Dynamic visual-response skill for opencode. Registers vision subagents from OpenCode's configured image-capable models and teaches a text-only orchestrator to extract visual intent, design a task-specific JSON response template, and delegate to a vision s