Full threadWhitneyLand·Could this task be a nice benchmark for computer use models?Would interesting to see the success rate for Claude Cowork or Codex’s equivalent feature.View on HN