name: visio-drawing-orchestrator description: Analyze a source figure, classify the diagram family, decompose regions and elements, and produce a drawing plan before any detailed Visio drawing begins. Use when a job starts from an input image and the agent must decide what to draw first, which drawskills to invoke, which visioskills capabilities are needed, and how round-by-round critique should be structured.
Visio Drawing Orchestrator
这个 skill 是总控层。
它负责先想清楚“这次要怎么画”，再把工作交给 drawskills 和 visioskills。
Input Contract
开始时优先读取：
../../../../Setup/draw-job.local.json../../../../Setup/jobs/<job_name>/run-summary.json- 输入图片
../../../../docs/architecture/LAYER_CONTRACTS.md../../../../docs/architecture/ARTIFACT_CONTRACTS.md- 相关 lessons
Output Contract
这个 skill 至少要产出两份东西：
analysis.jsonplan.json
如果还没产出这两份结构化结果，不要直接进入复杂绘图。
Responsibilities
- 识别图类型
- pipeline
- network architecture
- module-detail side panel
- mixed-layout figure
- 拆分区域
- 主画布
- 侧边模块说明
- 标题/legend/IO 区
- 抽取元素
- box
- text
- arrow
- image
- annotation
- 判断风格
- 字体
- 字号层级
- 线条类型
- 颜色语义
- 形成绘图策略
- 先画什么
- 后画什么
- 哪些 drawskills 负责布局
- 哪些 visioskills 能力是必要的
Planning Rule
planner 只做“策略和结构”的决定，不做这些事：
- 不直接硬编码所有坐标
- 不直接写 Visio 请求序列
- 不把本轮失败经验直接塞进长期规则
Round Loop
每一轮后都要回答：
- 本轮偏差是识别问题、布局问题，还是执行问题？
- 下一轮应该改 planner、drawskills，还是 visioskills？
- 是否需要新增一条 lesson？
Read These Files As Needed
../../../../docs/architecture/LAYER_CONTRACTS.md../../../../Setup/analysis.schema.json../../../../Setup/plan.schema.json../../drawskills/visio-figure-builder/SKILL.md../../learningskills/visio-iteration-journal/SKILL.mdreferences/planning-checklist.md