Agent/Skill 模板搜尋策略
Provides 20 search queries and heuristics to find AI agent/skill template files on GitHub and the web.
What this file does
Provides 20 search queries and heuristics to find AI agent/skill template files on GitHub and the web.
When to use it
- Building a library of reusable agent or skill templates
- Searching for structured code review or security agent definitions
- Validating whether a found file is a template or an article
- Creating your own agent template and wanting to see existing patterns
Agent/Skill 模板搜尋策略
基於 code-review.agent.md 的特徵,搜尋類似的軟體工程 agent/skill 指令檔模板。
📋 已知檔案特徵
- 格式: Markdown 檔案 (
.md) - 結構: YAML frontmatter + Markdown 內容
- Frontmatter 欄位:
name: agent 名稱agent: 類型標記description: 功能描述
- 內容特徵:
- 角色定義(Role)
- 審查/分析面向(多個維度)
- 結構化輸出格式
- 可能包含變數/參數(如
${input:...})
🔍 1. Google/Bing 搜尋查詢(10 組)
1. site:github.com "---" "name:" "agent:" "description:" filetype:md
2. site:github.com ".agent.md" "code review" OR "code-review"
3. site:github.com "copilot" "agent" "frontmatter" "---"
4. site:github.com ".skill.md" "YAML frontmatter" "role"
5. site:github.com "claude" "sub-agent" ".md" "---"
6. site:github.com "prompt template" "software engineering" "---" filetype:md
7. site:github.com ".github/agents" OR ".github/prompts" "*.agent.md"
8. site:github.com "AI agent template" "security" "performance" "architecture"
9. site:github.com "copilot instructions" ".md" "frontmatter"
10. site:github.com "agent-based" "code analysis" "template" "markdown"
🐙 2. GitHub Code Search 查詢(10 組)
使用新版 GitHub Code Search 語法:
1. path:**.agent.md "---" "name:" "description:"
2. path:**/.github/agents/ language:markdown "---"
3. filename:*.agent.md "role" "output format"
4. path:**/prompts/ "---" "agent:" extension:md
5. "name: code-review" OR "name: security-scan" path:**.md
6. path:**.skill.md "frontmatter" "YAML"
7. "Critical Issues" "Suggestions" "Good Practices" path:**.agent.md
8. path:**/.github/ filename:*.agent.md OR filename:*.skill.md
9. "${input:" path:**.md "---"
10. "separation of concerns" "design pattern" path:**agent**.md
GitHub 進階搜尋組合
repo:github/copilot* path:**.agent.md
org:microsoft path:**agents/ extension:md
user:anthropic "claude" ".md" "---"
stars:>100 "AI agent" "template" extension:md
🔤 3. 關鍵字同義詞
| 類別 | 同義詞 |
|---|---|
| Agent 相關 | agent, skill, subagent, sub-agent, capability, tool, function |
| Prompt 相關 | prompt, instruction, system-prompt, system_prompt, directive |
| 模板相關 | template, scaffold, boilerplate, blueprint, schema |
| 配置相關 | config, configuration, manifest, definition, spec |
| 文檔類型 | .agent.md, .skill.md, .prompt.md, .instruction.md |
| 框架相關 | copilot, claude, cursor, cody, aider, chatgpt |
組合搜尋詞
- "agent template"
- "skill definition"
- "prompt scaffold"
- "AI instruction template"
- "code review agent"
- "analysis skill"
- "engineering agent"
✅ 4. 判斷「真的是模板」的特徵
🟢 肯定特徵 (這是模板)
-
結構化 Frontmatter
- 有 YAML frontmatter (
---開頭結尾) - 包含 metadata 欄位 (name, description, author, version 等)
- 有 YAML frontmatter (
-
角色定義區塊
- 明確的 "Role" 或 "角色" 段落
- 使用第二人稱 ("你是...", "You are...")
-
系統化內容結構
- 有明確的章節劃分 (## 標題)
- 包含「分析面向」、「審查項目」、「檢查清單」等結構
-
可重用的輸出格式
- 定義標準化輸出格式
- 使用表情符號或標記 (🔴, 🟡, ✅)
-
參數化設計
- 包含變數佔位符 (
${input:...},{{variable}},[PLACEHOLDER]) - 可以替換的參數區塊
- 包含變數佔位符 (
-
檔案命名模式
.agent.md,.skill.md,.prompt.md- 檔案位於特定目錄 (
.github/agents/,prompts/,.claude/)
🔴 否定特徵 (不是模板,是文章/引用)
-
敘述性內容
- 是在「介紹」或「討論」agent,而非定義 agent
- 包含 "如何使用", "教學", "範例說明" 等字眼
-
文章結構
- 有作者署名、發布日期
- 包含「前言」、「結論」、「延伸閱讀」
-
程式碼範例
- 檔案中主要是程式碼,而非指令
- 是在展示「如何呼叫 agent」,不是定義 agent 本身
-
README 或文檔
- 檔案名稱為 README.md
- 是專案說明文件,內含多個 agent 的引用
-
外部引用
- 包含 "參考來源", "引用自", "inspired by"
- 是集合/列表文件 (如 "awesome-agents")
🔍 快速驗證方法
✓ 檢查前 20 行是否有 YAML frontmatter
✓ 搜尋 "你是" 或 "You are" (角色定義)
✓ 檢查是否有 ${input:} 或類似變數
✓ 檔案路徑包含 agents/, prompts/, .claude/
✗ 包含 "How to", "Tutorial", "Guide", "Example"
✗ 有多個 agent 並列說明 (代表是列表文件)
🎯 5. 推薦搜尋路徑
優先搜尋的 GitHub Repository
github/copilot-*
anthropic/claude-*
microsoft/vscode-*
cursor-ai/*
sourcegraph/cody-*
特定目錄
.github/agents/
.github/prompts/
.claude/
prompts/
agents/
skills/
.copilot/
📊 6. 執行順序建議
-
第一階段: GitHub Code Search (最精準)
- 使用前 5 組 Code Search 查詢
- 重點:path 和 filename 過濾
-
第二階段: Google/Bing (擴大範圍)
- 使用 site:github.com 限定
- 重點:特定關鍵字組合
-
第三階段: 手動驗證
- 檢查搜尋結果是否符合「模板特徵」
- 排除文章、教學、README
-
第四階段: 交叉驗證
- 查看找到的 repository 是否有更多類似檔案
- 追蹤 forks 和 related repositories
📝 範例比對
✅ 這是模板
---
name: security-audit
agent: security
description: 執行安全性稽核
---
## 角色
你是資深安全工程師...
## 審查項目
1. 輸入驗證
2. 權限控制
...
❌ 這不是模板(是文章)
# How to Use Code Review Agents
In this tutorial, we'll learn how to use agents...
Here's an example of a code-review agent:
[引用某個 agent 內容]
🚀 下一步
執行搜尋後,建議:
- 記錄找到的模板數量與品質
- 分類不同類型的 agent (code-review, security, performance, testing...)
- 分析共通模式,提煉最佳實踐
- 建立自己的模板庫
建立日期: 2025-12-28 用途: 搜尋 GitHub 上的 AI Agent/Skill 模板檔案
What's inside
6 sections: 20 search queries, keyword synonyms, template validation criteria, recommended repos, execution order, and example comparisons.
Change this for your project
- Replace
github/copilot-*with the organization or user you want to search - Replace
anthropic/claude-*with the organization or user you want to search - Replace
microsoft/vscode-*with the organization or user you want to search
Where it goes
Keep it in your repository where the agent or team that needs it will read it.
Worth borrowing
- Frontmatter with name, agent, and description fields for metadata
- Positive and negative feature lists to classify search results
- Multi-phase search strategy: code search first, then web, then manual verification
Related Documents
AI Trending Bot Implementation Plan
Defines a 5-stage pipeline to collect, score, analyze, report, and deliver trending AI content from 6 platforms using Python and Claude CLI.
template
Turns the AI into a Socratic tutor that guides study sessions without giving answers or doing homework.
Spec Template
Provides a front-matter schema and writing rules for Brim language specification documents, covering status lifecycle and content conventions.