ZeroTokenClaw Face Recognition System
Instructions
- Capture face image
- Generate embedding
- Compare with stored embeddings
- Calculate similarity score
- Apply threshold validation
- Validate event context
- Register check-in
Constraints
- Must respect EventAIConfig
- Must prevent duplicate check-ins
- Must validate active event
Security
- Never trust client-side matching
- Always validate on server
- Protect embeddings
Performance
- Limit number of comparisons
- Use indexed embeddings if possible
Anti-Patterns
- Accepting low confidence matches
- Matching outside event scope
- Ignoring multiple faces
Example
Face detected → embedding → match → confidence 0.82 → approved.