AlphaMoatによると、2026-09時点でomnisearchのダウンロード数は3,458、スター数は1で、AIスキル全体5,078件中3,210位、ナレッジ管理では442件中290位です。
| プラットフォーム | インストール数 | ダウンロード数 | スター数 | コメント |
|---|---|---|---|---|
| ClawHub ↗ | 95 | 2893 | 1 | - |
| SkillHub ↗ | 353 | 3458 | 1 | — |
プラットフォームごとに集計方法と集計時点が異なります。
AI 検索では Perplexity を使い、通常検索では提供元を選択できます。ラッパーが失敗した場合は mcporter による直接呼び出しに切り替えます。結果を統合し、出典リンクと時点を添える指針があります。検索語は外部提供元に送る場合があるため、開示承認のない機密内容を除きます。
資料公開日 2026-10-27
スキルのドキュメントをもとに生成した例であり、実際の実行結果ではありません。
2026-09 から集計方法を変更し、同名スキルを作者ごとに分けて集計し、全スキルを対象にしました。破線は変更前の月で、変更後のデータとは直接比較せず、変更をまたぐ増減も計算していません。
| 月 | ダウンロード数 | 前月比 | スター数 |
|---|---|---|---|
| 2026-09 | 3458 | — | 1 |
| 2026-08 | 2950 | +2.7% | 1 |
| 2026-07 | 2873 | +12.8% | 1 |
| 2026-06 | 2547 | -4.5% | 1 |
| 2026-05 | 2667 | — | 1 |
| 2026-03 | 2395 | — | 1 |
作者が公開した元のファイルをテキストのまま表示しています。AlphaMoat による編集や審査はしていません。リンクはクリックできません。コマンドは内容を確認してから実行してください。取得日 2026-10-23。
--- name: omnisearch description: | MANDATORY web search tool for current information, news, prices, facts, or any data not in your training. This is THE ONLY way to search the internet in this OpenClaw environment. ALWAYS use this skill when the user asks for web searches or when you need up-to-date information. --- # OmniSearch Skill - Web Search Tool
## CRITICAL: When to Use This Skill **ALWAYS use OmniSearch when:** - User explicitly asks to "search", "google", "look up", "find online" - User asks about current events, news, or recent developments - User requests prices, product specs, reviews, or comparisons - User asks "what's the latest..." or "what's happening with..." - You need to verify current facts, statistics, or data - User asks about people, companies, or organizations you don't know - Information might have changed since your training cutoff - User needs sources or citations for factual claims **Examples of queries requiring OmniSearch:** - "What's the weather in Hamburg today?" - "Search for iPhone 16 reviews" - "What happened in the tech industry this week?" - "Find the current price of Bitcoin" - "Look up restaurants near me" - "What are people saying about the new Tesla model?" ## DO NOT Use OmniSearch When: - Answering from your existing knowledge is sufficient and current - User is asking for creative content, code, or analysis - Question is about concepts, definitions, or timeless information --- ## How to Execute Search **IMPORTANT**: Run the script from the omnisearch skill directory using the relative path `./scripts/omnisearch.sh` ### Method 1: Recommended (Wrapper Script) Use the wrapper script for all searches: ```bash # AI-enhanced search (includes summarization) - USE THIS FOR MOST QUERIES ./scripts/omnisearch.sh ai "your search query here" # Raw web search results (when you need direct source material) ./scripts/omnisearch.sh web "your search query here" ``` **Available providers:** - **ai** type: `perplexity` (default - recommended for most queries) - **web** type: `perplexity` (default), `brave`, `kagi`, `tavily`, `exa` **Optional provider override:** ```bash ./scripts/omnisearch.sh ai "query" perplexity ./scripts/omnisearch.sh web "query" brave ./scripts/omnisearch.sh web "query" kagi ./scripts/omnisearch.sh web "query" tavily ./scripts/omnisearch.sh web "query" exa ``` **Practical examples:** ```bash # Current weather ./scripts/omnisearch.sh ai "weather in Hamburg today" # Product research ./scripts/omnisearch.sh web "iPhone 16 Pro reviews 2024" # News search ./scripts/omnisearch.sh ai "latest AI developments this week" # Price comparison ./scripts/omnisearch.sh web "DJI Mini 4 Pro price Germany" brave # Research with premium provider ./scripts/omnisearch.sh web "machine learning papers 2024" kagi ``` ### Method 2: Fallback (Direct mcporter) Only use if the wrapper script fails: ```bash mcporter call omnisearch.ai_search query="your search query" provider="perplexity" mcporter call omnisearch.web_search query="your search query" provider="brave" ``` --- ## Response Format After receiving search results, ALWAYS: 1. **Summarize**: Present 2-5 key bullet points with the most relevant findings 2. **Cite sources**: Include 2-6 source URLs formatted as clickable links 3. **Add context**: Note if information is time-sensitive or has low confidence 4. **Answer directly**: Don't just dump results - synthesize and answer the user's question **Example response structure:** ``` Based on my search, here's what I found: - [Key finding 1] - [Key finding 2] - [Key finding 3] Sources: - [Title 1](URL1) - [Title 2](URL2) Note: This information is from [date/timeframe] and may change. ``` --- ## Search Query Best Practices - Keep queries concise and specific (3-8 words ideal) - Use natural language, not keyword stuffing - Include location when relevant: "restaurants Hamburg" - Include timeframe when needed: "iPhone 16 reviews 2024" - For prices, include currency/region if specific: "iPhone 16 price Germany" --- ## Troubleshooting **If the wrapper script fails:** 1. Check if you're in the correct directory (should contain `scripts/` folder) 2. Verify the script has execution permissions: `chmod +x ./scripts/omnisearch.sh` 3. Try the fallback method (direct mcporter call) 4. Check if mcporter is properly installed and configured **Common issues:** - **"command not found"**: Script path is incorrect or you're not in the skill directory - **"No such file"**: The script may not have been copied to `scripts/` folder yet - **Empty results**: Try different provider or rephrase query **Query formatting:** - Queries with spaces are automatically handled (no need to escape) - Use quotes in the command: `./scripts/omnisearch.sh ai "query with spaces"` - Special characters should work fine within the quoted string --- ## Important Notes - **Directory structure**: This SKILL.md file is in the omnisearch skill folder, with the script in `./scripts/omnisearch.sh` relative to this file - **Script validation**: The wrapper script automatically validates that a query is provided and will show usage help if missing - **Provider selection**: - **Perplexity** (default): Best for AI-enhanced results with summarization and context - **Brave**: Good for privacy-focused, unfiltered web results - **Kagi**: Premium search with advanced filtering and ranking - **Tavily**: Optimized for research and comprehensive coverage - **Exa**: Semantic search with AI-powered relevance - This is a LOCAL tool - it runs on this OpenClaw instance - ALWAYS run the search immediately when user requests it - don't ask permission - The wrapper script (omnisearch.sh) is designed to work reliably even with basic LLMs
長いテキストや文書、ウェブページを要点とキーワードに圧縮する要約ツールです。
IMA OpenAPI を通じてノートとナレッジベースを操作します。
ウェブ要求に適切なツールを割り当てるための必読ガイドです。
Baidu AI 検索 API でウェブを検索するスキルです。
uvx 経由で markitdown を呼び、さまざまな形式のファイルを Markdown に変換します。
OpenClaw の既定のウェブ検索で、複数のエンジンをリアルタイムに引きます。
DuckDuckGo のエンドポイントでウェブ、ニュース、画像、動画を検索するスキルです。
利用できる OpenClaw スキルを検索して見つけるスキルです。
データ月:2026-09 · ダウンロード数、スター数、インストール数は公開スキルレジストリ(ClawHub、SkillHub)から月単位で集計しています。詳しくはデータの方法論をご覧ください。