Prompt Professionalizer

Prompt Professionalizer Skill

Use this skill to turn casual, vague, incomplete, or tool-specific generation requests into professional prompts that downstream tools can understand. This skill prepares prompts only; it does not directly invoke image generation, video generation, architecture diagram generation, data visualization, chart generation, or file generation tools.

Preflight Requirement

Use prompt-professionalizer before calling professional generation tools when the user request is oral, vague, incomplete, or not already shaped for the target tool. This preflight applies to image-generator, architecture-diagram-generator, chart-generator, animation/video generation tools, data visualization tools, and similar professional generators.

Skip this preflight only when the user already provides a complete structured prompt or complete tool payload with the required target type, content, style or rendering choices, and key parameters for the downstream tool.

Scope

Use this skill for:

• Image-generation prompts for tools such as Midjourney, Stable Diffusion, DALL-E, 即梦, 可灵, 通义万相, or similar tools.
• Animation or video-generation prompts for tools such as 可灵, Runway, Pika, 即梦视频, Sora-style tools, or similar tools.
• Technical architecture diagram prompts for Mermaid, PlantUML, Draw.io, Excalidraw, C4 diagrams, deployment diagrams, sequence diagrams, flowcharts, and ER diagrams.
• Data visualization prompts for ECharts, Tableau, Power BI, Datawrapper, chart generators, dashboards, and BI tools.
• General prompt optimization when the user asks to make a prompt more professional, precise, structured, controllable, or suitable for a specified tool.

Do not use this skill to:

• Directly generate images, videos, charts, architecture diagrams, dashboards, files, or code.
• Invent business facts, technical modules, data fields, metrics, dimensions, brands, people, or constraints not provided by the user.
• Change the user's original intent or silently add a new goal.
• Produce prompts for illegal, infringing, pornographic, graphic violent, deceptive identity, or clearly harmful outputs.

Workflow

1. Identify the target prompt type: image generation, animation/video generation, architecture diagram, data visualization, general prompt optimization, or unknown.
2. Check whether key information is missing for that target type.
3. If critical information is missing and the user did not say "do not ask", call request_user_input before producing the final professional prompt.
4. Read exactly one reference for the selected prompt type.
5. Combine the user's original request, any provided target tool, and any clarified answers into a professional prompt.
6. Output a structured result with 专业化 Prompt, 参数建议, optional 负面 Prompt, and 说明.
7. If the user explicitly asks to run a downstream generation tool after prompt preparation, route the prepared prompt to the corresponding skill or tool only when that tool is available and the request is explicit.
8. When another professional generation skill invokes this skill as preflight, return the optimized prompt and structured parameters that the downstream skill should use; do not produce the final artifact inside this skill.

Intent Detection

• Image generation: user asks for image, poster, cover, illustration, product scene, visual, logo draft, banner, 生图, 图片生成, 画面, 海报, 插画.
• Animation/video generation: user asks for video, animation, shot, camera motion, scene transition, 可灵, Runway, Pika, Sora, 视频, 动画, 镜头.
• Architecture diagram: user asks for architecture, system diagram, flowchart, deployment, sequence diagram, ER diagram, Mermaid, PlantUML, 架构图, 技术架构图, 流程图, 时序图.
• Data visualization: user asks for chart, dashboard, visualization, ECharts, Tableau, Power BI, 图表, 可视化, 数据看板.
• General optimization: user asks to optimize, professionalize, rewrite, improve, structure, adapt to a tool, 提示词优化, 更专业.

If the target type cannot be determined, call request_user_input.

Clarification Workflow

Call request_user_input instead of writing a plain clarification when missing information would materially reduce prompt quality. Ask at most 3 questions in one turn. Options must be concrete answers that can be used directly. A single question may contain at most 5 options.

Must ask when:

• The target prompt type or target tool is unknown.
• Image prompt lacks a subject or style.
• Video prompt lacks action or duration.
• Architecture diagram prompt lacks system modules or diagram type.
• Data visualization prompt lacks data fields or analysis goal.
• The user only says something like "make this more professional" without target tool or use case.
• The request is too short, such as "generate a tech image".

Do not ask when:

• The user explicitly says not to ask and to optimize directly.
• The prompt is already complete enough.
• The missing detail can be filled with a reasonable default and will not materially change the result.
• The user only wants a reusable prompt template.

When using defaults, label them under 默认假设.

Example request_user_input payload:

{
  "message": "我可以把需求整理成专业提示词，但需要先确认关键参数。",
  "questions": [
    {
      "id": "target_type",
      "question": "这个提示词主要给哪类工具使用？",
      "options": [
        { "label": "生图工具", "description": "用于图片、海报、插画、封面生成。" },
        { "label": "视频工具", "description": "用于动画、视频、镜头生成。" },
        { "label": "架构图工具", "description": "用于 Mermaid、PlantUML、Draw.io 等。" },
        { "label": "数据可视化", "description": "用于图表、看板、BI 工具。" },
        { "label": "通用优化", "description": "只提升表达、结构和可控性。" }
      ]
    },
    {
      "id": "primary_goal",
      "question": "最重要的生成目标是什么？"
    },
    {
      "id": "output_language",
      "question": "希望输出哪种语言的提示词？",
      "options": [
        { "label": "中文", "description": "输出中文专业提示词。" },
        { "label": "英文", "description": "输出英文专业提示词。" },
        { "label": "中英双版", "description": "同时输出中文和英文版本。" }
      ]
    }
  ]
}

After calling request_user_input, stop the turn and wait for the user's answer.

Output Rules

Default Chinese output:

## 专业化 Prompt

...

## 参数建议

- 工具类型：
- 风格：
- 画幅/尺寸：
- 输出形式：

## 负面 Prompt（可选）

...

## 说明

...

For architecture diagrams, prefer:

## 专业化 Prompt

请生成一张【图类型】，用于展示...

## 图表结构

- 用户层：
- 应用层：
- 服务层：
- 数据层：
- 外部系统：

## 关系说明

...

## 推荐图类型

Mermaid flowchart / C4 架构图 / 部署图

For data visualization, prefer:

## 专业化 Prompt

请基于以下数据生成一张...

## 图表配置建议

- 图表类型：
- X 轴：
- Y 轴：
- 分组字段：
- 筛选条件：
- 排序方式：
- 标题：

Language Rules

• Match the user's requested output language. If the user writes in Chinese or asks in Chinese, output all visible headings, field names, labels, and content in Chinese.
• If the user explicitly asks for English, output all visible headings, field names, labels, and content in English.
• If the user asks for both Chinese and English, keep each version internally consistent.
• Do not mix Chinese and English section headings in a single-language output.
• Keep target tool names, model names, product names, system names, code identifiers, data field names, and quoted source terms in their original language when translation would change meaning.

References

Read exactly one reference after identifying the target prompt type:

If the request combines multiple prompt types, read the reference for the primary downstream tool and include secondary guidance only when it does not conflict.

Constraints

• Do not expose internal reasoning or mention this skill.
• Do not directly call downstream generation tools unless the user explicitly asks after the prompt is prepared and the target tool is available.
• Before professional generation tools, use this skill as the required preflight when the request is vague, casual, incomplete, or not already structured for the target tool.
• Do not invent facts, data fields, metrics, dimensions, technical modules, deployment layers, user roles, actions, camera movements, or tool parameters not provided by the user.
• If defaults are used, mark them under 默认假设.
• Do not create prompts for illegal, infringing, pornographic, graphic violent, deceptive identity, or clearly harmful outputs.
• Do not simply polish wording; convert the request into a structured, professional prompt that a downstream tool can use.
• If the target tool has a known special format requirement from the user's request, adapt the prompt to that format.