Back to Details

skill-to-prompt

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Skill to Prompt Converter

技能转Prompt转换器

Converts a Claude skill (SKILL.md + references/) into ChatGPT Project format: one prompt instruction document (.docx, under 8,000 characters) and one knowledge file document (.docx).
将Claude技能（SKILL.md + references/目录）转换为ChatGPT Project格式：一份Prompt指令文档（.docx格式，字符数不超过8000）和一份知识文件文档（.docx格式）。

What This Produces

输出内容

For each conversion:
  1. Prompt Instructions (.docx) — The text that gets pasted into ChatGPT's "Instructions" field. Under 8,000 characters. Structured using the 9-component prompt formula.
  2. Knowledge File (1 .docx file) — Uploaded to the ChatGPT Project as reference material. Contains all detailed workflows, templates, frameworks, examples, and reference content consolidated into a single document.
All .docx files follow standard formatting: Arial font, 10pt normal text, H1=20pt, H2=16pt, H3=14pt, H4=12pt, 1.15 line spacing.
每次转换会生成：
  1. Prompt指令（.docx）—— 可粘贴至ChatGPT“Instructions”字段的文本，字符数不超过8000，采用九要素Prompt公式构建结构。
  2. 知识文件（1份.docx文件）—— 作为参考资料上传至ChatGPT Project，包含整合到单个文档中的所有详细工作流、模板、框架、示例及参考内容。
所有.docx文件遵循标准格式：Arial字体，正文字号10pt，一级标题20pt，二级标题16pt，三级标题14pt，四级标题12pt，行间距1.15。

Conversion Workflow

转换工作流

Step 1: Locate and Read Source Skill

步骤1：定位并读取源技能

If user provides a skill name (e.g., "copywriting"):
  • Look in 
    ~/.claude/skills/[skill-name]/SKILL.md
If user provides a path, use it directly.
Read everything:
  • SKILL.md
    (full content)
  • All files in 
    references/
    folder
  • Note any 
    scripts/
    or 
    assets/
    folders (these need special handling)
若用户提供技能名称（例如“copywriting”）：
  • 在
    ~/.claude/skills/[skill-name]/SKILL.md
    路径下查找
若用户提供路径，则直接使用该路径。
读取所有内容：
  • SKILL.md
    （完整内容）
  • references/
    目录下的所有文件
  • 记录所有
    scripts/
    或
    assets/
    目录（这些目录需要特殊处理）

Step 2: Analyze Skill Complexity

步骤2：分析技能复杂度

Count total source content to understand how much compression is needed for the single knowledge file:
ComplexityCriteriaCompression Needed
Simple<300 lines in SKILL.md, no reference filesMinimal — content fits easily in one knowledge file
Medium300-500 lines or 1-2 reference filesModerate — condense verbose sections, keep all frameworks and examples
Complex500+ lines or 3+ reference filesHeavy — prioritize frameworks, worked examples, and actionable content. Summarize repetitive sections
Regardless of complexity, always produce exactly 1 knowledge file.
统计源内容总量，以确定单个知识文件需要的压缩程度：
复杂度判断标准压缩需求
简单SKILL.md内容少于300行，无参考文件最低 — 内容可轻松容纳于单个知识文件
中等300-500行或1-2个参考文件中等 — 精简冗余章节，保留所有框架和示例
复杂超过500行或3个及以上参考文件重度 — 优先保留框架、实际示例和可执行内容，总结重复章节
无论复杂度如何，始终生成恰好1份知识文件。

Step 3: Map Content to the 9-Component Prompt Formula

步骤3：将内容映射到九要素Prompt公式

Read 
references/prompt-structure-guide.md
for the full mapping guide.
Extract from the Claude skill and map to ChatGPT instruction components:
Claude Skill SectionChatGPT Prompt Component
Frontmatter 
description
Role + Task summary
"When to Use" / trigger conditionsTask context
"Before Starting" / gather context stepsInputs/Variables
Core workflow sections (numbered steps)Task (condensed steps)
"Output Format" sectionFormatting Instructions
"Examples" sectionExample Output (1-2 max in instructions)
"Constraints" / "Rules" / "Never" statementsConstraints
Target audience mentionsAudience
Voice/tone guidanceStyle/Tone
Purpose / outcome descriptionsGoal
阅读
references/prompt-structure-guide.md
获取完整映射指南。
从Claude技能中提取内容并映射到ChatGPT指令组件：
Claude技能章节ChatGPT Prompt组件
前置元数据
description
角色 + 任务摘要
“适用场景”/触发条件任务上下文
“开始前准备”/收集上下文步骤输入/变量
核心工作流章节（编号步骤）任务（精简步骤）
“输出格式”章节格式说明
“示例”章节输出示例（指令中最多保留1-2个）
“约束”/“规则”/“禁止”声明约束条件
目标受众说明受众
语气风格指南风格/语气
目的/预期结果描述目标

Step 4: Split Content Between Instructions and Knowledge Files

步骤4：拆分指令与知识文件的内容

Apply the priority system to stay under 8,000 characters:
P1 — Always in the prompt instructions:
  • Role statement (who the AI is, what it specializes in)
  • Task overview (primary task + condensed workflow steps)
  • Goal (desired outcome)
  • Critical constraints (never/always rules)
  • Knowledge file references (what each file contains, when to consult it)
P2 — Include in instructions if character budget allows:
  • 1-2 key examples (keep short, move full examples to knowledge files)
  • Input variables (what the user must provide before starting)
  • Output formatting specs
  • Style/tone guidance
P3 — Always in knowledge files:
  • Full detailed workflow with multi-paragraph explanations
  • Complete example library
  • Templates and frameworks (from SKILL.md body and references/)
  • All reference material (from references/ folder)
  • Edge cases and advanced scenarios
  • Decision trees and detailed tables
  • Script code (from scripts/ folder, documented as reference)
Character budget: Target 7,500 characters for the instruction text (500-char safety buffer).
应用优先级系统以确保字符数不超过8000：
优先级1 — 必须包含在Prompt指令中：
  • 角色声明（AI的身份及专长）
  • 任务概述（主要任务+精简工作流步骤）
  • 目标（预期结果）
  • 关键约束（必须/禁止规则）
  • 知识文件引用（各文件包含内容及查阅时机）
优先级2 — 若字符预算允许则包含在指令中：
  • 1-2个关键示例（保持简短，完整示例移至知识文件）
  • 输入变量（用户开始前必须提供的内容）
  • 输出格式规范
  • 风格/语气指南
优先级3 — 必须包含在知识文件中：
  • 包含多段解释的完整详细工作流
  • 完整示例库
  • 模板和框架（来自SKILL.md正文及references/目录）
  • 所有参考资料（来自references/目录）
  • 边缘案例及高级场景
  • 决策树及详细表格
  • 脚本代码（来自scripts/目录，作为参考文档记录）
字符预算： 指令文本目标字符数为7500（预留500字符的安全缓冲）。

Step 5: Write Prompt Instructions

步骤5：编写Prompt指令

Draft the instruction text following this structure:
Every prompt instruction document must include all 9 headings below, in this exact order. No headings may be omitted or renamed.
undefined
按照以下结构编写指令文本：
每份Prompt指令文档必须包含以下所有9个标题，且顺序完全一致，不得省略或重命名标题。
undefined

Role

Role

You are [expertise extracted from skill description and intro]. You specialize in [specific domain/capabilities].
You are [expertise extracted from skill description and intro]. You specialize in [specific domain/capabilities].

Task

Task

Your primary task is to [main objective from the skill].
Follow this workflow:
  1. [Condensed step from skill workflow]
  2. [Condensed step]
  3. [Condensed step] ...
Your primary task is to [main objective from the skill].
Follow this workflow:
  1. [Condensed step from skill workflow]
  2. [Condensed step]
  3. [Condensed step] ...

Goal

Goal

[Desired outcome extracted from the skill's purpose/output sections]
[Desired outcome extracted from the skill's purpose/output sections]

Audience

Audience

[Target audience. If the skill does not specify an audience, write: "General users seeking [skill domain] assistance. Adjust complexity based on the user's apparent experience level."]
[Target audience. If the skill does not specify an audience, write: "General users seeking [skill domain] assistance. Adjust complexity based on the user's apparent experience level."]

Style/Tone

Style/Tone

[Voice/tone rules extracted from the skill. If the skill does not specify tone, write: "Professional and clear. Match the register of the target audience. Avoid jargon unless the audience expects it."]
Always include these writing rules regardless of skill-specific tone:
  • Avoid dashes unless absolutely necessary. Prefer commas, periods, or conjunctions to keep the tone natural.
  • Use en dashes (with spaces, e.g. word – word) or em dashes (without spaces, e.g. word—word) only when unavoidable, ideally no more than once per complete piece of output.
[Voice/tone rules extracted from the skill. If the skill does not specify tone, write: "Professional and clear. Match the register of the target audience. Avoid jargon unless the audience expects it."]
Always include these writing rules regardless of skill-specific tone:
  • Avoid dashes unless absolutely necessary. Prefer commas, periods, or conjunctions to keep the tone natural.
  • Use en dashes (with spaces, e.g. word – word) or em dashes (without spaces, e.g. word—word) only when unavoidable, ideally no more than once per complete piece of output.

Constraints

Constraints

  • [Never/always rule]
  • [Character/word limits]
  • [Format restrictions]
  • [Quality standards]
  • [Never/always rule]
  • [Character/word limits]
  • [Format restrictions]
  • [Quality standards]

Inputs/Placeholders

Inputs/Placeholders

Before starting, ask the user for:
  1. [Required input from the skill's context-gathering steps]
  2. [Required input]
Before starting, ask the user for:
  1. [Required input from the skill's context-gathering steps]
  2. [Required input]

Example Output

Example Output

[1-2 short examples showing expected output structure. Keep brief.]
[1-2 short examples showing expected output structure. Keep brief.]

Formatting Instructions

Formatting Instructions

[Output structure and formatting rules from the skill]

After the 9 required sections, always append a Knowledge Files section:
[Output structure and formatting rules from the skill]

在9个必填章节之后，必须添加一个“知识文件”章节：

Knowledge Files

Knowledge Files

This project includes 1 knowledge file. Consult it as needed:
  1. [SkillName]_Reference_Guide.docx — [What it contains and when to reference it]

After drafting, count characters. If over 7,500:
- Cut examples to bare minimum (but keep the Example Output heading with at least a brief example)
- Compress workflow steps further
- Shorten Audience and Style/Tone content (but never remove the headings)

If over 8,000 after compression: shorten section content further, but all 9 headings must remain. Move detailed explanations to the knowledge file and replace with concise summaries under each heading.
This project includes 1 knowledge file. Consult it as needed:
  1. [SkillName]_Reference_Guide.docx — [What it contains and when to reference it]

草稿完成后统计字符数。若超过7500：
- 将示例精简至最低限度（但需保留“输出示例”标题及至少一个简短示例）
- 进一步精简工作流步骤
- 缩短“受众”和“风格/语气”章节内容（但不得删除标题）

若压缩后仍超过8000：进一步缩短章节内容，但所有9个标题必须保留。将详细解释移至知识文件，在每个标题下替换为简洁摘要。

Step 6: Organize Knowledge File Content

步骤6：整理知识文件内容

Always produce exactly 1 knowledge file named 
[SkillName]_Reference_Guide.docx
.
Consolidate all P3 content into this single document, organized with clear heading hierarchy:
[SkillName]_Reference_Guide.docx
containing:
  • Full detailed workflow with complete explanations
  • All examples and worked examples
  • Templates and frameworks (from SKILL.md body and references/ folder)
  • All reference material (from references/ folder)
  • Guidelines, best practices, and edge cases
  • Decision trees and detailed tables
  • Script documentation (from scripts/ folder, noted as non-executable reference)
Organization within the single file:
  • Use H1 for major sections (Workflows, Templates, Examples, Reference Material)
  • Use H2-H3 for subsections within each major section
  • Use page breaks between major sections for readability
  • For complex skills with heavy reference content, prioritize frameworks, worked examples, and actionable content over verbose explanations
始终生成恰好1份名为
[SkillName]_Reference_Guide.docx
的知识文件。
将所有优先级3的内容整合到该文档中，采用清晰的标题层级结构：
[SkillName]_Reference_Guide.docx
包含：
  • 包含完整解释的详细工作流
  • 所有示例及实际操作案例
  • 模板和框架（来自SKILL.md正文及references/目录）
  • 所有参考资料（来自references/目录）
  • 指南、最佳实践及边缘案例
  • 决策树及详细表格
  • 脚本文档（来自scripts/目录，标注为不可执行参考）
单文件内的组织结构：
  • 一级标题用于主要章节（工作流、模板、示例、参考资料）
  • 二级至三级标题用于各主要章节下的子章节
  • 主要章节之间添加分页符以提升可读性
  • 对于参考内容较多的复杂技能，优先保留框架、实际示例和可执行内容，而非冗长解释

Step 7: Generate JSON Data

步骤7：生成JSON数据

Structure all content as a JSON object matching the docx-generator input format.
The JSON structure:
json
{
  "skillName": "Skill Display Name",
  "originalSkillName": "skill-name",
  "version": "1.0.0",
  "outputDir": "~/Downloads/",
  "instructionDoc": {
    "filename": "SkillName_Prompt_Instructions.docx",
    "title": "Skill Name — ChatGPT Project Instructions",
    "characterCount": 7234,
    "sections": [
      { "type": "heading", "level": 1, "text": "Section Title" },
      { "type": "paragraph", "text": "Regular paragraph text." },
      { "type": "bold-paragraph", "label": "Label:", "text": "Text after label" },
      { "type": "bullet", "text": "Bullet point", "level": 0 },
      { "type": "numbered", "text": "Numbered item", "number": 1 },
      { "type": "table", "headers": ["H1","H2"], "rows": [["a","b"]], "columnPcts": [30,70] },
      { "type": "divider" },
      { "type": "spacer", "size": 200 },
      { "type": "meta", "text": "Metadata or footnote text" }
    ]
  },
  "knowledgeFiles": [
    {
      "filename": "SkillName_Reference_Guide.docx",
      "title": "Skill Name Reference Guide",
      "sections": [...]
    }
  ]
}
Write this JSON to a temporary file in the current directory: 
[skill-name]-conversion.json
将所有内容整理为符合docx-generator输入格式的JSON对象。
JSON结构如下：
json
{
  "skillName": "Skill Display Name",
  "originalSkillName": "skill-name",
  "version": "1.0.0",
  "outputDir": "~/Downloads/",
  "instructionDoc": {
    "filename": "SkillName_Prompt_Instructions.docx",
    "title": "Skill Name — ChatGPT Project Instructions",
    "characterCount": 7234,
    "sections": [
      { "type": "heading", "level": 1, "text": "Section Title" },
      { "type": "paragraph", "text": "Regular paragraph text." },
      { "type": "bold-paragraph", "label": "Label:", "text": "Text after label" },
      { "type": "bullet", "text": "Bullet point", "level": 0 },
      { "type": "numbered", "text": "Numbered item", "number": 1 },
      { "type": "table", "headers": ["H1","H2"], "rows": [["a","b"]], "columnPcts": [30,70] },
      { "type": "divider" },
      { "type": "spacer", "size": 200 },
      { "type": "meta", "text": "Metadata or footnote text" }
    ]
  },
  "knowledgeFiles": [
    {
      "filename": "SkillName_Reference_Guide.docx",
      "title": "Skill Name Reference Guide",
      "sections": [...]
    }
  ]
}
将该JSON写入当前目录下的临时文件：
[skill-name]-conversion.json

Step 8: Generate .docx Files

步骤8：生成.docx文件

Run the generator:
bash
NODE_PATH=~/Documents/node_modules node ~/.claude/skills/skill-to-prompt/references/docx-generator.js [skill-name]-conversion.json
This produces all .docx files in the 
~/Downloads/
folder.
If the docx package is not available, install it first:
bash
cd ~/Documents && npm install docx && cd -
Fallback: If .docx generation fails entirely, output the content as formatted markdown that the user can paste into Google Docs or Word. Apply the same heading hierarchy and section structure.
运行生成器：
bash
NODE_PATH=~/Documents/node_modules node ~/.claude/skills/skill-to-prompt/references/docx-generator.js [skill-name]-conversion.json
所有.docx文件将生成在
~/Downloads/
目录下。
若docx包未安装，请先执行安装：
bash
cd ~/Documents && npm install docx && cd -
备选方案： 若.docx生成完全失败，则输出格式化的Markdown内容，用户可将其粘贴至Google Docs或Word中，采用相同的标题层级和章节结构。

Step 9: Validate and Present Results

步骤9：验证并展示结果

Check and report:
  • Instruction character count (must be under 8,000)
  • All .docx files generated successfully
  • Knowledge files contain complete referenced content
  • Cross-references between instructions and knowledge files are clear
Present to the user:
Conversion complete: [skill-name] → ChatGPT Project

Output saved to ~/Downloads/

Prompt Instructions:
  [SkillName]_Prompt_Instructions.docx (X,XXX characters)

Knowledge File:
  [SkillName]_Reference_Guide.docx — [brief description]

Setup in ChatGPT:
  1. Create a new Project in ChatGPT
  2. Paste the instruction text from the Instructions .docx into the project's Instructions field
  3. Upload the knowledge file to the project
  4. Test with a sample prompt
After presenting, clean up the temporary JSON file:
bash
rm [skill-name]-conversion.json
检查并报告：
  • 指令字符数（必须少于8000）
  • 所有.docx文件生成成功
  • 知识文件包含完整的参考内容
  • 指令与知识文件之间的交叉引用清晰明确
向用户展示：
Conversion complete: [skill-name] → ChatGPT Project

Output saved to ~/Downloads/

Prompt Instructions:
  [SkillName]_Prompt_Instructions.docx (X,XXX characters)

Knowledge File:
  [SkillName]_Reference_Guide.docx — [brief description]

Setup in ChatGPT:
  1. Create a new Project in ChatGPT
  2. Paste the instruction text from the Instructions .docx into the project's Instructions field
  3. Upload the knowledge file to the project
  4. Test with a sample prompt
展示完成后，清理临时JSON文件：
bash
rm [skill-name]-conversion.json

Conversion Rules

转换规则

What Transfers Directly

可直接迁移的内容

  • Role and purpose definitions
  • Workflow steps (condensed)
  • Constraints and rules
  • Output format specifications
  • Examples and templates
  • Reference content
  • 角色和目的定义
  • 工作流步骤（已精简）
  • 约束条件和规则
  • 输出格式规范
  • 示例和模板
  • 参考内容

What Requires Rewriting

需要重写的内容

  • Claude tool references (Read, Grep, Glob, Bash, Edit, Write) → Rewrite as natural language instructions. Example: "Read the user's file" becomes "Ask the user to paste or upload the content."
  • Dynamic commands (
    !command
    syntax) → Not available in ChatGPT. Document as manual steps.
  • File system operations → Replace with "ask the user to provide" patterns.
  • Progressive disclosure (loading references on demand) → All content must be in knowledge files upfront.
  • Claude工具引用（Read、Grep、Glob、Bash、Edit、Write）→ 改写为自然语言指令。示例：“Read the user's file”改为“请用户粘贴或上传内容”。
  • 动态命令（
    !command
    语法）→ ChatGPT不支持，记录为手动步骤。
  • 文件系统操作 → 替换为“请用户提供”的表述。
  • 渐进式披露（按需加载参考内容）→ 所有内容必须预先包含在知识文件中。

What Does Not Transfer

不可迁移的内容

  • Scripts (scripts/ folder) → Document the script's purpose in the knowledge file. Include the code as reference if useful, but note it cannot execute in ChatGPT.
  • Assets (assets/ folder) → Note in the output: "These assets must be manually transferred if needed."
  • Cross-skill references → Flag as dependencies: "This skill references [other-skill]. Consider converting that skill separately."
  • MCP server integrations → Not available in ChatGPT Projects.
  • Frontmatter fields (version, argument-hint, context, agent, allowed-tools) → These are Claude-specific metadata. Do not include in ChatGPT instructions.
  • 脚本（scripts/目录）→ 在知识文件中记录脚本用途。若有用可将代码作为参考包含，但需注明其无法在ChatGPT中执行。
  • 资源文件（assets/目录）→ 在输出中注明：“如需使用这些资源文件，必须手动迁移。”
  • 跨技能引用 → 标记为依赖项：“本技能引用了[other-skill]，请考虑单独转换该技能。”
  • MCP服务器集成 → ChatGPT Projects不支持。
  • 前置元数据字段（version、argument-hint、context、agent、allowed-tools）→ 这些是Claude专属元数据，请勿包含在ChatGPT指令中。

Output Naming Convention

输出命名规范

[SkillName]_Prompt_Instructions.docx
[SkillName]_Reference_Guide.docx
Where 
[SkillName]
is the skill name in Title Case with spaces replaced by underscores.
Examples:
  • copywriting
    → 
    Copywriting_Prompt_Instructions.docx
  • digital-product-video-ads
    → 
    Digital_Product_Video_Ads_Prompt_Instructions.docx
  • page-cro
    → 
    Page_CRO_Prompt_Instructions.docx
[SkillName]_Prompt_Instructions.docx
[SkillName]_Reference_Guide.docx
其中
[SkillName]
为标题格式的技能名称，空格替换为下划线。
示例：
  • copywriting
    → 
    Copywriting_Prompt_Instructions.docx
  • digital-product-video-ads
    → 
    Digital_Product_Video_Ads_Prompt_Instructions.docx
  • page-cro
    → 
    Page_CRO_Prompt_Instructions.docx

Document Formatting Standards

文档格式标准

All .docx output follows these standards:
ElementSetting
FontArial
Font colorBlack (default) for all text and headings
Normal text10pt
Heading 120pt
Heading 216pt
Heading 314pt
Heading 412pt
Line spacing1.15
Additional rules:
  • Headings are styled by size only. No extra bold on headings.
  • All text and headings must be default black. No additional colors on any text or headings.
  • Add space after each paragraph.
  • Tables use alternating row shading for readability.
  • When generating .docx, explicitly set Arial font, heading sizes, black color (no theme colors), and 1.15 line spacing on every element. Never rely on Word default styles.
所有.docx输出遵循以下标准：
元素设置
字体Arial
字体颜色所有文本和标题均为黑色（默认）
正文10pt
一级标题20pt
二级标题16pt
三级标题14pt
四级标题12pt
行间距1.15
附加规则：
  • 标题仅通过字号区分样式，无需额外加粗。
  • 所有文本和标题必须为默认黑色，不得使用其他颜色。
  • 每个段落之后添加间距。
  • 表格采用交替行底纹以提升可读性。
  • 生成.docx时，需为每个元素明确设置Arial字体、标题字号、黑色（不使用主题颜色）及1.15行间距，切勿依赖Word默认样式。

References

参考资料

  • references/prompt-structure-guide.md
    — 9-component prompt formula, character budget allocation, compression techniques
  • references/docx-generator.js
    — Node.js script for generating .docx files from JSON data
  • references/AI Prompting Techniques.pdf
    — Source material for the prompt structure framework
  • references/prompt-structure-guide.md
    — 九要素Prompt公式、字符预算分配、压缩技巧
  • references/docx-generator.js
    — 从JSON数据生成.docx文件的Node.js脚本
  • references/AI Prompting Techniques.pdf
    — Prompt结构框架的参考资料

Related Skills

相关技能

  • agent-skill-builder
    — The reverse direction: converts ChatGPT Projects into Claude skills
  • skill-creator
    — Guide for creating new Claude skills from scratch
  • agent-skill-builder
    — 反向转换：将ChatGPT Projects转换为Claude技能
  • skill-creator
    — 从零开始创建Claude技能的指南