Agent-native image workflow and optional prompt optimizer for `/bananahub` and generic agent image generation requests. Normalizes non-English prompts into English by default, generates or edits images across Gemini/Nano Banana, OpenAI GPT Image, chat-compatible routes, and Codex/host-native image tools when available, discovers or uses BananaHub templates, and captures successful multi-turn image iterations as reusable workflow templates. Explicit `/bananahub` requests execute the BananaHub workflow. Requests like "把刚才的调图流程沉淀成模板", "save this image workflow", or "capture workflow" should use the capture-workflow flow. Generic requests like "生成图片", "画一个", "create an image", or "edit this image" should load BananaHub as an optimization advisor: ask whether to optimize/template-route the prompt before generation, then continue with the user's chosen image channel.
2026-05-17