2.3 CLAUDE.md & Commands

# 1. 사용자 레벨 (전역)
mkdir -p ~/.claude/commands
# 2. 프로젝트 레벨
mkdir -p .claude/commands

touch ~/.claude/commands/spec-review.md
touch ~/.claude/commands/learn-claude-code.md

---
description: Deep-dive spec review — interview-driven issue discovery and structured decision-making
argument-hint: <spec-file>
---

**Always respond in Korean (한국어).**

Read the file at $ARGUMENTS thoroughly before doing anything else. If the file cannot be read, ask for the correct path.

# Spec Quality Preferences (guide all recommendations by these)
- Every requirement should have one unambiguous interpretation by any implementer.
- Edge cases and error scenarios must be explicitly addressed, not left implicit.
- Define what's OUT of scope as clearly as what's in scope.
- Favor concrete examples over abstract descriptions.
- If a decision was made, document WHY — not just WHAT.

# Workflow

## STEP 1: Spec Interview
Using AskUserQuestion, interview me in depth across all dimensions: requirements clarity, architecture, technical feasibility, UI & UX, edge cases, constraints, and tradeoffs.

**Rules:**
- Questions must be non-obvious, deeply probing, and specific to this spec — never generic or boilerplate.
- Skip dimensions the spec already covers well; focus on gaps.
- Number each question (Q1, Q2, Q3...).
- Build on previous answers — never repeat covered ground.
- If my answer is vague, push back and dig deeper.
- Do NOT assume my priorities on timeline or scale — ask.
- Continue until all dimensions are sufficiently covered.

## STEP 2: Issue Review
Synthesize interview findings into issues, organized by section:
1) Requirements Clarity — ambiguous language, multiple interpretations, missing acceptance criteria
2) Completeness — missing edge cases, error scenarios, boundary conditions, undefined behaviors
3) Architecture & Feasibility — technical risks, unrealistic constraints, unresolved dependencies
4) Scope & Boundaries — unclear in/out of scope, scope creep risks, undefined limits
5) UX & User Flows (if applicable) — incomplete flows, missing error/loading/empty states

**Issue format:**
- Number issues (1-1, 1-2 = section 1, issue 1 and 2).
- Describe the problem concretely with spec references.
- Present 2-3 options with LETTERS (A, B, C). Include "do nothing" where reasonable.
- For each option: effort to spec out, risk if left unaddressed, impact on implementation.
- **Recommended option is always A**, with reasoning mapped to my preferences above.
- In AskUserQuestion, label as `"1-1A: recommended approach", "1-1B: alternative", "1-1C: do nothing"`.

**Pause after each section** — use AskUserQuestion for feedback before moving to the next.

## STEP 3: Spec Update
Once all decisions are made:
- Edit the original spec file in-place with all decisions, clarifications, and new details.
- Add or refine sections as needed (edge cases, error scenarios, scope boundaries, constraints, etc.).
- Mark open items with `> [!WARNING] OPEN ITEM:` callouts.
- Do NOT remove existing content unless explicitly agreed.
- Provide a brief summary of what changed.

**Reminder: All responses must be in Korean (한국어).**

---
description: Claude Code 기능과 사용법을 학습하는 인터랙티브 가이드
argument-hint: <질문>
---

**Always respond in Korean (한국어).**

You are a Claude Code instructor helping the user learn Claude Code features and best practices.

## STEP 1: Answer the Question
Use the Task tool with `subagent_type='claude-code-guide'` to answer: $ARGUMENTS

Present the answer clearly with concrete examples.

**Important:** The claude-code-guide subagent occasionally returns outdated or inaccurate information. If the user challenges your answer or asks follow-up questions, verify by fetching the official documentation directly:

Base URL: `https://code.claude.com/docs/en/`

Key pages (use WebFetch on the relevant one):
- `overview` — Core concepts and getting started
- `settings` — Configuration, permissions, environment variables
- `memory` — CLAUDE.md files and memory hierarchy
- `slash-commands` — Skills, slash commands, SKILL.md
- `subagents` — Built-in and custom subagents
- `hooks` — Event-driven automation
- `mcp` — Model Context Protocol setup
- `common-workflows` — Practical usage patterns
- `best-practices` — Official recommendations
- `model-config` — Model selection and configuration
- `features-overview` — Extension system overview
- `output-styles` — Output style customization
- `plugins-reference` — Plugin system

When the official docs contradict the subagent's answer, always prefer the official docs and correct the information.

## STEP 2: Hands-On Quiz
After answering, use AskUserQuestion to give the user a practical quiz:
- The quiz should test whether they can **actually use** the feature, not just recall facts.
- Frame it as a concrete action they can try right now in their terminal.
- Include 2-3 options where one is correct and others represent common mistakes.

Example format:
> Q: CLAUDE.md를 사용자 레벨에 추가하려면 어디에 파일을 만들어야 할까요?
> A) ./CLAUDE.md
> B) ~/.claude/CLAUDE.md
> C) ~/.config/claude/CLAUDE.md

After they answer, confirm whether correct, explain why, and offer to go deeper or move to the next topic.

**Reminder: All responses must be in Korean (한국어).**