair	Noticeable wear/flaws, may need minor repair
E	For Parts	Significant damage, sold as-is

Key Rule

Be honest about flaws. Transparency is the foundation of C2C trust. A lost honest sale is better than a return from undisclosed flaws.

Script

scripts/recognize.py provides the structured prompt template and output schema.