Claude Code Custom Commands Guide

Claude Code supports custom slash commands that let you quickly run common operations. This chapter covers practical commands and how to use them.

---

What Are Custom Commands

Custom commands are Markdown files stored in .claude/commands/. You invoke them with /<command-name>.

Command File Structure

---
description: Command description (shown in command list)
argument-hint: [parameter hint]
allowed-tools: Read, Edit, Bash(git:*)
---

# Command Title

This is the command's prompt content...
Use $ARGUMENTS to reference user-provided arguments

Command File Location

.claude/
└── commands/
    ├── think-harder.md
    ├── reflection.md
    └── gh/
        ├── review-pr.md
        └── fix-issue.md

---

Analysis & Thinking Commands

/think-harder - Enhanced Deep Analysis

For deep analysis of complex problems.

Usage:

/think-harder How should I design a high-concurrency order system?

Core flow:

1. Problem clarification - Define core problem, identify hidden assumptions
2. Multi-dimensional analysis
   • Structural decomposition: break into base components and dependencies
   • Stakeholder perspectives: consider all affected parties
   • Temporal analysis: short-term vs long-term impact
   • Causal reasoning: map cause-effect chains and feedback loops
3. Critical evaluation - Challenge assumptions, identify cognitive biases
4. Synthesis - Connect cross-domain insights

Output format:

• Problem reframe
• Key insights
• Reasoning chain
• Alternative approaches
• Uncertainties
• Actionable recommendations

---

/think-ultra - Ultra-Deep Analysis

A deeper pass than /think-harder. For extremely complex problems.

Usage:

/think-ultra How would you build an AI programming assistant from scratch?

---

/reflection - Instruction Reflection

Analyzes the current session and improves instructions in CLAUDE.md.

Usage:

/reflection

Core flow:

1. Analysis phase - Review conversation history, check current CLAUDE.md
2. Identify issues
   • Inconsistencies in Claude's responses
   • Misunderstandings of user requests
   • Areas where more detail could help
3. Interactive phase - Suggest improvements, wait for confirmation
4. Implementation phase - Modify CLAUDE.md

Best practices:

• Use TodoWrite to track analysis progress
• Read current CLAUDE.md thoroughly before modifying
• Consider edge cases and common scenarios

---

/reflection-harder - Deep Session Analysis

End-to-end analysis of the entire session, capturing learnings and improvement points.

Usage:

/reflection-harder

---

/eureka - Record Technical Breakthroughs

Use this when you have an important technical insight worth saving.

Usage:

/eureka Discovered that React 18 Suspense perfectly solves the data-fetching waterfall problem

---

GitHub Integration Commands

/gh:review-pr - PR Code Review

Auto-fetches PR details and runs a code review.

Usage:

/gh:review-pr 123

Core flow:

1. Fetch PR details: gh pr view 123
2. Get code diff: gh pr diff 123
3. Analyze changes and provide review feedback
4. Post comments directly on GitHub

Review focus:

• Code correctness
• Project convention compliance
• Performance impact
• Test coverage
• Security considerations

Comment format:

# Get commit ID
gh api repos/OWNER/REPO/pulls/123 --jq '.head.sha'

# Post comment
gh api repos/OWNER/REPO/pulls/123/comments \
    --method POST \
    --field body="Suggestion: consider using useMemo here for optimization" \
    --field commit_id="abc123" \
    --field path="src/components/List.tsx" \
    --field line=42 \
    --field side="RIGHT"

---

/gh:fix-issue - Issue Fix Workflow

Complete Issue fix workflow.

Usage:

/gh:fix-issue 456

Core flow:

1. Fetch Issue details
2. Analyze root cause
3. Create fix branch
4. Implement fix
5. Create PR linked to Issue

---

Utility Commands

/translate - Technical Translation

Translates English/Japanese technical content to Chinese.

Usage:

/translate This article explains the fundamentals of RAG...

Features:

• Preserves technical terms
• Maintains code formatting
• Proper technical expression in the target language

---

/cc:create-command - Create New Command

Quickly creates a new custom command.

Usage:

/cc:create-command deploy "Automated deployment to production"

---

Creating Your Own Commands

Step 1: Create the Command File

mkdir -p .claude/commands
touch .claude/commands/my-command.md

Step 2: Write the Command Content

---
description: My custom command description
argument-hint: [parameter description]
allowed-tools: Read, Write, Bash
---

# Command Title

You are a professional [role].

## Task

Based on the user's $ARGUMENTS, perform the following:

1. Step one
2. Step two
3. Step three

## Output Format

-   Use clear headings
-   Provide specific recommendations

Step 3: Use the Command

/my-command argument content

---

Command Config Parameters

Common allowed-tools Values

• Read - Read files
• Write - Write files
• Edit - Edit files
• Glob - File search
• Grep - Content search
• Bash - Execute commands
• Bash(git:*) - Git commands only
• Bash(gh:*) - gh commands only
• TodoWrite - Task management
• WebFetch - Web fetching

---

Best Practices

1. Name clearly - command name should directly express the function
2. Describe accurately - description should be instantly understandable
3. Limit tools - use allowed-tools to restrict permissions
4. Structured output - define a clear output format
5. Error handling - consider edge cases

---

Recommended Command Set

---

Reference: feiskyer/claude-code-settings