Community plugin for DeepSeek Harness: give text-only models eyes - paste images natively, described via an OpenAI-compatible vision API