跳到主内容
智客 ZICQ

技能库 智客分类:Agent 工作流 skill-router

Skill Router

当用户询问要使用或需要帮助在所支持的主机之间选择哪种技能时,将代理请求路由到最窄的匹配可移植技能.

864 安装量

官方网址:skills.sh

技能介绍

先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。

做什么

当用户询问要使用或需要帮助在所支持的主机之间选择哪种技能时,将代理请求路由到最窄的匹配可移植技能.

何时用

官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。

代理如何加载

按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Skill Router、When This Skill Activates、Available Skills Catalog、Meta & Workflow、Core Development、Documentation & Testing。 其中含规范建议的小节:分步指令。

文件分析

文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。

官方 description(原文)

Routes an agent request to the narrowest matching portable skill when the user asks which skill to use or needs help selecting one across supported hosts.

Skill RouterWhen This Skill ActivatesAvailable Skills CatalogMeta & WorkflowCore DevelopmentDocumentation & TestingArchitecture & DevOpsPlanning & AnalysisDesign & UXRouting ProcessStep 1: Intent AnalysisStep 2: Skill Matching

· allowed-tools:Read, AskUserQuestion, WebSearch, Grep

来源分类:skills.sh agent-skill

SKILL.md 与 Agent 调用

官方规范 ↗
name
skill-router
description
Routes an agent request to the narrowest matching portable skill when the user asks which skill to use or needs help selecting one across supported hosts.
allowed-tools
Read, AskUserQuestion, WebSearch, Grep实验字段,支持情况取决于客户端;字段声明本身不会授予工具权限。
  1. 发现技能客户端向 Agent 提供名称与描述目录。
  2. 匹配与调用用户指定或任务匹配后,载入 SKILL.md 指令。
  3. 按需加载按步骤读取参考文档、使用脚本与素材。

具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗

安装这个技能

Skills CLI ↗

先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。

交给 Agent 安装

复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。

把 Agent Skill「skill-router」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-466760ce5d616c52-%E6%8A%80%E8%83%BD%E8%B7%AF%E7%94%B1%E5%99%A8.html
请存为 .cursor/skills/skill-router/SKILL.md 或 .claude/skills/skill-router/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。

GitHub 完整包 ↗

终端安装 · Skills CLI

需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。

npx skills add 'https://github.com/zhaono1/agent-playbook' --list

npx skills add 'https://github.com/zhaono1/agent-playbook' --skill 'skill-router'

CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。

阅读排版

name: skill-router description: Routes an agent request to the narrowest matching portable skill when the user asks which skill to use or needs help selecting one across supported hosts. allowed-tools: Read, AskUserQuestion, WebSearch, Grep metadata: hooks: after_complete: - trigger: session-logger mode: auto reason: "Log skill routing decisions"

Skill Router

Analyze the request and recommend the narrowest matching skill exposed by the current host. Do not assume Claude Code, Codex, Gemini, or DeepSeek Harness capabilities that are not visible in the current session.

When This Skill Activates

This skill activates when you:

  • Ask "which skill should I use?" or "what skill can help with...?"
  • Say "use a skill" without specifying which one
  • Express a need but aren't sure which skill fits
  • Mention "skill router" or "help me find a skill"

Available Skills Catalog

Meta & Workflow

| Skill | Best For | |-------|----------| | skill-router | Finding and selecting the right skill for a task | | create-pr | Creating pull requests with bilingual documentation checks | | session-logger | Saving session summaries and activity logs | | workflow-orchestrator | Coordinating multi-skill workflows and triggers | | self-improving-agent | Learning from completed skill workflows | | auto-trigger | Declaring trigger relationships between skills |

Core Development

| Skill | Best For | |-------|----------| | commit-helper | Writing Git commit messages, formatting commits | | code-reviewer | Reviewing PRs, code changes, quality checks | | debugger | Diagnosing bugs, errors, unexpected behavior | | refactoring-specialist | Improving code structure, reducing technical debt |

Documentation & Testing

| Skill | Best For | |-------|----------| | documentation-engineer | Writing README, technical docs, code documentation | | api-documenter | Creating OpenAPI/Swagger specifications | | test-automator | Writing tests, setting up test frameworks | | qa-expert | Test strategy, quality gates, QA processes |

Architecture & DevOps

| Skill | Best For | |-------|----------| | api-designer | Designing REST/GraphQL APIs, API architecture | | security-auditor | Security audits, vulnerability reviews, OWASP Top 10 | | performance-engineer | Performance optimization, speed analysis | | deployment-engineer | CI/CD pipelines, deployment automation |

Planning & Analysis

| Skill | Best For | |-------|----------| | prd-planner | Creating PRDs with persistent file-based planning | | prd-implementation-precheck | Prechecking PRDs/specs before implementation | | architecting-solutions | Non-PRD solution design, architecture, and requirements analysis | | planning-with-files | Multi-step task planning, persistent file-based organization | | long-task-coordinator | Multi-session, delegated, or resumable work that needs explicit state and recovery |

Design & UX

| Skill | Best For | |-------|----------| | figma-designer | Analyzing Figma designs and producing implementation-ready visual specs/PRDs |

Routing Process

Step 1: Intent Analysis

Analyze the user's request to identify:

  • Task Type: What does the user want to accomplish?
  • Context: What is the working domain (web, mobile, data, etc.)?
  • Complexity: Is this a simple task or complex workflow?

Step 2: Skill Matching

Match the identified intent to the most relevant skill(s) using:

  • Keyword matching: Compare request keywords with skill descriptions
  • Semantic similarity: Understand the meaning behind the request
  • Context awareness: Consider project state and previous actions

Step 3: Interactive Clarification

If the request is ambiguous, guide the user with targeted questions:

  • What is the primary goal?
  • What type of output is expected?
  • Are there specific constraints or preferences?

Step 4: Recommendation & Execution

Present the recommended skill with:

  • Skill name and brief description
  • Why it fits the current request
  • Option to proceed or ask for alternatives

Routing Examples

Example 1: Clear Intent

User: "I need to review this pull request"

Router Analysis:

  • Keywords: "review", "pull request"
  • Intent: Code review
  • Recommendation: code-reviewer

Example 2: Ambiguous Intent

User: "Use a skill to help with my project"

Router Questions:

  1. What type of task are you working on?
  2. Are you designing, coding, testing, or documenting?

Based on answers → Recommend appropriate skill

Example 3: Multi-Skill Scenario

User: "I'm building a new API and need help with the full workflow"

Router Recommendation: Consider using multiple skills in sequence:

  1. api-designer - Design the API structure
  2. api-documenter - Document endpoints with OpenAPI
  3. test-automator - Set up API tests
  4. code-reviewer - Review implementation

Interactive Question Templates

When user intent is unclear, use these question patterns:

Goal Clarification

  • "What are you trying to accomplish with this task?"
  • "What would the ideal outcome look like?"

Domain Identification

  • "What area does this relate to: development, testing, documentation, or deployment?"
  • "Are you working on code, APIs, infrastructure, or something else?"

Stage Assessment

  • "What stage are you at: planning, implementing, testing, or maintaining?"

Preference Confirmation

  • "Do you want a quick solution or a comprehensive approach?"
  • "Are there specific tools or frameworks you're using?"

Best Practices

1. Start Broad, Then Narrow

  • Begin with general category questions
  • Drill down into specifics based on responses

2. Explain Your Reasoning

  • Tell the user why a particular skill is recommended
  • Build trust through transparency

3. Offer Alternatives

  • Present the top recommendation
  • Mention 1-2 alternatives if applicable

4. Handle Edge Cases

  • If no skill fits perfectly, suggest the closest match
  • Offer to help without a specific skill if better

5. Learn from Context

  • Consider previous interactions
  • Remember user preferences for future routing

Advanced Routing Patterns

Semantic Routing

Use semantic similarity when keywords don't match directly:

  • "clean up my code" → refactoring-specialist
  • "make my app faster" → performance-engineer
  • "check for security issues" → security-auditor
  • "resume this interrupted workflow" → long-task-coordinator

Multi-Skill Orchestrations

Suggest skill combinations for complex workflows:

  • New Feature: architecting-solutions → debugger → code-reviewer
  • API Project: api-designer → api-documenter → test-automator
  • Production Readiness: security-auditor → performance-engineer → deployment-engineer

Confidence Levels

Indicate confidence in recommendations:

  • High: Direct keyword match, clear intent
  • Medium: Semantic similarity, reasonable inference
  • Low: Ambiguous request, clarification needed

Error Recovery

If the recommended skill doesn't fit:

  1. Acknowledge the mismatch
  2. Ask follow-up questions to refine understanding
  3. Provide alternative recommendations
  4. Fall back to general assistance if needed

Output Format

When recommending a skill, use this format:

## Recommended Skill: {skill-name}

{brief description of why this skill fits}

**What it does:** {one-sentence skill description}

**Best for:** {specific use cases}

---

Would you like me to activate this skill, or would you prefer to see other options?

References

相关技能

Agent 工作流

Skill Creator

创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.

Agent 工作流

Clawdhub

使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.

Agent 工作流

Agent Team Orchestration

管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.

Agent 工作流

Superpowers

Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…