Decomposes complex user requests into executable subtasks, identifies required capabilities, searches for existing skills at skills.sh, and creates new skills when no solution exists. This skill should be used when the user submits a complex multi-step request, wants to automate workflows, or needs help breaking down large tasks into manageable pieces.

# Task Decomposer & Skill Generator

此技能有助于将复杂的用户请求分解为可执行的子任务，识别每个任务所需的能力，从开放技能生态系统中搜索现有技能，并在没有可用解决方案时自动创建新技能。

## 核心工作流

``` User Request → Task Decomposition → Capability Identification → Skill Search → Gap Analysis → Skill Creation → Execution Plan ```

## 第一阶段：任务分析与分解

收到用户请求时，请遵循以下步骤：

### 步骤 1：理解用户意图

分析请求以识别： - **核心目标**：最终目标是什么？ - **涉及领域**：需要哪些专业知识领域？ - **触发机制**：一次性、计划任务还是事件驱动？

分析示例： ``` User Input: "Help me get email summaries every morning and send them to Slack"

Analysis: - Core objective: Automated email digest delivery to Slack - Domains: Email access, content summarization, messaging - Trigger: Scheduled (daily morning) ```

### 步骤 2：分解为原子任务

将复杂任务拆解为最小的可执行单元：

```yaml Task Decomposition: - task_id: 1 name: "Access and retrieve email list" type: "data_retrieval" input: "Email credentials/session" output: "List of emails with metadata" dependencies: [] - task_id: 2 name: "Extract key information from emails" type: "data_extraction" input: "Email list" output: "Structured email data" dependencies: [1] - task_id: 3 name: "Generate email summary" type: "content_generation" input: "Structured email data" output: "Formatted summary text" dependencies: [2] - task_id: 4 name: "Send message to Slack" type: "message_delivery" input: "Summary text, Slack webhook/token" output: "Delivery confirmation" dependencies: [3] - task_id: 5 name: "Configure scheduled execution" type: "scheduling" input: "Workflow script, schedule config" output: "Active scheduled job" dependencies: [4] ```

## 第二阶段：能力识别

将每个子任务映射到通用能力分类中的一种能力类型。

### 通用能力类型

| 能力 | 描述 | 搜索关键词 | |------------|-------------|-----------------| | `browser_automation` | 网页导航、交互、抓取 | browser, selenium, puppeteer, playwright, scrape | | `web_search` | 互联网搜索和信息检索 | search, google, bing, duckduckgo | | `api_integration` | 第三方 API 通信 | api, rest, graphql, webhook, {service-name} | | `data_extraction` | 解析和提取结构化数据 | parse, extract, scrape, ocr, pdf | | `data_transformation` | 转换、清洗、转换数据 | transform, convert, format, clean, etl | | `content_generation` | 创建文本、图像或其他内容 | generate, write, create, summarize, translate | | `file_operations` | 读取、写入、操作文件 | file, read, write, csv, excel, json, pdf | | `message_delivery` | 发送通知或消息 | notify, send, email, slack, discord, telegram | | `scheduling` | 基于时间的任务执行 | schedule, cron, timer, daily, weekly | | `authentication` | 身份和访问管理 | auth, oauth, login, token, credentials | | `database_operations` | 数据库 CRUD 操作 | database, sql, mongodb, query, store | | `code_execution` | 运行脚本或程序 | execute, run, script, shell, python | | `version_control` | Git 和代码仓库操作 | git, github, gitlab, commit, pr, review | | `testing` | 自动化测试和 QA | test, jest, pytest, e2e, unit | | `deployment` | 应用部署和 CI/CD | deploy, docker, kubernetes, ci-cd, release | | `monitoring` | 系统和应用监控 | monitor, alert, log, metrics, health |

### 能力识别流程

对于每个子任务： 1. 分析任务描述和要求 2. 匹配到一个或多个能力类型 3. 生成搜索关键词以发现技能

示例： ```yaml Task: "Send message to Slack" Capability: message_delivery Search Keywords: ["slack", "notification", "message", "webhook"] ```

## 第三阶段：技能搜索

使用 Skills CLI 在 https://skills.sh/ 搜索现有技能。

### 搜索流程

对于每个能力需求，使用相关关键词进行搜索：

```bash # Search for skills matching the capability npx skills find <keyword>

# Examples: npx skills find slack notification npx skills find browser automation npx skills find pdf extract npx skills find github api ```

### 评估搜索结果

返回结果时： ``` Install with npx skills add <owner/repo@skill>

owner/repo@skill-name └ https://skills.sh/owner/repo/skill-name ```

评估每个结果： - **相关性**：是否符合所需能力？ - **完整性**：是否覆盖所有必需功能？ - **质量**：文档是否完善且维护良好？

### 生成能力映射

```yaml Capability Mapping: - task_id: 1 capability: browser_automation search_query: "browser email automation" found_skills: - name: "anthropic/claude-skills@browser-use" url: "https://skills.sh/anthropic/claude-skills/browser-use" match_score: high recommendation: "Install browser-use skill" - task_id: 4 capability: message_delivery search_query: "slack notification" found_skills: [] recommendation: "Create new skill: slack-notification" ```

## 第四阶段：差距分析

识别没有匹配技能的任务：

### 内置能力（无需技能）

这些能力通常由代理的原生能力处理： - `content_generation` - LLM 的原生文本生成 - `data_transformation` - 通过代码进行基本数据操作 - `code_execution` - 直接脚本执行 - `scheduling` - 系统级 cron/调度器配置

### 需要技能

对于没有内置支持的能力，确定： 1. **技能存在**：从 skills.sh 安装 2. **技能未找到**：创建新技能

## 第五阶段：技能创建

当没有现有技能匹配所需能力时，创建一个新技能。

### 技能创建流程

1. **定义范围**：确定技能应该做什么 2. **设计接口**：定义输入、输出和使用模式 3. **创建 SKILL.md**：编写技能定义文件 4. **添加资源**：根据需要包含脚本、参考或资产

### 技能模板

```markdown --- name: {skill-name} description: {Clear description of what the skill does and when to use it. Written in third person.} ---

# {Skill Title}

{Brief introduction explaining the skill's purpose.}

## When to Use

{Describe scenarios when this skill should be triggered.}

## Prerequisites

{List any required installations, configurations, or credentials.}

## Usage

{Detailed usage instructions with examples.}

### Basic Usage

```bash {Basic command or code example} ```

### Advanced Usage

{More complex examples and options.}

## Configuration

{Any configuration options or environment variables.}

## Examples

### Example 1: {Use Case}

{Step-by-step example with code.}

## Troubleshooting

{Common issues and solutions.} ```

### 初始化新技能

```bash # Create skill using the skills CLI npx skills init <skill-name>

# Or manually create the structure: # skill-name/ # ├── SKILL.md (required) # ├── scripts/ (optional) # ├── references/ (optional) # └── assets/ (optional) ```

## 第六阶段：生成执行计划

将所有信息编译成结构化的执行计划：

```yaml Execution Plan: title: "{Task Description}" prerequisites: - "{Prerequisite 1}" - "{Prerequisite 2}" skills_to_install: - skill: "owner/repo@skill-name" command: "npx skills add owner/repo@skill-name -g -y" url: "https://skills.sh/owner/repo/skill-name" skills_to_create: - name: "{new-skill-name}" capability: "{capability_type}" description: "{What it does}" execution_steps: - step: 1 task: "{Task name}" skill: "{skill-name | built-in}" action: "{Specific action to take}" - step: 2 task: "{Task name}" skill: "{skill-name | built-in}" action: "{Specific action to take}" verification: - "{How to verify step 1 succeeded}" - "{How to verify step 2 succeeded}" ```

## 任务分解原则

### 原则 1：原子性

每个子任务应是具有清晰输入和输出的最小可执行单元。

### 原则 2：独立性

最小化任务之间的依赖关系，以便在可能的情况下允许并行执行。

### 原则 3：可验证性

每个任务都应有明确的方式来验证成功完成。

### 原则 4：可重用性

识别可重用的模式，并优先创建通用技能。

### 原则 5：单一职责

每个任务应做好一件事。

## 输出格式

以结构化格式呈现分解结果：

``` ════════════════════════════════════════════════════════════════ 📋 TASK DECOMPOSITION REPORT ════════════════════════════════════════════════════════════════

🎯 Original Request: {User's original request}

──────────────────────────────────────────────────────────────── 📊 SUBTASKS ──────────────────────────────────────────────────────────────── ┌─────┬────────────────────────┬───────────────────┬───────────┐ │ ID │ Task │ Capability │ Status │ ├─────┼────────────────────────┼───────────────────┼───────────┤ │ 1 │ {task name} │ {capability} │ Found │ │ 2 │ {task name} │ {capability} │ Built-in │ │ 3 │ {task name} │ {capability} │ Create │ └─────┴────────────────────────┴───────────────────┴───────────┘

──────────────────────────────────────────────────────────────── 🔍 SKILL SEARCH RESULTS ──────────────────────────────────────────────────────────────── Task 1: {task name} Search: npx skills find {keywords} Found: owner/repo@skill-name URL: https://skills.sh/owner/repo/skill-name Task 3: {task name} Search: npx skills find {keywords} Found: No matching skills Action: Create new skill

──────────────────────────────────────────────────────────────── 🛠️ SKILLS TO CREATE ──────────────────────────────────────────────────────────────── 1. {skill-name} Capability: {capability_type} Description: {what it does}

──────────────────────────────────────────────────────────────── 📝 EXECUTION PLAN ──────────────────────────────────────────────────────────────── Prerequisites: • {prerequisite 1} • {prerequisite 2}

Steps: 1. {action} using {skill} 2. {action} using {skill} 3. {action} using {skill}

════════════════════════════════════════════════════════════════ ```

## 示例

### 示例 1：工作流自动化

**用户请求：** ``` Create a workflow that monitors GitHub issues, summarizes new issues, and posts notifications to Discord ```

**分解：** ```yaml Subtasks: 1. Monitor GitHub repository for new issues Capability: api_integration Search: "npx skills find github issues" 2. Extract issue content and metadata Capability: data_extraction Status: Built-in (code) 3. Generate issue summary Capability: content_generation Status: Built-in (LLM) 4. Send notification to Discord Capability: message_delivery Search: "npx skills find discord notification" 5. Configure webhook or polling trigger Capability: scheduling Status: Built-in (system) ```

### 示例 2：数据管道

**用户请求：** ``` Search for AI research papers, download PDFs, extract key findings, and save to Notion ```

**分解：** ```yaml Subtasks: 1. Search for AI research papers Capability: web_search Search: "npx skills find academic search" 2. Download PDF files Capability: browser_automation Search: "npx skills find browser download" 3. Extract text from PDFs Capability: data_extraction Search: "npx skills find pdf extract" 4. Generate summaries of key findings Capability: content_generation Status: Built-in (LLM) 5. Save to Notion database Capability: api_integration Search: "npx skills find notion" ```

## 最佳实践

1. **从技能搜索开始**：在创建新技能之前务必检查 https://skills.sh/ 2. **使用具体的搜索词**：将能力关键词与领域术语结合使用 3. **利用内置能力**：不要为代理原生能做的事情创建技能 4. **创建可重用技能**：尽可能将新技能设计为通用目的 5. **完善文档**：新技能应有清晰的使用说明 6. **执行前验证**：在执行任务之前确认技能安装 7. **优雅地处理错误**：在执行计划中包含回退策略

## 与 find-skills 集成

此技能与 `find-skills` 技能配合使用，用于发现现有解决方案：

```bash # Search the skills ecosystem npx skills find <query>

# Install a discovered skill npx skills add <owner/repo@skill> -g -y

# Browse all available skills # Visit: https://skills.sh/ ```

## 注意事项

- 在创建新技能之前务必搜索现有技能 - 内置能力（LLM、基础代码）不需要技能 - 创建技能需要用户确认才能继续 - 复杂的工作流可能需要多个技能协同工作