technical-image-generation-skill 用于制作有明确解释任务的技术插图。它定义了一套可复用的水彩信息图语言:米白纸纹、细墨线、低饱和蓝绿紫橙、编号卡片、清楚的箭头和克制的图例。
它不绑定任何产品、工作流或发布渠道。你可以把它用于文章、演示文稿、设计文档、产品说明、培训材料或技术复盘;前提是图片确实能让读者更快理解一个问题。
这张总览图把七种解释任务放在同一张图里,并用底部图例强调清晰文字、结构关系和统一视觉语言。
- 全景概览与心智模型 / ecosystem overviews and mental models
- 流程、生命周期与反馈回路 / processes, lifecycles, and feedback loops
- 系统架构与边界说明 / system architecture and boundaries
- 方案对比与取舍 / option comparison and trade-offs
- 单一机制下钻 / single-mechanism deep dives
- 抽象概念的技术解释 / technical explanation of abstract concepts
- 数据趋势与决策提示 / data trends and decision cues
-
阅读 SKILL.md,明确图片要回答的问题,并选择对应的
references/模板。 -
写出一句话结论,列出必须逐字保留的文字。
-
按模板填充内容,再用主 Skill 的 Prompt 框架组织成完整请求。
-
生成后依照 quality-checklist.md 检查文字、关系、版式与风格。
-
每轮只修一个最影响理解的问题;连续三轮仍无改善时,缩小范围或拆图。
-
Read SKILL.md, state the question the visual must answer, and choose the relevant
references/template. -
Write the one-sentence takeaway and list text that must remain verbatim.
-
Fill the selected template, then assemble a complete request using the core prompt frame.
-
After generation, check copy, relationships, layout, and style with quality-checklist.md.
-
Repair only the single issue that most harms comprehension per round; after three unsuccessful rounds, narrow or split the visual.
每个图型保留一张代表图。图片先以宫格展示,随后说明各图型的构图要点。
One representative image is kept for each illustration family. The grid appears first, followed by notes on each composition pattern.
一张合格图不是“内容很多的漂亮图片”。它应让读者在几秒内知道主题,在半分钟内走完整条主路径,并能在回看时准确找到关键边界、反馈或例外。文字、数字与连线出错时,宁可重做,也不要拿“手绘感”搪塞。







