behavioral sandbox / any.run for agent skills

Submit an AI-agent skill. An isolated analyst container reads it and simulates what would happen if an agent followed it — then rates severity. Nothing executes on real systems.