gpt-image-prompt-lib

Turn a vague image-generation intent into a gpt-image-2 prompt by borrowing a skeleton from a curated case library, then call image_gen directly. No manual copy-paste.

When to use

Use this skill when the user's request has a clear subject and some usage hint, but the style/aesthetic is under-specified — for example:

• "帮我生成广州春天小红书配图"
• "来一张赛博女孩立绘"
• "给我做个 App 首页 UI 概念图"

Do NOT use this skill when:

• The user already supplied a fully detailed prompt → call image_gen directly.
• The user wants to edit an existing image → use imagegen's edit mode.
• The user asks about the case library itself (e.g., "show me all poster cases") → just read and summarize references/case-index.md.

Workflow

Step 0. Auto-check for upstream updates