DeepSeek Harness Update: Better Multimodal Support for Visual Models

DeepSeek Harness got a big update on day 7, focusing on better multimodal support. Now you can turn on native image requests for DeepSeek models, even though DeepSeek v4 doesn't have vision yet. If you switch to a third-party model that supports vision, you can manually enable it. Also, commands like /goal and /plan now accept images, and you can reference past chats. So one question can include screenshots, files, and old conversations.