ccf-visual-composer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

CCF Visual Composer

Invocation Controls

调用控制

CCFA Handoff Mode: PARTIAL (Recommended). Follow

metadata.ccf_skill_controls.handoff_question_mode

and

../ccf-common/references/handoff-modes.md

Treat image generation and editable reconstruction as two separate stages. For method, architecture, system, pipeline, framework, and illustrative scientific diagrams, use GPT Image 2 as the default first-pass renderer unless the user explicitly says not to use it. A request to create such a diagram authorizes the default image-generation stage; do not add a redundant pre-generation confirmation. After every generated image, show the raster draft and ask whether the user wants it reconstructed as editable SVG, vector PDF, editable PPTX, or a combination. Do not start reconstruction until the user agrees. Use the pure-SVG-first route only when the user explicitly rejects GPT Image 2 or explicitly asks for pure SVG/code-first output.

Classify the destination before drawing:

paper mechanism figure

PPT/Poster graphic

, or

README/outreach infographic

. A paper architecture request must expose the computation graph and pass

references/paper-vs-presentation-diagrams.md

; do not use slide-style stage cards, hero headings, callout dashboards, or presentation prose merely because they look polished. Presentation grammar is allowed only for an explicitly presentation-facing artifact and must be labeled accordingly.

CCFA 交接模式：PARTIAL（推荐）。遵循

metadata.ccf_skill_controls.handoff_question_mode

以及

../ccf-common/references/handoff-modes.md

中的规则。

将图像生成和可编辑重构视为两个独立阶段。对于方法图、架构图、系统图、流程图、框架图以及说明性科学图表，默认使用GPT Image 2作为首轮渲染工具，除非用户明确表示不使用。创建此类图表的请求即默认授权图像生成阶段，无需额外添加生成前确认步骤。每次生成图像后，展示栅格草稿并询问用户是否需要将其重构为可编辑SVG、矢量PDF、可编辑PPTX或组合格式。需获得用户同意后方可开始重构。仅当用户明确拒绝使用GPT Image 2或明确要求纯SVG/代码优先输出时，才采用纯SVG优先路线。

绘制前需明确目标场景：

论文机制图

、

PPT/海报图形

或

README/推广信息图

。论文架构图请求必须呈现计算图，并符合

references/paper-vs-presentation-diagrams.md

中的规范；不得仅因样式精美就使用幻灯片式阶段卡片、醒目标题、标注仪表盘或演示文稿式文字。仅当明确面向演示场景的成果时，才可使用演示文稿语法，并需相应标注。

Core Rule

核心规则

Make every visual evidence-bearing, readable, and integrated with the manuscript. Start from a visual contract, not a template. Never invent data, numbers, statistics, baselines, sample sizes, architecture modules, data flows, training signals, labels, images, captions that imply unsupported results, or official venue rules.

Preserve strong user prompts. When an architecture prompt already specifies the scientific topology, components, labels, and visual intent, keep it verbatim and append only a compact content-fit refinement. Do not replace it with a longer house-style prompt. When the input is a manuscript or sparse notes instead, extract a factual diagram specification and build the prompt from that evidence. Treat prompt length as a cost and latency surface.

Use natural title case or sentence case for ordinary English displayed inside a figure, while preserving canonical uppercase for acronyms, initialisms, and standardized technical abbreviations. For example, use

Visual Composer

GPT Image 2

Structure QA

, and

Editable SVG/PDF/PPTX

, never

VISUAL COMPOSER

Gpt Image 2

Structure Qa

, or

Editable Svg/Pdf/Pptx

. Acronyms such as

CCF

AI

GPT

QA

SVG

PDF

PPTX

, and

PNG

remain uppercase. Do not use all caps for complete ordinary-language titles, module names, legends, axes, annotations, badges, or table headers merely for emphasis.

确保所有可视化内容具备证据支撑、可读性强，并与手稿内容融为一体。从视觉契约出发，而非模板。严禁编造数据、数字、统计信息、基线、样本量、架构模块、数据流、训练信号、标签、图像、暗示无依据结果的说明文字，或官方会议规则。

保留用户的明确提示。当架构提示已明确指定科学拓扑结构、组件、标签和视觉意图时，需严格保留原文，仅添加紧凑的内容适配优化。不得用冗长的内部风格提示替代原文。若输入为手稿或零散笔记，则从中提取事实性图表规范并基于此构建提示。需将提示长度视为成本和延迟因素。

图表内显示的普通英文采用自然标题大小写或句首大小写，同时保留首字母大写的缩写词、首字母缩略词和标准化技术缩写。例如，使用

Visual Composer

、

GPT Image 2

、

Structure QA

和

Editable SVG/PDF/PPTX

，而非

VISUAL COMPOSER

、

Gpt Image 2

、

Structure Qa

或

Editable Svg/Pdf/Pptx

。

CCF

、

AI

、

GPT

、

QA

、

SVG

、

PDF

、

PPTX

和

PNG

等缩写词需保持大写。不得仅为强调而将完整的普通英文标题、模块名称、图例、坐标轴、注释、徽章或表格标题全部大写。

Modes

模式

```
visual-contract
```
: define core claim, reviewer question, evidence layer, source data, panel/table map, caption role, and output constraints.
```
figure-design
```
: design multi-panel figures, chart families, image plates, schematics, legends, labels, color, and export specs from supplied evidence.
```
architecture-generation
```
: turn supplied paper/method content into a scientific diagram specification and tailored prompt; invoke GPT Image 2 as the default first-pass renderer and inspect the result.
```
pure-svg-generation
```
: generate diagram structure directly as SVG/code only when the user explicitly declines GPT Image 2 or explicitly requests a pure-SVG/code-first artifact; label the result and any evaluation based on it as pure SVG generation.
```
icon-system
```
: select one coherent public SVG family for common concepts and generate isolated custom icons only for method-specific concepts that the public family cannot express.
```
reference-layout-blueprint
```
: derive a content-fit wireframe from user-authorized or public scientific references without copying their exact composition, icons, or styling.
```
editable-reconstruction
```
: after the required post-generation confirmation, rebuild an approved architecture image as semantic SVG/PDF and/or an editable PPTX with native objects and isolated icon assets.
```
python-plotting
```
: write or adapt Python plotting code using bundled recipes, standard-library SVG output, analytical chart recipes, composite dashboards, or optional libraries available in the user's environment.
```
table-design
```
: design publication tables, numeric precision, grouping, ordering, notes, width strategy, and LaTeX table structure from supplied values.
```
layout-integration
```
: place figures/tables near first discussion, align captions/cross-references, choose single-column/full-width floats, and keep visuals connected to text.
```
render-qa
```
: compile or render when files exist; inspect clipping, overlap, float order, font, contrast, rasterization, and source-data traceability.

```
visual-contract
```
：定义核心论点、评审问题、证据层、源数据、面板/表格映射、说明文字作用及输出约束。
```
figure-design
```
：基于提供的证据设计多面板图表、图表系列、图像板、示意图、图例、标签、配色及导出规范。
```
architecture-generation
```
：将提供的论文/方法内容转换为科学图表规范和定制提示；默认调用GPT Image 2作为首轮渲染工具并检查结果。
```
pure-svg-generation
```
：仅当用户明确拒绝使用GPT Image 2或明确要求纯SVG/代码优先成果时，才直接以SVG/代码形式生成图表结构；需将结果及相关评估标记为纯SVG生成。
```
icon-system
```
：为通用概念选择一套连贯的公开SVG图标库，仅当公开图标库无法表达方法特定概念时，才生成独立的定制图标。
```
reference-layout-blueprint
```
：从用户授权或公开的科学参考资料中推导适配内容的线框图，但不得复制其精确构图、图标或样式。
```
editable-reconstruction
```
：在获得生成后的必要确认后，将已批准的架构图重构为语义化SVG/PDF和/或包含原生对象及独立图标资产的可编辑PPTX。
```
python-plotting
```
：使用内置模板、标准库SVG输出、分析图表模板、复合仪表盘或用户环境中可用的可选库编写或适配Python绘图代码。
```
table-design
```
：基于提供的数值设计出版级表格，包括数值精度、分组、排序、注释、宽度策略及LaTeX表格结构。
```
layout-integration
```
：将图表/表格放置在首次提及的位置附近，对齐说明文字/交叉引用，选择单栏/通栏浮动布局，确保可视化内容与文本关联。
```
render-qa
```
：当文件存在时进行编译或渲染；检查裁剪、重叠、浮动顺序、字体、对比度、栅格化及源数据可追溯性。

Workflow

工作流

Identify target venue/family, manuscript context, supplied data/results or method content, artifact type, output format, and whether the user wants creation, redesign, or QA.

Load

../ccf-common/references/task-modes.md

and

../ccf-common/references/privacy-and-evidence.md

when the task touches manuscript files, private results, or project artifacts.

If claims, evidence, source data, or result values are missing, mark the gap and hand off to
```
ccf-experiment-designer
```
; do not fill the gap by invention. If architecture topology, labels, or method content are missing, inspect the user-authorized project sources or ask a targeted question instead of inventing modules.
Load
```
references/visual-contract.md
```
and write the visual contract before changing layout or style.
Load
```
references/palette-and-accessibility.md
```
before choosing colors; prefer accessible scientific palettes and semantic consistency over decorative color.
Normalize visible English to natural title or sentence case before plotting, prompting, or reconstruction, while preserving canonical uppercase acronyms and initialisms. Reject both all-caps ordinary phrases and incorrectly lowercased acronyms during render QA.
Route ordinary quantitative/result figures to deterministic, reproducible plotting code. Load
```
references/python-plot-recipes.md
```
and use
```
resources/python/ccfa_plot_recipes.py
```
as a runnable starting point. Prefer analytical plot families when the evidence calls for them: pie/donut for composition, grouped bars for categorical comparisons, volcano plots for effect-size/significance screening, correlation heatmaps for relationship matrices, and composite dashboards for multi-view analysis. If a better plot grammar is needed, load
```
references/plot-inspiration-map.md
```
and invent a new evidence-bound chart without copying external code.
For a method, model, system, pipeline, framework, architecture diagram, or illustrative scientific schematic, load
```
references/architecture-diagram-generation.md
```
,
```
references/adaptive-architecture-style.md
```
,
```
references/reference-layout-blueprint.md
```
, and
```
references/paper-vs-presentation-diagrams.md
```
. Freeze the delivery context, build the semantic topology, and create a low-fidelity aligned blueprint before rendering. For paper figures, map inputs, representations, operators, branches, merges, and outputs explicitly; reject a blueprint that is only a sequence of explanatory cards. Apply the prompt-specificity ladder: preserve detailed user prompts and add at most a short refinement; use full content-derived prompt construction only for manuscripts, notes, or materially incomplete prompts. Invoke the verified GPT Image 2 capability by default and inspect the raster result. If the user explicitly opts out of GPT Image 2 or requests pure SVG/code-first output, use the deterministic pure-SVG route and label it clearly. If the selected backend cannot be verified as GPT Image 2, state the limitation; do not substitute an unknown image model or silently switch to pure SVG.
Load
```
references/icon-system.md
```
when icons are useful. Prefer one coherent, licensed public SVG family for common concepts. Generate a custom icon only for a method-specific concept that cannot be expressed clearly with public icons or native diagram primitives; generate it as a separate asset, remove its background, inspect its edges and semantics, and never accept noisy or invented internal detail merely because it looks creative.
After GPT Image 2 generation, inspect the actual raster image against the diagram specification and show it to the user. Then ask the mandatory editable-deliverable question from
```
references/architecture-diagram-generation.md
```
. On agreement, use the approved raster as the aesthetic and layout reference, load
```
references/editable-pptx.md
```
for PPTX requests, and reconstruct semantic groups, native shapes, connectors, live text, and independent icon assets. Preserve useful method-specific visual assets as separately movable, background-free elements instead of flattening the full figure. Derive PDF from vector source when requested. A full-slide raster, an embedded raster inside SVG, or an auto-traced bitmap alone does not satisfy editability.
Load
```
references/figure-table-layout.md
```
for multi-panel composition, LaTeX float/table choices, caption/cross-reference placement, and manuscript integration.
Load
```
references/render-qa.md
```
; when source files exist, compile/render and inspect the actual output. When only a spec is requested, include a QA checklist and no-fabrication status.
Hand off to
```
ccf-paper-writer
```
for prose rewrites or narrative placement text,
```
ccf-integrity-auditor
```
for number/claim consistency, and
```
ccf-submission-checker
```
for final venue/package compliance.

确定目标会议/类型、手稿上下文、提供的数据/结果或方法内容、成果类型、输出格式，以及用户需求是创建、重设计还是QA。

当任务涉及手稿文件、私有结果或项目成果时，加载

../ccf-common/references/task-modes.md

和

../ccf-common/references/privacy-and-evidence.md

。

若论点、证据、源数据或结果数值缺失，标记缺口并移交至
```
ccf-experiment-designer
```
；不得通过编造填补缺口。若架构拓扑结构、标签或方法内容缺失，检查用户授权的项目源文件或提出针对性问题，而非编造模块。
加载
```
references/visual-contract.md
```
并在更改布局或样式前编写视觉契约。
加载
```
references/palette-and-accessibility.md
```
后再选择配色；优先选择符合 accessibility 的科学调色板和语义一致性，而非装饰性色彩。
在绘图、生成提示或重构前，将可见英文标准化为自然标题大小写或句首大小写，同时保留规范的大写缩写词和首字母缩略词。在渲染QA过程中，拒绝全部大写的普通短语和小写错误的缩写词。
将常规定量/结果图表路由至确定性、可复现的绘图代码。加载
```
references/python-plot-recipes.md
```
并使用
```
resources/python/ccfa_plot_recipes.py
```
作为可运行的起点。当证据需要时，优先选择分析型图表类型：饼图/环形图用于展示构成，分组条形图用于分类比较，火山图用于效应量/显著性筛选，相关热图用于关系矩阵，复合仪表盘用于多视图分析。若需要更合适的图表语法，加载
```
references/plot-inspiration-map.md
```
并基于证据创建新图表，不得复制外部代码。
对于方法图、模型图、系统图、流程图、框架图、架构图或说明性科学示意图，加载
```
references/architecture-diagram-generation.md
```
、
```
references/adaptive-architecture-style.md
```
、
```
references/reference-layout-blueprint.md
```
和
```
references/paper-vs-presentation-diagrams.md
```
。确定交付场景，构建语义拓扑结构，在渲染前创建低保真对齐线框图。对于论文图表，需明确映射输入、表示、操作符、分支、合并和输出；拒绝仅为解释性卡片序列的线框图。遵循提示特异性层级：保留详细的用户提示，最多添加简短优化；仅对手稿、笔记或严重不完整的提示，才基于内容构建完整提示。默认调用已验证的GPT Image 2能力并检查栅格结果。若用户明确选择不使用GPT Image 2或要求纯SVG/代码优先输出，则采用确定性纯SVG路线并明确标记。若所选后端无法验证为GPT Image 2，需说明限制；不得替换为未知图像模型或静默切换至纯SVG。
当需要使用图标时，加载
```
references/icon-system.md
```
。优先选择一套连贯、有授权的公开SVG图标库。仅当公开图标或原生图元无法清晰表达方法特定概念时，才生成定制图标；将其作为独立资产生成，移除背景，检查边缘和语义，不得因看起来有创意就接受含噪或编造的内部细节。
GPT Image 2生成后，对照图表规范检查实际栅格图像并展示给用户。然后提出
```
references/architecture-diagram-generation.md
```
中规定的必选可交付格式问题。获得同意后，以已批准的栅格图像为美学和布局参考，若为PPTX请求则加载
```
references/editable-pptx.md
```
，重构语义组、原生形状、连接线、可编辑文本和独立图标资产。将有用的方法特定视觉资产保留为可单独移动、无背景的元素，而非将整个图表扁平化。若有需求，从矢量源导出PDF。仅全幻灯片栅格图、SVG中嵌入的栅格图或自动追踪的位图不符合可编辑要求。
加载
```
references/figure-table-layout.md
```
用于多面板构图、LaTeX浮动/表格选择、说明文字/交叉引用放置及手稿集成。
加载
```
references/render-qa.md
```
；当源文件存在时，编译/渲染并检查实际输出。若仅请求规范，则包含QA检查清单和无编造声明。
若需重写散文或叙述性文本，移交至
```
ccf-paper-writer
```
；若需检查数值/论点一致性，移交至
```
ccf-integrity-auditor
```
；若需最终会议/包合规性检查，移交至
```
ccf-submission-checker
```
。

Output Contract

输出契约

Return the requested artifact first. For a full visual-composition request, use this structure:

text

Mode:
Target venue / format:
Visual contract:
Panel or table map:
Plot recipe or code path:
Architecture prompt / generation status:
Icon and asset manifest:
Reference-layout provenance:
Editable SVG/PDF/PPTX status:
Palette and accessibility:
LaTeX / manuscript placement:
Caption and cross-reference plan:
Render QA ledger:
Missing evidence or data:
No-fabrication status:
Next CCFA owner:

优先返回请求的成果。对于完整的可视化构图请求，采用以下结构：

text

模式:
目标会议/格式:
视觉契约:
面板或表格映射:
绘图模板或代码路径:
架构提示/生成状态:
图标和资产清单:
参考布局来源:
可编辑SVG/PDF/PPTX状态:
调色板与 accessibility:
LaTeX/手稿放置:
说明文字与交叉引用计划:
渲染QA记录:
缺失的证据或数据:
无编造声明:
下一任CCFA负责人:

References

参考资料

```
references/visual-contract.md
```
: figure/table contract, evidence hierarchy, panel map, source-data traceability, and anti-loop state files.
```
references/palette-and-accessibility.md
```
: top-journal/scientific palettes, color-vision safety, print/grayscale checks, and semantic color rules.
```
references/python-plot-recipes.md
```
: bundled Python recipe library, chart-selection rules, and custom plot invention prompt.
```
references/plot-inspiration-map.md
```
: conceptual map from open-source visualization projects to CCFA-native plotting decisions.
```
references/architecture-diagram-generation.md
```
: architecture-content extraction, scientific prompt construction, GPT Image 2 default generation, explicit pure-SVG opt-out, post-generation editable-format question, semantic SVG/PDF/PPTX reconstruction, and architecture QA.
```
references/adaptive-architecture-style.md
```
: prompt-preservation ladder, compact augmentation budget, content-fit style selection, transferable scientific diagram principles, and anti-imitation checks.
```
references/icon-system.md
```
: public-icon selection, custom icon micro-specs, transparent asset extraction, consistency rules, licensing records, and icon QA.
```
references/reference-layout-blueprint.md
```
: reference selection, composition distillation, grid/alignment strategy, layout-first wireframes, and anti-imitation controls.
```
references/paper-vs-presentation-diagrams.md
```
: mandatory context classification and the paper-mechanism acceptance gate that prevents slide or poster layouts from being presented as conference-paper architecture figures.
```
references/editable-pptx.md
```
: PowerPoint editability levels, native-object reconstruction, separate icon assets, deterministic generation, packaging, and PPTX QA.
```
references/figure-table-layout.md
```
: multi-panel design, table design, LaTeX float placement, captions, cross-references, and manuscript integration.
```
references/render-qa.md
```
: render-visible QA checklist, escalation rules, and visual issue ledger.
```
resources/python/ccfa_plot_recipes.py
```
: runnable standard-library SVG plotting recipes for paper-ready data-analysis figures.

```
references/visual-contract.md
```
：图表/表格契约、证据层级、面板映射、源数据可追溯性及防循环状态文件。
```
references/palette-and-accessibility.md
```
：顶级期刊/科学调色板、色觉友好性、打印/灰度检查及语义色彩规则。
```
references/python-plot-recipes.md
```
：内置Python模板库、图表选择规则及定制图表生成提示。
```
references/plot-inspiration-map.md
```
：从开源可视化项目到CCFA原生绘图决策的概念映射。
```
references/architecture-diagram-generation.md
```
：架构内容提取、科学提示构建、GPT Image 2默认生成、明确纯SVG退出选项、生成后可编辑格式询问、语义化SVG/PDF/PPTX重构及架构QA。
```
references/adaptive-architecture-style.md
```
：提示保留层级、紧凑优化预算、内容适配样式选择、可迁移科学图表原则及防模仿检查。
```
references/icon-system.md
```
：公开图标选择、定制图标微规范、透明资产提取、一致性规则、授权记录及图标QA。
```
references/reference-layout-blueprint.md
```
：参考资料选择、构图提炼、网格/对齐策略、布局优先线框图及防模仿控制。
```
references/paper-vs-presentation-diagrams.md
```
：强制场景分类及论文机制图验收标准，防止将幻灯片或海报布局作为会议论文架构图提交。
```
references/editable-pptx.md
```
：PowerPoint可编辑级别、原生对象重构、独立图标资产、确定性生成、打包及PPTX QA。
```
references/figure-table-layout.md
```
：多面板设计、表格设计、LaTeX浮动布局、说明文字、交叉引用及手稿集成。
```
references/render-qa.md
```
：渲染可见QA检查清单、升级规则及视觉问题记录。
```
resources/python/ccfa_plot_recipes.py
```
：可运行的标准库SVG绘图模板，用于生成符合出版要求的数据分析图表。