# Agent Role Guide

> Use when the user wants clear sub-agent role separation for project work, especially around planning, design, development, and QA responsibilities. This skill defines what each role owns, which built-in agent type to use for each role, how to split work cleanly, and how to avoid overlap between parallel agents.

- Skill: `kim-dh2417/agent-role-guide` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add kim-dh2417/agent-role-guide`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kim-dh2417/agent-role-guide/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: kim-dh2417 (https://skillmd.com/u/kim-dh2417)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/kim-dh2417/agent-role-guide

---


# Agent Role Guide

## Overview

Use this skill to organize sub-agents into four project roles: planning, design, development, and QA.

The goal is to keep responsibilities clear, reduce duplicated work, and make delegation easier in multi-step projects.

## Role Model

### 1. Planning Agent

- Purpose:
  - requirements review
  - scope clarification
  - policy and workflow definition
  - missing requirement detection
- Recommended built-in type:
  - `default` for reasoning-heavy planning
  - `explorer` for document lookup and codebase impact scan
- Typical outputs:
  - requirement summary
  - changed rules
  - open questions
  - scope boundaries

### 2. Design Agent

- Purpose:
  - IA review
  - menu and route structure review
  - screen design and user flow alignment
  - wireframe or prototype interpretation
- Recommended built-in type:
  - `explorer`
- Typical outputs:
  - screen mapping
  - menu coverage check
  - route plan
  - design mismatch notes

### 3. Development Agent

- Purpose:
  - frontend or backend code implementation
  - refactor
  - route/component/service creation
  - test fixes tied to implementation
- Recommended built-in type:
  - `worker`
- Typical outputs:
  - changed files
  - implementation summary
  - verification result
  - unresolved blockers

### 4. QA Agent

- Purpose:
  - compare implementation against docs
  - browser walkthrough
  - missing or partial feature detection
  - verification of expected flows
- Recommended built-in type:
  - `explorer` for review and coverage analysis
  - `worker` only if this role also owns test execution or automated verification scripts
- Typical outputs:
  - findings list
  - missing screens/features
  - browser notes
  - residual risks

## Delegation Rules

Use this split:

- planning owns rules and scope
- design owns IA, menus, routes, screens
- development owns code changes
- QA owns verification and findings

Do not overlap ownership:

- do not have planning and design edit the same artifact blindly
- do not have QA re-implement development work
- do not have multiple development agents edit the same files unless explicitly coordinated

## Recommended Order

For most projects:

1. planning agent
2. design agent
3. development agent
4. QA agent

For faster iterations:

- planning and design can run in parallel if the scope is already clear
- development starts after planning/design direction is stable
- QA starts when enough implementation exists to verify

## Example Mapping

- `기획 검토 에이전트` -> `default` or `explorer`
- `디자인 검토 에이전트` -> `explorer`
- `프론트 개발 에이전트` -> `worker`
- `백엔드 개발 에이전트` -> `worker`
- `기능 리뷰 에이전트` -> `explorer`
- `테스트/QA 에이전트` -> `explorer` or `worker`

## Notes

- Use this skill as an operating guide for delegation.
- This skill does not replace project-specific skills such as frontend, backend, prototype, or review skills.