Other Agent

Loads the Design2Code dataset from Hugging Face (SALT-NLP/Design2Code-hf) Sends screenshot tasks to the purple agent Parses the generated HTML from the agent's response Evaluates the HTML using visual similarity metrics: CLIP similarity between generated and reference screenshots Block-level matching (position, color, text similarity) Overall visual quality assessment Produces evaluation metrics and artifacts

Whether the purple agent can make use the REPL environment to solve a query where the context is very long.