Skip to content
Back to search

CCF-Figure

CCF-Figure —— 为 AI / 计算机科学论文自动生成顶会级科研配图的智能助手。

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/deepshare-official--ccf-figure/badge.svg)](/listings/deepshare-official--ccf-figure)

HTML

<a href="/listings/deepshare-official--ccf-figure"><img src="/api/listings/deepshare-official--ccf-figure/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Claude Code

# Option 1: Install as a user-level Claude Code skill, available across projects
git clone https://github.com/Deepshare-Official/CCF-Figure \
  ~/.claude/skills/ccf-figure

# Option 2: Install as a project-level Claude Code skill, available only in the current project
git clone https://github.com/Deepshare-Official/CCF-Figure \
  .claude/skills/ccf-figure

Codex

# Option 1: Install as a user-level Codex skill, available across projects
git clone https://github.com/Deepshare-Official/CCF-Figure \
  ~/.agents/skills/ccf-figure

# Option 2: Install as a project-level Codex skill, available only in the current project
git clone https://github.com/Deepshare-Official/CCF-Figure \
  .agents/skills/ccf-figure

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 165 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: scientific-figure-generator
  3. version: "2.0"
  4. description: >
  5. Generate publication-ready scientific figures for AI/CS research papers.
  6. Automatically classifies paper type, selects the optimal diagram structure,
  7. and produces figures meeting the visual standards of NeurIPS, ICML, CVPR,
  8. Nature Machine Intelligence, and other top venues.
  9. platforms:
  10. - claude
  11. - codex
  12. author: Deepshare / Deepshare
  13. source: https://mp.weixin.qq.com/s/qPqfvBILEu5ascyZe6guAw
  14. ---
  15. # Scientific Figure Generator — Skill Instructions
  16. ## Role
  17. You are an expert scientific illustrator specializing in computer science and AI research.
  18. Your mission is **not** to simply beautify paper content, but to transform complex scientific
  19. text into logically accurate, structurally clear, information-dense, low-noise figures that
  20. meet the visual standards of top-tier conferences and journals.
  21. You must first understand the paper's research type, core problem, method mechanism,
  22. experimental logic, and key contributions — **then** decide on the most appropriate diagram
  23. structure. You are **prohibited** from mechanically applying the "left-input → center-model →
  24. right-output" three-column template unless the paper genuinely fits that structure.
  25. ---
  26. ## SOP — Standard Operating Procedure
  27. ### Step 1 · Input Validation
  28. Ask the user to provide one or more of the following (the more the better):
  29. - Paper title
  30. - Abstract
  31. - Method / approach section (key paragraphs)
  32. - Core contributions (bullet points)
  33. - Any specific visual requirement (e.g., "focus on the feedback loop", "show the three modules side-by-side")
  34. If the user provides **only an abstract**, warn them:
  35. > "Abstracts describe results, not structure. For a better figure, please also paste the method section or describe the key modules and their relationships."
  36. ---
  37. ### Step 2 · Paper Type Classification
  38. Classify the paper into **one primary type** (and optionally a secondary type):
  39. | ID | Type | Trigger Keywords / Signals |
  40. |----|------|---------------------------|
  41. | A | **Method Paper** | proposes a new model/framework/algorithm/loss/training strategy |
  42. | B | **Mechanism / Analysis Paper** | ablation study, module analysis, probing, attention visualization |
  43. | C | **Benchmark / Evaluation Paper** | new dataset, evaluation suite, leaderboard, task taxonomy |
  44. | D | **Scaling Law / Empirical Analysis** | scaling curves, compute budget, performance trend, law fitting |
  45. | E | **Robot / Embodied AI Paper** | perception-action loop, manipulation, navigation, sim-to-real |
  46. | F | **Interdisciplinary / AI for X** | medical AI, AI for science, smart manufacturing, cybersecurity |
  47. | G | **Survey / Position Paper** | literature taxonomy, research roadmap, comparison matrix |
  48. > See `references/figure-types.md` for the full type × structure mapping table.
  49. ---
  50. ### Step 3 · Diagram Structure Selection
  51. Based on the paper type, **automatically select** the most appropriate diagram structure:
  52. | Paper Type | Recommended Structure(s) |
  53. |-----------|--------------------------|
  54. | A — Method | Method overview, Model architecture, Layered pipeline |
  55. | B — Mechanism | Local zoom-in diagram, Comparison diagram, Ablation matrix |
  56. | C — Benchmark | Evaluation flow, Multi-panel comparison, Data construction pipeline |
  57. | D — Scaling | Experimental matrix, Trend curves panel, Variable-relationship diagram |
  58. | E — Robot/Embodied | **Closed-loop feedback diagram** (perception → language → planning → action → environment) |
  59. | F — Interdisciplinary | Cross-domain application framework (domain data + AI model + task + metrics) |
  60. | G — Survey | Taxonomy tree, Timeline, Comparison matrix |
  61. ---
  62. ### Step 4 · Generate the Full Image Prompt
  63. Assemble the final prompt by combining:
  64. 1. The **base visual specification** (from `references/prompt-template.md`, Section A)
  65. 2. The **paper-type-specific instruction** (from `references/prompt-template.md`, Section B)
  66. 3. The **user's paper content** (pasted verbatim at the end)
  67. Then pass this assembled prompt to the image generation model.
  68. **For Claude**: Use the built-in image generation capability directly.
  69. **For Codex**: Output the assembled prompt and instruct the user to paste it into
  70. ChatGPT (GPT Image 2 / DALL·E 3) or another image generation API.
  71. ---
  72. ### Step 5 · Self-Check Before Delivery
  73. Before presenting the result, verify against the **Five Common Failure Modes**
  74. (see `references/figure-types.md`, Section 3):
  75. - [ ] Did I use only the abstract? (if yes → request method details)
  76. - [ ] Did I request vague style words like "high-end" or "sci-fi"? (if yes → replace with specific venue style)
  77. - [ ] Did I tell the model the paper type? (if no → re-run Step 2)
  78. - [ ] Is the information density excessive? (if yes → trim to one clear visual storyline)
  79. - [ ] Is there excessive decoration? (if yes → enforce flat vector rules)
  80. ---
  81. ### Step 6 · Iteration Protocol
  82. If the user requests revisions:
  83. 1. Ask which specific element to change (structure, color, text, emphasis)
  84. 2. Diagnose whether the change is **structural** (re-run Step 3) or **cosmetic** (tweak prompt)
  85. 3. Generate a revised prompt and re-run generation
  86. 4. Maximum 3 revision rounds before suggesting a fundamentally different diagram type
  87. ---
  88. ## Quick-Start Template
  89. When a user wants to generate a figure immediately, respond with:
  90. ```
  91. Please provide the following for your paper:
  92. 1. **Title** (required)
  93. 2. **Abstract** (required)
  94. 3. **Method section** or key module descriptions (strongly recommended)
  95. 4. **Main contributions** (optional but helpful)
  96. 5. **Any specific visual preference** (optional — e.g., "emphasize the two-stage pipeline", "use a closed-loop structure")
  97. I will automatically classify your paper type, select the best diagram structure,
  98. and generate a top-conference-quality figure prompt for you.
  99. ```
  100. ---
  101. ## Visual Specification Summary
  102. | Property | Rule |
  103. |----------|------|
  104. | Background | Pure white (`#FFFFFF`) or very light gray (`#F5F5F5`) |
  105. | Style | 2D flat vector — NO 3D perspective, NO neon gradients, NO decorative textures |
  106. | Color usage | Functional only (encode logic hierarchy, not decoration) |
  107. | Color palette | Max 3 color groups; large areas = low-saturation cool gray / light blue / off-white |
  108. | Accent color | Sparse use of orange, cyan-green, or deep blue for key innovations |
  109. | Typography | Horizontal only; max 2 font sizes; Chinese for labels, English for proper nouns |
  110. | Symbols | Data/input = parallelogram or card stack; Model = rectangle; Decision = diamond; Feedback = closed-loop arrow |
  111. | Target venues | NeurIPS, ICML, ICLR, CVPR, ICCV, ACL, KDD, SIGIR, AAAI, Nature MI, Science Robotics, IEEE TPAMI |
  112. ---
  113. ## File Structure
  114. ```
  115. scientific-figure-generator/
  116. ├── SKILL.md ← this file (main workflow)
  117. ├── references/
  118. │ ├── prompt-template.md ← full image generation prompt (copy-paste ready)
  119. │ └── figure-types.md ← 11 diagram types × paper type table + 5 failure modes
  120. └── README.md ← GitHub-ready bilingual documentation
  121. ```
README.md · README · 267 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # CCF Figure
  2. <p align="center">
  3. <img src="https://img.shields.io/badge/Platform-Claude%20%7C%20Codex-blue?style=flat-square" />
  4. <img src="https://img.shields.io/badge/Version-2.0-brightgreen?style=flat-square" />
  5. <img src="https://img.shields.io/badge/Source-DeepShare%20深度之眼-orange?style=flat-square" />
  6. <img src="https://img.shields.io/badge/License-MIT-lightgrey?style=flat-square" />
  7. </p>
  8. <p align="center">
  9. A skill for generating publication-ready scientific figures from AI/CS research papers.<br>
  10. Classify paper type first, then select the optimal diagram structure automatically.
  11. </p>
  12. ---
  13. <!-- Chinese version (default) -->
  14. ## 中文说明
  15. <details open>
  16. <summary>点击展开 / 收起</summary>
  17. ### 是什么
  18. `CCF Figure` 是一个专门为 AI / 计算机科学领域的研究人员设计,帮助你将论文内容转化为**顶会顶刊级科研配图**的工具。
  19. 核心理念来自「深度之眼」2026年6月11日发布的文章《科研绘图提示词第二弹》:
  20. **先判断论文类型,再选择最合适的图示结构**,而不是千篇一律地套用“左输入-中模型-右输出”三段式模板。
  21. ### 支持平台
  22. | 平台 | 使用方式 |
  23. |-----|---------|
  24. | **Claude Code** | 安装到 skills 目录后,在项目中自动调用 |
  25. | **Codex** | 安装到 skills 目录后,生成完整提示词并辅助调用图像模型 |
  26. ### 功能特性
  27. - **自动分类**:识别 7 种论文类型(方法 / 机制 / 评测 / 规律 / 机器人 / 交叉 / 综述)
  28. - **结构自选**:从 11 种图示结构中自动匹配最优布局
  29. - **完整提示词**:内置中英文双语提示词模板,可直接复制使用
  30. - **五大翻车预防**:内置 Self-Check 清单,避免 AI 绘图的常见错误
  31. - **迭代协议**:标准化修改流程,最多 3 轮迭代
  32. ### 支持的论文类型 × 图示结构
  33. | 论文类型 | 推荐图示结构 |
  34. |--------|------------|
  35. | A · 方法类 | 方法总览图 / 模型架构图 |
  36. | B · 机制 / 分析类 | 机制示意图(局部放大)/ 对比式方法图 |
  37. | C · 评测 / 基准类 | 评测流程图 / 数据构建流程图 |
  38. | D · 规律发现类 | 实验规律总结图 / 实验矩阵图 |
  39. | E · 机器人 / 具身智能 | **闭环反馈图**(感知 → 规划 → 执行 → 反馈) |
  40. | F · 交叉方向 | 跨学科应用框架图 |
  41. | G · 综述 / 位置 | 分类树 / 时间线 / 对比矩阵 |
  42. ### 视觉规范摘要
  43. | 属性 | 规则 |
  44. |-----|------|
  45. | 背景 | 纯白 `#FFFFFF` 或极浅灰 `#F5F5F5` |
  46. | 风格 | 二维扁平化矢量 — 禁止 3D 透视 / 霓虹渐变 / 装饰纹理 |
  47. | 配色 | 功能性配色;最多 3 个色彩群组;强调色限橙 / 青绿 / 深蓝 |
  48. | 文字 | 水平排布;最多 2 种字号;中文标签 + 英文专有名词 |
  49. | 目标刊会 | NeurIPS, ICML, ICLR, CVPR, ACL, Nature MI, IEEE TPAMI 等 |
  50. ### 安装方式
  51. #### Claude Code
  52. ```bash
  53. # 方式一:安装到 Claude Code 个人级 skills 目录(推荐,所有项目通用)
  54. git clone https://github.com/Deepshare-Official/CCF-Figure \
  55. ~/.claude/skills/ccf-figure
  56. # 方式二:安装到当前项目的 Claude Code skills 目录(仅当前项目可用)
  57. git clone https://github.com/Deepshare-Official/CCF-Figure \
  58. .claude/skills/ccf-figure
  59. ```
  60. #### Codex
  61. ```bash
  62. # 方式一:安装到 Codex 个人级 skills 目录(推荐,所有项目通用)
  63. git clone https://github.com/Deepshare-Official/CCF-Figure \
  64. ~/.agents/skills/ccf-figure
  65. # 方式二:安装到当前项目的 Codex skills 目录(仅当前项目可用)
  66. git clone https://github.com/Deepshare-Official/CCF-Figure \
  67. .agents/skills/ccf-figure
  68. ```
  69. ### 使用方式
  70. 安装完成后,在 Claude Code 或 Codex 中直接说:
  71. ```text
  72. 帮我为这篇论文生成科研配图:
  73. [发送你的论文原文,或粘贴论文标题 + 摘要 + 方法部分]
  74. ```
  75. `CCF Figure` 会自动:
  76. 1. 判断论文类型
  77. 2. 选择最合适的图示结构
  78. 3. 组装完整提示词
  79. 4. 生成图像或输出可直接用于图像模型的完整提示词
  80. ### 文件结构
  81. ```text
  82. CCF-Figure/
  83. ├── SKILL.md ← 主工作流(Skill 入口)
  84. ├── references/
  85. │ ├── prompt-template.md ← 完整提示词模板(中英双语,可直接复制)
  86. │ └── figure-types.md ← 11 种图示类型规范 + 五大翻车问题
  87. └── README.md ← 本文件
  88. ```
  89. ### 建议的 SKILL.md 头部配置
  90. 为了让 Claude Code 和 Codex 更稳定识别,建议 `SKILL.md` 开头使用如下 metadata:
  91. ```markdown
  92. ---
  93. name: ccf-figure
  94. description: Generate publication-ready scientific figures for AI and computer science papers. Use when the user provides a paper title, abstract, method section, or full paper and wants a top-conference-style research diagram.
  95. ---
  96. # CCF Figure
  97. ```
  98. </details>
  99. ---
  100. <!-- English version -->
  101. ## English Documentation
  102. <details>
  103. <summary>Click to expand / collapse</summary>
  104. ### What It Is
  105. `CCF Figure` is a Skill for researchers in AI and computer science.
  106. It transforms paper content into **publication-quality scientific figures** that meet the visual standards of top conferences and journals.
  107. The core methodology comes from DeepShare’s June 2026 article:
  108. **Classify the paper type first, then select the optimal diagram structure** — rather than mechanically applying a generic “left-input → center-model → right-output” template.
  109. ### Supported Platforms
  110. | Platform | How it works |
  111. |----------|-------------|
  112. | **Claude Code** | Install it into the skills directory, then invoke it naturally in your project |
  113. | **Codex** | Install it into the skills directory, then use it to assemble complete scientific-figure prompts |
  114. ### Features
  115. - **Auto-classification**: Identifies 7 paper types — method, mechanism, benchmark, scaling, robotics, interdisciplinary, and survey
  116. - **Structure auto-selection**: Automatically matches the best layout from 11 diagram structures
  117. - **Complete prompt library**: Bilingual Chinese / English prompt templates, ready to copy and use
  118. - **5-failure-mode prevention**: Built-in self-check checklist to avoid common AI figure mistakes
  119. - **Iteration protocol**: Standardized revision workflow, up to 3 rounds
  120. ### Paper Type × Diagram Structure
  121. | Paper Type | Recommended Diagram |
  122. |-----------|---------------------|
  123. | A · Method | Method Overview / Model Architecture |
  124. | B · Mechanism / Analysis | Mechanism Zoom-in / Contrastive Diagram |
  125. | C · Benchmark / Evaluation | Evaluation Pipeline / Data Construction Pipeline |
  126. | D · Scaling Law / Empirical | Trend Panel / Experimental Matrix |
  127. | E · Robot / Embodied AI | **Closed-Loop Feedback Diagram** |
  128. | F · Interdisciplinary / AI for X | Cross-domain Application Framework |
  129. | G · Survey / Position | Taxonomy Tree / Timeline / Comparison Matrix |
  130. ### Visual Specification
  131. | Property | Rule |
  132. |----------|------|
  133. | Background | Pure white `#FFFFFF` or very light gray `#F5F5F5` |
  134. | Style | 2D flat vector — NO 3D, NO neon gradients, NO decorative textures |
  135. | Color | Functional only; max 3 color groups; accent colors limited to orange, cyan-green, and deep blue |
  136. | Typography | Horizontal only; max 2 font sizes; Chinese labels + English proper nouns |
  137. | Target venues | NeurIPS, ICML, ICLR, CVPR, ACL, Nature MI, IEEE TPAMI, etc. |
  138. ### Installation
  139. #### Claude Code
  140. ```bash
  141. # Option 1: Install as a user-level Claude Code skill, available across projects
  142. git clone https://github.com/Deepshare-Official/CCF-Figure \
  143. ~/.claude/skills/ccf-figure
  144. # Option 2: Install as a project-level Claude Code skill, available only in the current project
  145. git clone https://github.com/Deepshare-Official/CCF-Figure \
  146. .claude/skills/ccf-figure
  147. ```
  148. #### Codex
  149. ```bash
  150. # Option 1: Install as a user-level Codex skill, available across projects
  151. git clone https://github.com/Deepshare-Official/CCF-Figure \
  152. ~/.agents/skills/ccf-figure
  153. # Option 2: Install as a project-level Codex skill, available only in the current project
  154. git clone https://github.com/Deepshare-Official/CCF-Figure \
  155. .agents/skills/ccf-figure
  156. ```
  157. ### Usage
  158. After installation, simply say in Claude Code or Codex:
  159. ```text
  160. Generate a scientific figure for this paper:
  161. [Paste your paper, or paste the paper title + abstract + method section]
  162. ```
  163. `CCF Figure` will automatically:
  164. 1. Classify the paper type
  165. 2. Select the most appropriate diagram structure
  166. 3. Assemble the complete image-generation prompt
  167. 4. Generate the figure or output a complete prompt ready for an image model
  168. ### File Structure
  169. ```text
  170. CCF-Figure/
  171. ├── SKILL.md ← Main workflow / Skill entry point
  172. ├── references/
  173. │ ├── prompt-template.md ← Full prompt template, bilingual and copy-paste ready
  174. │ └── figure-types.md ← 11 diagram type specs + 5 failure modes
  175. └── README.md ← This file
  176. ```
  177. ### Recommended SKILL.md Metadata
  178. For better compatibility with Claude Code and Codex, it is recommended to start `SKILL.md` with the following metadata:
  179. ```markdown
  180. ---
  181. name: ccf-figure
  182. description: Generate publication-ready scientific figures for AI and computer science papers. Use when the user provides a paper title, abstract, method section, or full paper and wants a top-conference-style research diagram.
  183. ---
  184. # CCF Figure
  185. ```
  186. </details>
  187. ---
  188. ## License
  189. MIT License — feel free to use, modify, and distribute with attribution.
  190. ---
  191. <p align="center">
  192. Made with ❤️ for the AI research community · 深度之眼 Deepshare
  193. </p>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.