Scientific Diagram Content, Icons, and Light Morandi Layout

中文包说明

默认将用户的文字材料设计为一份可直接复制的详细英文科研图 Prompt，也支持从用户提供的图像、手绘草稿、论文或已有图表中提取内容。重点保留科学术语、公式与拓扑关系，并组织相对布局、图标语义、浅色莫兰迪配色及 Times New Roman 字体要求。本包不捆绑图像生成模型、第三方参考图片或矢量绘图软件；实际出图、编辑或导出 SVG/PPT/Visio 需要运行本 Skill 的 Agent 具备相应工具。仅在用户提供其有权使用的模板时参考该模板；未提供模板时采用中性的出版导向布局。

Goal

Turn papers, notes, screenshots, hand-drawn drafts, or existing figures into scientifically accurate diagram content. Supply the modules, labels, formulas, topology, relative layout, icon plan, and light Morandi color roles. Let the user's own template control its shape vocabulary, framing, icon-drawing treatment, connector treatment, and rendering finish. Keep Times New Roman, the light palette, icon-first communication, and topological clarity as defaults unless the user explicitly overrides them.

Scientific meaning and topology take priority over appearance.

Load only the needed resources

1. Read references/transcription-and-topology.md when extracting content or validating a source figure.
2. Read references/prompt-template.md when composing a diagram-generation prompt.
3. Read references/style-spec.md when assigning colors or describing relative content layout.
4. Inspect every user-supplied content image and template at original detail.
5. Reference a visual template only when the user supplies one they are authorized to use. No legacy reference-image template is bundled; without a supplied template, use the neutral publication-layout fallback.

Select the operating mode

• Prompt mode: Return one copy-ready detailed English prompt by default. Do not generate an image unless asked.
• Render mode: Generate or edit the figure, then validate content, topology, colors, and text.
• Audit mode: Compare a prompt or rendered figure against the content sources and report or fix drift.
• Vector mode: When editable SVG, PPT, or Visio output is requested, use a deterministic vector-authoring workflow.

If the user says 给我 prompt, 同样的, or supplies explanatory notes, use Prompt mode. If the user says 画, 生成, 重绘, or 出图, use Render mode.

Assign input roles

Identify inputs explicitly:

• CONTENT SOURCE: authoritative for scientific claims, modules, labels, formulas, and relationships.
• CONTENT TARGET: an existing diagram whose visible content and topology should be preserved.
• TEMPLATE REFERENCE: authoritative for composition character, shape vocabulary, framing, icon-drawing treatment, connector treatment, and rendering finish only. It does not override explicit user instructions or the skill defaults of Times New Roman, light Morandi colors, icon-first communication, and clear topology unless the user explicitly asks it to.
• SUPPORTING INPUT: optional source material to insert, compare, or cite.

Never copy scientific semantics from a template. Describe references by their content rather than image number so attachment order cannot invert their roles.

Workflow

1. Extract the scientific story

Determine:

• the figure's purpose and one-sentence message;
• inputs, transformations, learned modules, controls, losses, outputs, and validation;
• exact terminology, acronyms, formulas, and visible labels;
• what is observed, inferred, trained, frozen, retrieved, filtered, bypassed, or rejected.

Before synthesis, perform literal transcription before scientific interpretation. Inspect dense or handwritten regions at original resolution and zoomed detail, then create an exact-source ledger for official module names, operation names, acronyms, enumerated category lists, variables, and formulas. For each fragile item, record its source location, literal reading, and confidence. Treat OCR as a candidate reading only, never as authority. Treat every source enumeration as a closed set: verify its item count and order; do not substitute plausible synonyms, add categories, drop items, or summarize the list away. Preserve distinctions such as regularization versus normalization exactly. If a term cannot be read or translated with confidence, retain the visible source-language text or mark it uncertain instead of guessing from domain expectations. A prior explicit user correction overrides OCR and inferred terminology.

For design-from-notes tasks, compress prose into concise figure labels without adding mechanisms, equations, thresholds, or claims absent from the sources.

2. Reconstruct the topology

Write every important route explicitly. Record arrow direction, solid/dashed meaning, branch and merge points, bypasses, feedback, gates, memory access, and operation nodes. State the relationships most likely to drift as explicit invariants.

3. Propose relative content layout

Specify only what the template needs to place the content correctly:

• reading direction and panel order;
• relative panel widths or heights;
• upper/lower lanes, sidebars, insets, and comparison strips;
• which modules sit together, branch, merge, or bypass;
• where key outputs and benefit callouts belong.

Do not prescribe border radius, container shape, line weight, shadow, or decorative treatment when a template is supplied. The template owns those decisions. Always specify Times New Roman for visible text and formulas, and request icons wherever they can replace explanatory prose without losing scientific meaning.

If no template is supplied, recommend a neutral publication layout without presenting it as a fixed house style.

4. Plan icon-first visual encoding

Convert scientific nouns, data types, physical objects, states, and familiar operations into meaningful icons, miniature schematics, or compact scientific glyphs wherever doing so reduces text. Keep a short exact label beside or beneath an icon when the identity would otherwise be ambiguous. Preserve formulas, official module names, acronyms, closed category lists, and critical state labels as text; never replace them with an unlabeled decorative icon.

Use icons as semantic carriers, not decoration. Reuse the same icon for the same entity across panels. Prefer compact icon-plus-label units over sentences. Replace explanatory paragraphs with visual sequences, branches, comparison motifs, small plots, matrices, maps, camera/ray sketches, database stacks, masks, filters, or output previews when supported by the source. Do not invent a visual object that implies unsupported data, dimensionality, or computation.

5. Assign Morandi color roles

Use references/style-spec.md. Map colors by scientific meaning, not by panel number. Keep the same meaning in the same color across the figure. Use only light, low-saturation, gray-toned fills on a warm light background. Do not use dark backgrounds, dark panels, black fills, saturated blocks, or strong gradients. Reserve a restrained neutral gray only for text, formulas, and connectors so the figure remains readable.

Preserve a template palette exactly only when the user explicitly says it is mandatory. Otherwise translate its hue relationships into the default light Morandi palette.

6. Compose the prompt

Use references/prompt-template.md. In Prompt mode, return one detailed English prompt containing only:

1. input roles and template-content separation;
2. scientific content and exact labels;
3. relative content layout;
4. explicit topology;
5. icon-first visual encoding;
6. light Morandi semantic color mapping;
7. Times New Roman typography;
8. critical invariants;
9. text-fidelity behavior and integrated negative constraints.

Do not automatically add a Chinese prompt, short prompt, standalone negative prompt, or parameter guide. Provide those only when explicitly requested.

For a content and color layout only request, omit instructions about borders, cards, shadows, rendering engines, or stylistic motifs. Retain the skill defaults of Times New Roman, light Morandi colors, icon-first communication, and clear topology unless the user explicitly overrides them.

7. Protect text and math

• Use strict preservation for a clear existing figure.
• Use manual typesetting for dense or ambiguous formulas.
• Set all visible titles, labels, annotations, numerals, and mathematical text in Times New Roman. Do not mix in sans-serif display fonts.
• When a glyph is unresolved, state that the source is authoritative and reserve a clean label area rather than guessing.
• Never accept plausible pseudo-text or corrupted equations.

8. Render or audit when requested

Validate node by node:

• all required modules and labels are present;
• formulas and fragile notation are correct;
• every arrow, branch, merge, bypass, and solid/dashed distinction is preserved;
• semantic colors are consistent and sufficiently distinct;
• all module fills and background areas remain light, with no dark visual blocks;
• icons replace avoidable prose without removing required scientific labels;
• all visible text uses Times New Roman;
• template semantics have not leaked into the content;
• no content is clipped, overlapped, duplicated, or invented.

Before any Prompt-mode delivery, perform the same term audit against the sources: verify every official name, operation, closed-list item, variable, and formula word for word. Correct omissions and synonym drift before returning the prompt.

Non-negotiable rules

• Do not impose a fixed visual motif by default. Reference only a user-supplied template the user is authorized to use; no bundled image template is required.
• Do not override the user's template with new shape, framing, connector, or icon-rendering rules; Times New Roman, light Morandi colors, and icon-first content encoding remain defaults unless explicitly overridden.
• Do not copy labels, modules, formulas, or scientific logic from a template reference.
• Do not invent formulas, architectural details, metrics, thresholds, losses, or claims.
• Do not rename source operations or replace closed category lists with domain-plausible alternatives.
• Do not omit a closed source list merely because its items are difficult to translate; preserve or flag them.
• Preserve exact terminology and math unless the user explicitly requests translation or editing.
• Preserve all meaningful arrows, branches, junctions, bypasses, and control semantics.
• Use Morandi colors as semantic grouping, not decoration.
• Use only light Morandi fills and light canvas colors; never use a dark theme, dark background, black panel, or large dark color block.
• Use icons, miniature scientific schematics, and compact glyphs to replace nonessential explanatory prose, while retaining exact labels wherever an icon alone could be ambiguous.
• Use Times New Roman for every visible title, label, annotation, numeral, and formula.
• Keep the reading direction obvious, minimize connector crossings, and make every split, merge, bypass, rejection, and feedback route visually explicit.
• Do not claim a raster output is editable vector artwork.

Completion check

Before delivering, verify:

• content and template roles are unambiguous;
• the prompt focuses on content, topology, relative layout, and color roles;
• the template remains authoritative for shape vocabulary, framing, connector treatment, icon rendering, and finish;
• every critical invariant appears in the prompt;
• fills and backgrounds use a coherent light Morandi mapping with no dark blocks unless the user explicitly requested another palette;
• icons carry as much nonverbal meaning as possible without obscuring exact scientific terminology;
• all visible text is explicitly set to Times New Roman;
• topology is immediately readable and connectors do not create ambiguous relationships;
• the response contains only the deliverables requested.