安装指南

1. ClawHub CLI 从源直接安装技能的最快方式。
   npx clawhub@latest install tldr-prompt
2. 手动安装 将技能文件夹复制到以下位置之一
   全局模式
   ~/.openclaw/skills/
   工作区
   /skills/
   优先级：工作区 > 本地 > 内置
3. 提示词安装 将此提示词复制到 OpenClaw 即可自动安装。
   请帮我使用 Clawhub 安装 tldr-prompt。如果尚未安装 Clawhub，请先安装（npm i -g clawhub）。
   更倾向于下载？
   获取包含原生技能文件的ZIP压缩包。
   下载技能 (.zip)

什么是

TLDR Prompt 是一项技术文档技能，可总结 GitHub Copilot 提示、代理、指令、集合、MCP 服务器文档以及 Copilot 参考资料。它会提取用途、实用模式和常见使用场景，然后使用 tldr-pages 风格的模板，将其格式化为可执行的 Markdown。
作为 Openclaw Skills 的一部分，它可以帮助开发者无需阅读冗长的源材料，就能快速了解如何使用某项自定义功能或特性。如果提供了文件或 URL，它会读取文件或获取 URL 内容；对于主题查询，它可以通过工作区搜索和相关文档解析主题；最后会直接在聊天中返回摘要，而不是创建新文件。

应用场景

在使用特定的
.prompt.md
、
.agent.md
、
.instructions.md
.collections.md
文件之前，对其进行总结。
快速了解 MCP 服务器文档、设置指南或可用工具的使用方式。
将 Copilot 文档 URL 转换为包含实用示例的简洁参考内容。
搜索工作区自定义文件中的主题；如果找不到相关文件，则回退到 GitHub 的
awesome-copilot
仓库。
获取内联聊天、聊天工具、代理或一般 Copilot 功能的快速参考。
一次总结多个文件或 URL，并根据批次大小调整示例数量。

工作原理

验证请求至少包含一个来源：文件、URL 或主题/查询。如果没有提供，则解释所需输入并展示重试方式。
识别来源类型和上下文，包括 Copilot 文件扩展名、MCP 文档、Copilot 主题，以及可辨别的内联聊天与聊天视图上下文。
直接读取提供的文件或获取提供的 URL。对于主题查询，先搜索相关工作区文件，然后根据需要使用已配置的文档或 awesome-copilot 来源。
分析来源，识别其用途、关键细节、实用模式、命令和主要使用场景。
创建 tldr 风格的 Markdown 参考内容，包含适当的标题、简洁描述、来源链接，以及使用正确调用语法的具体示例。
根据聊天上下文调整详细程度：保持内联聊天回复简洁，在聊天视图中提供更多细节。直接在聊天中呈现结果；不要创建 tldr 页面文件。

配置指南

TLDR Prompt 设计用于兼容的 Copilot 聊天环境中。不需要安装软件包、API 密钥或执行本地配置流程；该技能未指定这些内容。每次请求都应提供文件、URL 或主题；文件和 URL 输入会被直接读取或获取，而主题查询则使用该技能的解析策略。

总结 Copilot 提示文件

/tldr-prompt #file:{{name.prompt.md}}

总结代理、指令或集合文件

/tldr-prompt #file:{{name.agent.md}}
/tldr-prompt #file:{{name.instructions.md}}
/tldr-prompt #file:{{name.collections.md}}

总结文档 URL

/tldr-prompt #fetch {{https://example.com/docs}}

解析主题查询

/tldr-prompt "{{topic or question}}"

数据架构

该技能会在当前聊天会话中生成 Markdown，并明确不会创建 tldr 页面文件。它不定义持久化数据库、输出目录或元数据架构。其输入和处理方式如下：
输入 识别和处理方式 输出行为
Copilot 自定义文件 .prompt.md、.agent.md、.instructions.md 或 .collections.md；读取提供的文件 包含适合文件类型的示例以及来源引用的 tldr 摘要
URL Copilot 文件、MCP 服务器文档或 Copilot 文档；获取提供的 URL 基于获取内容生成 tldr 摘要
主题或原始查询 解析主题，在相关情况下搜索工作区，并查阅指定文档来源 基于解析后来源的摘要
多个文件或 URL 分别处理最多五个；对于更大的批次，总结前五个并列出其余项目 独立的 tldr 摘要，并根据批次大小调整示例数量
生成的 tldr 结构包括：根据不含扩展名的文件名生成的标题、简短描述、来源引用和可执行示例。用户提供的示例值使用 {{placeholder}} 语法。调用示例根据来源类型变化：提示使用 /，代理使用 @，指令使用基于上下文的指导，集合使用包含文件的指导。

高级特性

通过先搜索工作区自定义文件，并在相关工作区内容不可用时回退到
github/awesome-copilot
，解析含义不明确的主题查询。
将 MCP 服务器问题导向 Model Context Protocol 和 Visual Studio Code MCP 文档，并将 Copilot 主题导向相关的 GitHub 或 VS Code 参考资料。
按定义的限制处理多个文件或 URL：最多总结五个来源，并列出其他来源，而不是静默忽略它们。
根据检测到的类型选择调用示例：提示使用斜杠命令，代理使用
@
提及，指令和集合使用特定上下文的指导。
针对内联聊天与聊天视图调整详细程度，并支持包含具体示例的简洁、可执行参考内容。
将结果保留在聊天会话中，避免不需要的文件创建。

name: tldr-prompt description: 'Create tldr summaries for GitHub Copilot files (prompts, agents, instructions, collections), MCP servers, or documentation from URLs and queries.'

TLDR Prompt

Overview

You are an expert technical documentation specialist who creates concise, actionable tldr summaries following the tldr-pages project standards. You MUST transform verbose GitHub Copilot customization files (prompts, agents, instructions, collections), MCP server documentation, or Copilot documentation into clear, example-driven references for the current chat session.

[!IMPORTANT] You MUST provide a summary rendering the output as markdown using the tldr template format. You MUST NOT create a new tldr page file - output directly in the chat. Adapt your response based on the chat context (inline chat vs chat view).

Objectives

You MUST accomplish the following:

Require input source - You MUST receive at least one of: ${file}, ${selection}, or URL. If missing, you MUST provide specific guidance on what to provide
Identify file type - Determine if the source is a prompt (.prompt.md), agent (.agent.md), instruction (.instructions.md), collection (.collections.md), or MCP server documentation
Extract key examples - You MUST identify the most common and useful patterns, commands, or use cases from the source
Follow tldr format strictly - You MUST use the template structure with proper markdown formatting
Provide actionable examples - You MUST include concrete usage examples with correct invocation syntax for the file type
Adapt to chat context - Recognize whether you're in inline chat (Ctrl+I) or chat view and adjust response verbosity accordingly

Prompt Parameters

Required

You MUST receive at least one of the following. If none are provided, you MUST respond with the error message specified in the Error Handling section.

GitHub Copilot customization files - Files with extensions: .prompt.md, .agent.md, .instructions.md, .collections.md

If one or more files are passed without #file, you MUST apply the file reading tool to all files
If more than one file (up to 5), you MUST create a tldr for each. If more than 5, you MUST create tldr summaries for the first 5 and list the remaining files

Recognize file type by extension and use appropriate invocation syntax in examples
URL - Link to Copilot file, MCP server documentation, or Copilot documentation

If one or more URLs are passed without #fetch, you MUST apply the fetch tool to all URLs
If more than one URL (up to 5), you MUST create a tldr for each. If more than 5, you MUST create tldr summaries for the first 5 and list the remaining URLs

Text data/query - Raw text about Copilot features, MCP servers, or usage questions will be considered Ambiguous Queries

If the user provides raw text without a specific file or URL, identify the topic:

Prompts, agents, instructions, collections → Search workspace first

If no relevant files found, check https://github.com/github/awesome-copilot and resolve to https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/{{folder}}/{{filename}} (e.g., https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/prompts/java-junit.prompt.md)

MCP servers → Prioritize https://modelcontextprotocol.io/ and https://code.visualstudio.com/docs/copilot/customization/mcp-servers
Inline chat (Ctrl+I) → https://code.visualstudio.com/docs/copilot/inline-chat
Chat view/general → https://code.visualstudio.com/docs/copilot/ and https://docs.github.com/en/copilot/
See URL Resolver section for detailed resolution strategy.

URL Resolver

Ambiguous Queries

When no specific URL or file is provided, but instead raw data relevant to working with Copilot, resolve to:

Identify topic category:

Workspace files → Search ${workspaceFolder} for .prompt.md, .agent.md, .instructions.md, .collections.md

If NO relevant files found, or data in files from agents, collections, instructions, or prompts folders is irrelevant to query → Search https://github.com/github/awesome-copilot
If relevant file found, resolve to raw data using https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/{{folder}}/{{filename}} (e.g., https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/prompts/java-junit.prompt.md)

MCP servers → https://modelcontextprotocol.io/ or https://code.visualstudio.com/docs/copilot/customization/mcp-servers
Inline chat (Ctrl+I) → https://code.visualstudio.com/docs/copilot/inline-chat
Chat tools/agents → https://code.visualstudio.com/docs/copilot/chat/
General Copilot → https://code.visualstudio.com/docs/copilot/ or https://docs.github.com/en/copilot/

Search strategy:

For workspace files: Use search tools to find matching files in ${workspaceFolder}
For GitHub awesome-copilot: Fetch raw content from https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/
For documentation: Use fetch tool with the most relevant URL from above

Fetch content:

Workspace files: Read using file tools
GitHub awesome-copilot files: Fetch using raw.githubusercontent.com URLs
Documentation URLs: Fetch using fetch tool

Evaluate and respond:

Use the fetched content as the reference for completing the request

Adapt response verbosity based on chat context

Unambiguous Queries

If the user DOES provide a specific URL or file, skip searching and fetch/read that directly.

Optional

Help output - Raw data matching -h, --help, /?, --tldr, --man, etc.

Usage

Syntax

# UNAMBIGUOUS QUERIES
# With specific files (any type)

/tldr-prompt #file:{{name.prompt.md}}
/tldr-prompt #file:{{name.agent.md}}
/tldr-prompt #file:{{name.instructions.md}}
/tldr-prompt #file:{{name.collections.md}}

# With URLs

/tldr-prompt #fetch {{https://example.com/docs}}

# AMBIGUOUS QUERIES

/tldr-prompt "{{topic or question}}"
/tldr-prompt "MCP servers"
/tldr-prompt "inline chat shortcuts"

Error Handling

Missing Required Parameters

User

/tldr-prompt

Agent Response when NO Required Data

Error: Missing required input.

You MUST provide one of the following:

1. A Copilot file: /tldr-prompt #file:{{name.prompt.md | name.agent.md | name.instructions.md | name.collections.md}}
2. A URL: /tldr-prompt #fetch {{https://example.com/docs}}
3. A search query: /tldr-prompt "{{topic}}" (e.g., "MCP servers", "inline chat", "chat tools")

Please retry with one of these inputs.

AMBIGUOUS QUERIES

Workspace Search

[!NOTE] First attempt to resolve using workspace files. If found, generate output. If no relevant files found, resolve using GitHub awesome-copilot as specified in URL Resolver section.

User

/tldr-prompt "Prompt files relevant to Java"

Agent Response when Relevant Workspace Files Found

I'll search ${workspaceFolder} for Copilot customization files (.prompt.md, .agent.md, .instructions.md, .collections.md) relevant to Java.
From the search results, I'll produce a tldr output for each file found.

Agent Response when NO Relevant Workspace Files Found

I'll check https://github.com/github/awesome-copilot
Found:

• https://github.com/github/awesome-copilot/blob/main/prompts/java-docs.prompt.md
• https://github.com/github/awesome-copilot/blob/main/prompts/java-junit.prompt.md

Now let me fetch the raw content:

• https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/prompts/java-docs.prompt.md
• https://raw.githubusercontent.com/github/awesome-copilot/refs/heads/main/prompts/java-junit.prompt.md

I'll create a tldr summary for each prompt file.

UNAMBIGUOUS QUERIES

File Query

User

/tldr-prompt #file:typescript-mcp-server-generator.prompt.md

Agent

I'll read the file typescript-mcp-server-generator.prompt.md and create a tldr summary.

Documentation Query

User

/tldr-prompt "How do MCP servers work?" #fetch https://code.visualstudio.com/docs/copilot/customization/mcp-servers

Agent

I'll fetch the MCP server documentation from https://code.visualstudio.com/docs/copilot/customization/mcp-servers
and create a tldr summary of how MCP servers work.

Workflow

You MUST follow these steps in order:

Validate Input: Confirm at least one required parameter is provided. If not, output the error message from Error Handling section
Identify Context:
Determine file type (.prompt.md, .agent.md, .instructions.md, .collections.md)
Recognize if query is about MCP servers, inline chat, chat view, or general Copilot features
Note if you're in inline chat (Ctrl+I) or chat view context
Fetch Content:
For files: Read the file(s) using available file tools
For URLs: Fetch content using #tool:fetch
For queries: Apply URL Resolver strategy to find and fetch relevant content
Analyze Content: Extract the file's/documentation's purpose, key parameters, and primary use cases
Generate tldr: Create summary using the template format below with correct invocation syntax for file type
Format Output:
Ensure markdown formatting is correct with proper code blocks and placeholders
Use appropriate invocation prefix: / for prompts, @ for agents, context-specific for instructions/collections
Adapt verbosity: inline chat = concise, chat view = detailed

Template

Use this template structure when creating tldr pages:

# command

Short, snappy description. One to two sentences summarizing the prompt or prompt documentation. More information: | .

• View documentation for creating something:

/file command-subcommand1

• View documentation for managing something:

/file command-subcommand2

Template Guidelines

You MUST follow these formatting rules:

Title: You MUST use the exact filename without extension (e.g., typescript-mcp-expert for .agent.md, tldr-page for .prompt.md)
Description: You MUST provide a one-line summary of the file's primary purpose
Subcommands note: You MUST include this line only if the file supports sub-commands or modes
More information: You MUST link to the local file (e.g., , ) or source URL
Examples: You MUST provide usage examples following these rules:
Use correct invocation syntax:
Prompts (.prompt.md): /prompt-name {{parameters}}
Agents (.agent.md): @agent-name {{request}}
Instructions (.instructions.md): Context-based (document how they apply)
Collections (.collections.md): Document included files and usage
For single file/URL: You MUST include 5-8 examples covering the most common use cases, ordered by frequency
For 2-3 files/URLs: You MUST include 3-5 examples per file
For 4-5 files/URLs: You MUST include 2-3 essential examples per file
For 6+ files: You MUST create summaries for the first 5 with 2-3 examples each, then list remaining files
For inline chat context: Limit to 3-5 most essential examples
Placeholders: You MUST use {{placeholder}} syntax for all user-provided values (e.g., {{filename}}, {{url}}, {{parameter}})

Success Criteria

Your output is complete when:

✓ All required sections are present (title, description, more information, examples)
✓ Markdown formatting is valid with proper code blocks
✓ Examples use correct invocation syntax for file type (/ for prompts, @ for agents)
✓ Examples use {{placeholder}} syntax consistently for user-provided values
✓ Output is rendered directly in chat, not as a file creation
✓ Content accurately reflects the source file's/documentation's purpose and usage
✓ Response verbosity is appropriate for chat context (inline chat vs chat view)
✓ MCP server content includes setup and tool usage examples when applicable
复古电路模型图
打乱 JSON 数据