VoltAgent Subagent Directory: Choosing and Installing Claude Code Subagents

## What This Is / What Problem It Solves Claude Code's subagents draw their power from 'handing the right task to the right specialist,' but writing a good agent from scratch (role, tool permissions, trigger conditions) has a learning curve and easily leads to reinventing the wheel. VoltAgent/awesome-claude-code-subagents is a curated 'awesome list' containing 154+ ready-made, categorized Claude Code subagents, letting you 'pick first, then lightly customize' instead of starting from a blank page. This article organizes it into a 'how to pick + how to install' method, along with a fill-in-the-blank standard subagent template. ## Why This Source Is Worth Using VoltAgent is an active open-source organization building AI agent frameworks, and this list is MIT-licensed (commercial use allowed), organized into 10 major categories: (1) Core Development (2) Language Specialists (3) Infrastructure (4) Quality & Security (5) Data & AI (6) Developer Experience (7) Specialized Domains (8) Business & Product (9) Meta & Orchestration (10) Research & Analysis. Every subagent follows Claude Code's official 'YAML frontmatter + Markdown system prompt' format, so once you pick one it's nearly plug-and-play. For beginners, it doubles as both a directory and a how-to-write guide. ## How to Choose (Four Selection Questions) 1. Which category does the task fall under? Start by locating it in the 10 categories above (security review -> Quality & Security; writing in a specific language -> Language Specialists). 2. Are the trigger conditions clear? Check the agent's `description` field in the frontmatter — this is what Claude Code uses to auto-delegate, so the closer it matches your actual use case, the better. 3. Are tool permissions minimized? `tools:` should only grant what's necessary (don't give Write/Edit to a pure reviewer) — this is both a security control and a quality control. 4. Is the model routing sensible? The convention is opus for heavy reasoning (architecture, security), sonnet for everyday coding, and haiku for quick, small tasks. ## How to Install - Global (all projects): place it at `~/.claude/agents/<name>.md`. - Project-specific: place it at `.claude/agents/<name>.md` — a project-level file overrides the global one with the same name. - Multiple installation paths: copy the file directly, use the repository's install script, or install through the Claude Code plugin marketplace. - Usage: once in place, Claude Code auto-delegates based on the description; to invoke manually, say 'use the <name> subagent to handle this.' Subagents run in an isolated context window and won't pollute the main conversation. ## When to Use It When you're just starting to build your own subagent library (copy curated ones first, then customize), when you want a ready-made specialist for a recurring task (DB migration review, API design, documentation generation), or when you want to learn what a well-built subagent looks like. 📎 Source: VoltAgent/awesome-claude-code-subagents (by VoltAgent, MIT license) — this article is a Traditional Chinese adaptation; the template is a fill-in-the-blank version rewritten to match the official format, not a verbatim copy of repository entries. See the link above for the original list.

填下面的欄位，上方 prompt 會即時替換 [方括號] 內容。填好後按「複製組好的 prompt」直接丟進工具。

---
name: {{AGENT_NAME}}
description: >-
  {{WHEN_TO_USE}}。
  描述要寫成「何時該委派給這個代理」——這段是 Claude Code 自動路由的依據，
  寫得越具體、觸發越準（例如「在使用者要審查 SQL migration 或資料庫 schema 變更時使用」）。
tools: Read, Grep, Glob, Edit, Bash
model: sonnet
---

你是一位 {{ROLE}}，專精 {{EXPERTISE}}。

## 你的職責
- {{RESPONSIBILITY_1}}
- {{RESPONSIBILITY_2}}
- {{RESPONSIBILITY_3}}

## 工作流程
1. 先釐清任務範圍與既有專案慣例，不要套用與本專案無關的偏好。
2. 用 Read / Grep / Glob 收集足夠脈絡再動手；需要時用 Bash 驗證（跑測試、查狀態）。
3. 產出時標明：你做了什麼、為什麼、影響哪些檔案。
4. 不確定的地方明說「需確認」，不要編造。

## 邊界
- 只在 {{SCOPE}} 範圍內行動。
- 完成定義（Definition of Done）：{{DONE_CRITERIA}}。

專案脈絡：{{PROJECT_CONTEXT}}