skill-manager

name: skill-manager description: "管理所有已安装的 AI 技能(Skills)，支持 Claude Code、OpenClaw、Codex、Trae、Hermes 等多种智能体。在编程/开发任务的计划阶段自动分析项目并推荐技能组合，推荐尽可能简洁且必须基于已有的技能。若现有技能无法满足需求，再推荐可下载的技能。询问用户是否按以上技能组合，想替换哪些技能？如果有的话可以说。在同一任务的后续步骤中，询问用户是保持原有技能组合还是推荐新的组合。安装新 Skill 时自动同步到 Excel 仓库。用户询问技能管理、计划推荐、技能调换时触发：列出、推荐、分类、调换。Manage all installed AI skills for Claude Code, OpenClaw, Codex, Trae, Hermes and other agents. Automatically analyzes projects and recommends skill combos during task planning. Recommendations must be concise and based on already installed skills first. Only recommend downloadable skills when existing skills cannot meet needs. Ask user whether to use this combo and which skills to replace if any. In follow-up steps of the same task, asks user whether to keep existing combo or recommend new one. Auto-syncs to Excel database when new skills are installed."

Skill Manager — 技能管理器 · Skill Manager

📖 概述 · Overview

中文 | 本技能是全套 Skill 管理系统的核心入口，支持 Claude Code、OpenClaw、Codex、Trae、Hermes 等多种智能体平台。当你开始一个编程任务时，它会自动分析项目上下文、推荐最优技能组合、并让你自由调换。同时，每当有新技能被安装时，它会自动同步到 Excel 仓库。

English | This skill is the core entry point for the Skill Management system, supporting Claude Code, OpenClaw, Codex, Trae, Hermes and other agent platforms. When you start a programming task, it automatically analyzes the project context, recommends the best skill combination, and lets you freely swap skills. It also auto-syncs to the Excel database whenever new skills are installed.

支持的智能体平台 · Supported Agent Platforms

核心设计理念 · Core Design Principles

中文:

• 🎯 统一技能管理：跨平台技能统一管理，不必为每个平台单独维护
• 🧠 智能推荐：基于项目类型和技术栈自动匹配最佳技能组合
• 🔄 灵活切换：同一任务的后续步骤可保持或重新选择技能组合
• 📊 可视化仓库：Excel 表格直观展示所有技能分类和信息

English:

• 🎯 Unified Skill Management : Manage skills across platforms, no need to maintain separately for each
• 🧠 Smart Recommendation : Automatically match best skill combination based on project type & stack
• 🔄 Flexible Switching : Keep or re-select skill combo for follow-up steps in same task
• 📊 Visual Repository : Excel spreadsheet for intuitive skill category & info display

三大核心能力 · Three Core Capabilities

🎯 触发时机 · When to Trigger

✅ 必须触发 · Must Trigger

⏹ 不触发 · Do NOT Trigger

🔄 后续任务触发 · Follow-up Task Trigger

🔄 核心工作流 · Core Workflows

流程 A：首次下载 → 自动检索已有技能

Flow A: First Install → Auto-scan Existing Skills

中文 | 当用户首次安装 skill-manager 时（检测到 Skill仓库.xlsx 不存在），自动执行以下步骤：

English | When the user first installs skill-manager (detected by missing Skill仓库.xlsx), automatically run:

1. Auto-run: python scripts/sync-skills-repo.py
2. Scan ~/.agents/skills/ and ~/.claude/skills/ for all installed skills
3. Read each skill's SKILL.md, extract name and description
4. Classify into 16 categories (keyword matching + weighted scoring)
5. Generate Skill仓库.xlsx (visual Excel spreadsheet)
6. Generate skills-index.json (index for recommendation engine)
7. Report total skills found and category distribution

流程 B：安装新技能 → 自动填入

Flow B: New Skill Install → Auto-register

中文 | 通过 npx skills add 安装新技能时，PostToolUse hook 自动触发同步：

English | When installing via npx skills add, the PostToolUse hook auto-triggers sync:

Trigger conditions:
  - .claude/settings.json has PostToolUse hook configured
  - Hook matches "npx skills (add|install|remove|update)"
  - Auto-executes: python scripts/sync-skills-repo.py

Manual trigger:
  python scripts/sync-skills-repo.py
  Or double-click sync-skills.bat (Windows)

流程 C：编程任务计划 → 技能推荐（核心流程）

Flow C: Task Planning → Skill Recommendation (Core)

中文 | 当用户提出一个编程/开发任务时，按照以下步骤执行：

English | When the user starts a programming/development task, follow these steps:

Step 1: Analyze project context
   Run: python scripts/skill-recommend.py --project <path> --output json
   Read output for:
   - Project type: frontend / backend / fullstack / AI / CLI
   - Tech stack: React, Python, Node.js...
   - Language: TypeScript, Python, Go...
   - Has tests? Has database?

Step 2: Format and display recommendations

   =================================================================
     >> Skill 组合推荐 · Skill Combo Recommendation <<
     项目 Project: my-app
     类型 Type: 全栈项目 Full-stack
     技术栈 Stack: React, FastAPI, TypeScript
   =================================================================

     [推荐技能 Recommended Skills]
       -> 基于已安装技能推荐 · Based on installed skills
       1. skill-manager    - 技能管理 · Skill management
       2. frontend-design  - 前端界面设计 · Frontend UI design

   -----------------------------------------------------------------
   是否使用以上技能组合？想替换哪些技能？如果有的话可以说。
   Use this combo? Which skills to replace if any?
     1. 直接开始 · Start with this combo
     2. 替换技能 · Swap skills
     3. 跳过推荐 · Skip
   -----------------------------------------------------------------

Step 3: Wait for user choice
   - [1] -> Confirm combo, start task
   - [2] -> Enter skill swap flow
   - [3] -> Skip recommendation

Step 4: Start task
   Continue original task logic after user confirms.

流程 D：技能调换（组合自定义）

Flow D: Skill Swapping (Combo Customization)

中文 | 当用户选择 "替换组合中的技能" 时：

English | When the user chooses to "swap skills":

1. List all skills in the current recommendation (numbered)
2. List all available skills (grouped by category)
3. Ask user:
   "Remove which skill? (number/name, comma-separated, empty=skip)
    要移除哪个技能？(编号/名称, 多个用逗号分隔, 空=跳过)"
   "Add which skill? (name, comma-separated, empty=skip)
    要添加哪个技能？(名称, 多个用逗号分隔, 空=跳过)"
4. Adjust combo based on input
5. Show the new combo
6. Ask again: "Satisfied? [y/n] · 是否满意？[y/n]"
   - y: Confirm combo
   - n: Back to step 1
7. Optional: "Save preference? · 是否保存此偏好？"
   -> Auto-applied for similar projects next time

调换规则 · Swap Rules

流程 E：同一任务后续步骤 · 选择技能组合

Flow E: Same Task Follow-up · Choose Skill Combo

中文 | 当用户在同一任务下提出后续编程/修改需求时，按照以下步骤执行：

English | When user presents follow-up programming/modification needs in the same task, follow these steps:

Step 1: Detect task context
   - Is this a follow-up to the same original task?
   - Has a skill combo already been selected for this task?

Step 2: Display confirmation prompt

   =================================================================
     >> 技能组合选择 · Skill Combo Choice <<
     检测到这是同一任务的后续步骤
     Detected: this is a follow-up to the same task
   =================================================================

     当前技能组合 Current Combo:
       1. skill-manager    - 技能管理
       2. frontend-design  - 前端界面设计

   -----------------------------------------------------------------
   选择操作 Choose action:
     1. 保持原有技能组合 · Keep current combo
     2. 重新推荐技能组合 · Recommend new combo
     3. 手动调整技能组合 · Adjust manually
   -----------------------------------------------------------------

Step 3: Wait for user choice
   - [1] -> Keep combo, continue with current skills
   - [2] -> Re-run full recommendation (Flow C)
   - [3] -> Enter skill swap flow (Flow D)

Step 4: Continue with task
   Proceed with original task logic based on user's choice.

🛠 CLI 工具参考 · CLI Reference

# 同步技能到 Excel（首次运行自动创建）
# Sync skills to Excel (auto-creates on first run)
python scripts/sync-skills-repo.py

# 分析项目并推荐技能（交互模式）
# Analyze project & recommend skills (interactive)
python scripts/skill-recommend.py --project . --interactive

# 列出所有技能 · List all skills
python scripts/skill-recommend.py --list-all

# 查看项目信息（JSON 格式）
# View project info (JSON format)
python scripts/skill-recommend.py --project . --output json

# 添加/移除偏好 · Add/remove preferences
python scripts/skill-recommend.py --add frontend-design
python scripts/skill-recommend.py --remove lark-vc

📦 安装 · Installation

前置依赖 · Prerequisites

pip install openpyxl

安装 skill-manager · Install the Skill

# 1. 克隆或下载本仓库
#    Clone or download this repository
git clone <repo-url> ~/xielaoban
cd ~/xielaoban

# 2. 安装技能到 Claude
#    Install skill to Claude
mkdir -p ~/.claude/skills/skill-manager
cp skills/skill-manager/SKILL.md ~/.claude/skills/skill-manager/

# 3. （可选）安装到 agents 目录
#    (Optional) Install to agents directory
mkdir -p ~/.agents/skills/skill-manager
cp skills/skill-manager/SKILL.md ~/.agents/skills/skill-manager/

# 4. 首次同步 — 自动创建 Excel 仓库
#    First sync — auto-create Excel database
python scripts/sync-skills-repo.py

# 5. 测试推荐功能 · Test recommendation
python scripts/skill-recommend.py --list-all

自动触发配置 · Auto-trigger Configuration

中文 | 项目已内置 .claude/settings.json，配置了 PostToolUse hook，安装新技能后自动同步：

English | The project includes .claude/settings.json with a PostToolUse hook for auto-sync after installing new skills:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "npx skills (add|install|remove|update)",
        "hook": "python scripts/sync-skills-repo.py"
      }
    ]
  }
}

🔗 与其它技能的协作 · Ecosystem Integration

🏷️ 分类系统 · Classification System

中文 | 技能通过 scripts/skill-classifier.json 中的规则进行智能分类，采用 多维度关键词匹配 + 加权评分 算法：

English | Skills are classified using scripts/skill-classifier.json rules with multi-dimensional keyword matching + weighted scoring:

16 个分类 · 16 Categories

📁 项目文件结构 · Project Structure

xielaoban/
├── skills/
│   └── skill-manager/
│       └── SKILL.md              ← 技能文件 Skill file (核心 Core)
├── scripts/
│   ├── sync-skills-repo.py       ← 同步引擎 Sync engine
│   ├── skill-recommend.py        ← 推荐引擎 Recommend engine
│   └── skill-classifier.json     ← 16类分类规则 Classification rules
├── .claude/
│   └── settings.json             ← PostToolUse hook 配置 Hook config
├── .gitignore
├── CLAUDE.md                     ← 工作区指令 Workspace instructions
└── README.md                     ← 项目说明 Project README

自动生成的文件（不提交 Git）· Auto-generated Files (Not Committed)

🧪 技术细节 · Technical Details

推荐算法 · Recommendation Algorithm

中文:

1. 扫描项目目录，检测 package.json、requirements.txt、Cargo.toml、go.mod 等文件
2. 识别技术栈（React, Vue, FastAPI, Django 等）
3. 判断项目类型（前端/后端/全栈/AI/CLI）
4. 从 skills-index.json 中按分类匹配技能
5. 按优先级分组输出（核心 > 前端 > 后端 > 测试 > AI > 通用）

English:

1. Scan project directory for package.json, requirements.txt, Cargo.toml, go.mod, etc.
2. Detect tech stack (React, Vue, FastAPI, Django, etc.)
3. Determine project type (frontend/backend/fullstack/AI/CLI)
4. Match skills from skills-index.json by category
5. Group by priority (Core > Frontend > Backend > Testing > AI > General)

分类算法 · Classification Algorithm

score =
  name keyword matches × 3 × weight +
  desc keyword matches × 2 × weight +
  dir keyword matches × 2 × weight

confidence = normalize(score, 0-100)