Multimodal tool suite for DeepSeek Harness

Key features

• modlens turns image files into logged structured evidence for text-only models through local Codex, an OpenAI-compatible endpoint, or local Ollama.
• comfy-local inspects and runs local ComfyUI workflows through Comfy's first-party MCP and CLI tools.
• They can pass those files to ModLens when visual interpretation is needed.
• Native multimodal adapters remain in control of their own image input.