Skill Finder

Full-featured Agent Skills management tool with search, install, star, and update capabilities.

When to Use

• Looking for skills for a specific task or domain
• Finding and installing skills locally
• Managing favorite skills with star feature
• Keeping your skill index up-to-date
• Discovering similar skills by category

Features

• 🔍 Search - Local index (35+ skills) + GitHub API + Web fallback
• 🏷️ Tags - Search by category tags (#azure #bicep)
• 📦 Install - Download skills to local directory
• ⭐ Star - Mark and manage favorite skills
• 📊 Stats - View index statistics
• 🔄 Update - Sync all sources from GitHub
• 💡 Similar - Get category-based recommendations

Quick Start

Search

# Keyword search
python scripts/search_skills.py "pdf"
pwsh scripts/Search-Skills.ps1 -Query "pdf"

# Tag search (filter by category)
python scripts/search_skills.py "#azure #development"
pwsh scripts/Search-Skills.ps1 -Query "#azure #bicep"

Skill Management

# Show detailed info (includes SKILL.md content)
python scripts/search_skills.py --info skill-name

# Install to local directory
python scripts/search_skills.py --install skill-name

# Star favorite skills
python scripts/search_skills.py --star skill-name
python scripts/search_skills.py --list-starred

Index Management

# Update all sources
python scripts/search_skills.py --update

# Add new source repository
python scripts/search_skills.py --add-source https://github.com/owner/repo

# View statistics
python scripts/search_skills.py --stats

List Options

python scripts/search_skills.py --list-categories
python scripts/search_skills.py --list-sources
python scripts/search_skills.py --similar skill-name

Add New Source

When you find a good repository, add it to your index:

python scripts/search_skills.py --add-source https://github.com/owner/repo
pwsh scripts/Search-Skills.ps1 -AddSource -RepoUrl "https://github.com/owner/repo"

This will:

1. Add the repository as a source
2. Search for skills in skills/, .github/skills/, .claude/skills/
3. Auto-add found skills to your index

Command Reference

Command	Description
--info SKILL	Show skill details with SKILL.md content
--install SKILL	Download skill to ~/.skills or custom dir
--star SKILL	Add skill to favorites
--unstar SKILL	Remove from favorites
--list-starred	Show all starred skills
--similar SKILL	Find skills with matching categories
--stats	Show index statistics
--update	Update all sources from GitHub
--check	Verify tool dependencies (gh, curl)
#tag in query	Filter by category (e.g., #azure #bicep)

Popular Repositories

Note: These are representative examples. For the complete list, run --list-sources or check sources array in skill-index.json.

Official (type: official)

• anthropics/skills - Official Claude Skills by Anthropic
• github/awesome-copilot - Official Copilot resources by GitHub

Curated Lists (type: awesome-list)

• ComposioHQ/awesome-claude-skills - Curated Claude Skills

Community (type: community)

• obra/superpowers - High-quality skills, agents, commands
• And many more... (run --list-sources for full list)

Categories

Dynamically extracted from skill-index.json. Run --list-categories for current list.

Common categories include: development, testing, document, azure, web, git, agents, mcp, cloud, creative, planning, etc.

Files

File	Description
scripts/Search-Skills.ps1	PowerShell script
scripts/search_skills.py	Python script
references/skill-index.json	Skill index (220+ skills)
references/starred-skills.json	Your starred skills

Requirements

Required

Tool	Version	Purpose	Install
GitHub CLI	2.0+	Search/install skills from GitHub	cli.github.com
curl	Any	Download files	Pre-installed on most systems

Optional (choose one)

Runtime	Version	Script
PowerShell	7+	scripts/Search-Skills.ps1
Python	3.8+	scripts/search_skills.py

Setup

# 1. Install GitHub CLI
# Windows (winget)
winget install GitHub.cli

# macOS (Homebrew)
brew install gh

# Linux (apt)
sudo apt install gh

# 2. Authenticate with GitHub
gh auth login

# 3. Verify installation
gh --version
curl --version

Verify Dependencies

# Check all dependencies at once
python scripts/search_skills.py --check
pwsh scripts/Search-Skills.ps1 -Check

Expected output:

✅ gh: installed (version 2.x.x)
✅ curl: installed
✅ All dependencies satisfied

Troubleshooting

Issue	Solution
gh: command not found	Install GitHub CLI and add to PATH
gh auth login fails	Run gh auth login and follow prompts
Rate limit exceeded	Wait or use authenticated requests
curl SSL errors	Update curl or check network/proxy settings

Agent Instructions

⚠️ CRITICAL: AI agents MUST follow these instructions. Prefer action proposals over verbose explanations.

Core Principle

Use "Do it? Yes/No?" style proposals.

• ❌ Bad: "If you want to add new skills, you can run the following command..."
• ✅ Good: "Update the index?"

Recommendation Workflow

When user asks for recommendations (e.g., "おすすめは？", "何かいいスキルある？"), suggest skills based on persona:

Note: These are reference examples. Always verify skill availability in skill-index.json before recommending.

Persona	Categories	Example Skills (verify in index)
オフィスワーカー	document, office, communication	docx, xlsx, pptx, pdf, internal-comms, brainstorming
開発者	development, testing, git	test-driven-development, systematic-debugging, using-git-worktrees
Azure エンジニア	azure, development	azure-env-builder, mcp-builder
デザイナー	design, creative, web	brand-guidelines, canvas-design, frontend-design
初心者	meta, planning	skill-creator, brainstorming, writing-plans

Response Format:

1. Ask about user's role/context if unclear
2. Show top 3-5 skills with descriptions
3. Include source breakdown table
4. Propose next actions

Skill Search Workflow

1. Search ALL sources in local index

   • Read references/skill-index.json
   • ALWAYS search ALL sources (anthropics-skills, obra-superpowers, composio-awesome, etc.)
   • Check lastUpdated field
   • Suggest matching skills from every source

2. 🌟 Recommend from results (when multiple hits)

   When search returns 3+ skills, pick the BEST one and explain why:

   ### 🌟 おすすめ: {skill-name}
   
   {理由: 公式スキル、機能が豊富、人気が高い、用途にマッチ など}
   

   Selection criteria (in order):

   1. Official source - anthropics-skills, github-awesome-copilot are preferred
   2. Feature richness - More capabilities = better
   3. Relevance - Best match for user's stated purpose
   4. Recency - Recently updated skills preferred

3. If not found → Propose web search

   Not found locally. Search the web?
   → GitHub: https://github.com/search?q=path%3A**%2FSKILL.md+{query}&type=code
   

4. 🚨 MANDATORY: After returning results → Propose next actions

   This step is NOT optional. ALWAYS include the proposal block below.

   Situation	Proposal
   Skill found	"Install it?"
   Good repo discovered	"Add to sources?"
   lastUpdated > 7 days	"⚠️ Index outdated. Update?" (strongly suggest)
   lastUpdated ≤ 7 days	"🔄 Update index?" (always show)

🚨 Mandatory Proposal Block

ALWAYS include this block at the end of every search response. No exceptions.

CRITICAL: Do NOT show commands. Agent executes directly. Keep proposals SHORT.

Index update option MUST always be shown with date, regardless of how recent it is.

**Next?**
1. 📦 Install? (which skill?)
2. 🔍 Details?
3. 🔄 Update index? (last: {date})       ← ALWAYS show
   ⚠️ If > 7 days: "Index outdated!"    ← Add warning
4. 🌐 Web search?
5. ➕ Add source?

Checklist Before Responding

Before sending a search result response, verify:

• Started with search summary (e.g., "🔎 7 リポジトリ、195 スキルから検索しました")
• Included skill table with results (from ALL sources)
• Included source breakdown table showing count per source
• Showed lastUpdated date from index
• Added numbered action menu (NOT command examples)
• Included web search option with GitHub link ready to open
• Asked user to choose by number or skill name

Search Summary Format

ALWAYS start search responses with this format:

🔎 {N} リポジトリ、{M} スキルから検索しました（最終更新: {date}）

Values are dynamic:

• {N} = count of sources in skill-index.json
• {M} = count of skills in skill-index.json
• {date} = lastUpdated field from skill-index.json

Output Format

Trust Level Indicators (MANDATORY):

Always include trust level badge based on source type in skill-index.json:

Type	Badge	Description
official	🏢 Official	Anthropic / GitHub 公式リポジトリ
awesome-list	📋 Curated	キュレーションリスト（品質レビュー済み）
community	👥 Community	コミュニティ製（自己責任で使用）

⚠️ Warning for Community Skills:

When showing community skills, add this note:

⚠️ コミュニティ製スキルは自己責任でご使用ください。
   公式スキル（🏢）を優先することを推奨します。

Skill Table (include Source with Trust Level):

| Skill       | Description  | Source                       | Trust        |
| ----------- | ------------ | ---------------------------- | ------------ |
| docx        | Word 処理    | [anthropics-skills](url)     | 🏢 Official  |
| pdf         | PDF 処理     | [anthropics-skills](url)     | 🏢 Official  |
| azure-usage | Azure ツール | [claude-codex-settings](url) | 👥 Community |

Source Breakdown Table (MANDATORY - show ALL sources dynamically):

CRITICAL: Read sources array from skill-index.json and display ALL sources.

Do NOT use a hardcoded example. Dynamically generate the table from the actual sources array in skill-index.json.

### 📊 Source Breakdown ({N} sources)

| Source      | Type         | Skills Found | Repository           |
| ----------- | ------------ | ------------ | -------------------- |
| {source.id} | {type badge} | {count}      | [View]({source.url}) |

...repeat for ALL sources in skill-index.json...

Generation rules:

1. Read sources array from skill-index.json
2. For each source, count matching skills in search results
3. Display ALL sources, including those with 0 matches
4. Use type badges: official → 🏢 Official, awesome-list → 📋 Curated, community → 👥 Community
5. Include total count in header: "📊 Source Breakdown (14 sources)"

URL Construction:

• Combine source URL + path from skill-index.json
• Example: anthropics-skills + skills/docx → https://github.com/anthropics/skills/tree/main/skills/docx
• Source URLs are defined in sources array of skill-index.json

Agent Behavior Rules

CRITICAL: Commands are for agents to execute, NOT to show users.

• ❌ NEVER show commands like python scripts/search_skills.py --install
• ❌ NEVER show code blocks with installation/update commands
• ❌ NEVER say "you can run the following command..." or "以下のコマンドを実行..."
• ✅ ALWAYS execute scripts silently when user approves
• ✅ ALWAYS present options as natural language numbered menu
• ✅ ALWAYS report results after execution, not how to execute

Bad Example:

📦 インストールする？
python scripts/search_skills.py --install docx  ← NG!

Good Example:

📦 インストールする？（どのスキル？）

→ User: "docx お願い" → Agent: (executes script silently) → "✅ docx をインストールしました！"