Claude Code Advanced Guide: From Beginner to Power User

Introduction

After using Claude Code for a while, you'll realize it's far more than "ChatGPT in a terminal." It's a programmable agent runtime — configured properly, it can multiply your productivity several times over.

This post distills the advanced techniques I've picked up from daily use, from configuration optimization to workflow design, to help you get the most out of Claude Code.

1. CLAUDE.md — Your First Productivity Lever

1.1 Three-Layer Configuration System

Claude Code's CLAUDE.md has three levels, from highest to lowest priority:

~/.claude/CLAUDE.md              # Global config (applies to all projects)
project-root/CLAUDE.md           # Project-level config (shared with team)
subdirectory/CLAUDE.md           # Directory-level config (module-specific)

Best practices:

• Global CLAUDE.md: general preferences (language, code style, commit conventions)
• Project CLAUDE.md: tech stack, architecture conventions, key paths
• Subdirectory CLAUDE.md: module-specific rules (e.g., components/CLAUDE.md for component guidelines)

1.2 Writing an Effective CLAUDE.md

Don't write CLAUDE.md like documentation — treat it as a "work manual" for the AI:

# Project: My Blog
 
## Tech Stack
- Next.js 16 (App Router) + TypeScript
- Tailwind CSS v4 (no tailwind.config.js, uses CSS @import)
- next-intl for i18n, supports zh/en
 
## Code Conventions
- Components use PascalCase, utility functions use camelCase
- Prefer server components; only use 'use client' when interactivity is needed
- Commit messages in English, format: type: description
 
## Key Paths
- Blog content: content/zh/*.mdx and content/en/*.mdx
- Route structure: app/[locale]/blog/[slug]/page.tsx
- i18n config: i18n/routing.ts
 
## Do Not
- Do not modify proxy.ts (middleware config)
- Do not use useState/useEffect in server components

Core principle: Be specific, actionable, and constrained. Vague instructions ("write good code") are less effective than explicit rules ("functions should not exceed 30 lines").

2. Hooks — Automate Your Workflow

Hooks are Claude Code's "event hooks" — they automatically execute commands before or after specific operations.

2.1 Configuration Location

~/.claude/settings.json              # Global hooks
project-root/.claude/settings.json   # Project-level hooks

2.2 Practical Hook Examples

Auto-format after saving a file:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "command": "npx prettier --write \"$CLAUDE_FILE_PATH\""
      }
    ]
  }
}

Auto-lint before committing:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "command": "if echo \"$CLAUDE_TOOL_INPUT\" | grep -q 'git commit'; then npx eslint . --quiet; fi"
      }
    ]
  }
}

Block dangerous operations:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "command": "if echo \"$CLAUDE_TOOL_INPUT\" | grep -qE 'rm -rf|git push --force|git reset --hard'; then echo 'BLOCKED: Dangerous operation intercepted' >&2; exit 1; fi"
      }
    ]
  }
}

2.3 Hook Event Types

3. Context Management — The Most Underrated Skill

Claude Code's context window is finite. Managing it well directly determines output quality.

3.1 Core Commands

/clear          # Clear current conversation context (most used)
/compact        # Compress context, keeping key information
/context        # View current context usage
/resume         # Resume a previous session

3.2 When to /clear

• Switching to a completely different task
• Context is nearly full (slower responses are a signal)
• Previous conversation led you down the wrong path

3.3 When to /compact

• Current task isn't done yet, but context is nearly full
• You want to keep key decisions and progress, but discard intermediate steps

3.4 Tips for Feeding Context

# Feed file contents directly to Claude Code
cat src/lib/api.ts | claude "analyze the error handling in this file"
 
# Feed git diff for code review
git diff main | claude "review these changes"
 
# Feed error logs for debugging
npm run build 2>&1 | claude "help me analyze these build errors"

4. Multi-Agent Parallelism — The Real Productivity Multiplier

4.1 Plan Mode

Press Shift+Tab twice to enter Plan Mode — Claude will analyze before acting:

• Great for complex refactors and new feature design
• Review the plan first, then confirm execution
• Avoids the "write a bunch then start over" trap

4.2 Parallel Agents

Claude Code can spin up multiple sub-agents to handle independent tasks simultaneously:

You: Do these three things in parallel:
1. Add skeleton loading to the PostCard component
2. Optimize getPostsByLocale performance
3. Write an RSS feed generator

Claude Code will automatically determine these tasks are independent and process them in parallel.

4.3 Multi-Terminal Parallelism

How Boris, the creator of Claude Code, works:

• Run 2-3 Claude Code terminals locally, each handling different modules
• Open 5-10 web sessions on claude.ai/code simultaneously
• Use --resume to switch context between local and web

# Terminal 1: Frontend work
claude "refactor the Navigation component"
 
# Terminal 2: Backend work
claude "optimize MDX parsing performance"
 
# Terminal 3: Testing
claude "write unit tests for posts.ts"

5. Custom Slash Commands (Skills)

5.1 Creating Custom Commands

Create a .claude/commands/ directory in your project root. Each .md file becomes a command:

.claude/commands/
  review.md       # /review
  test.md         # /test
  deploy.md       # /deploy

Example: /review command

<!-- .claude/commands/review.md -->
Please code review the changes in the current git diff:
 
1. Check for security vulnerabilities (XSS, injection, etc.)
2. Check for performance issues
3. Check compliance with project code conventions
4. Provide improvement suggestions
 
Use git diff --staged to get the staged changes.

Example: /test command

<!-- .claude/commands/test.md -->
Write unit tests for $ARGUMENTS:
 
- Use the vitest framework
- Cover happy paths and edge cases
- Mock external dependencies
- Place test files in __tests__ under the same directory

Usage: /test src/lib/posts.ts — $ARGUMENTS gets replaced with whatever follows the command.

5.2 Team-Shared Commands

Commit .claude/commands/ to Git, and everyone on the team can use the same commands for a unified workflow.

6. MCP Servers — Extending Claude Code's Capabilities

MCP (Model Context Protocol) lets Claude Code connect to external tools and data sources.

6.1 Configuration

Add to .claude/settings.json:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "your-token"
      }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/docs"]
    }
  }
}

6.2 Popular MCP Servers

6.3 Real-World Scenarios

You: Check the 5 most recent GitHub issues, filter for bugs, and analyze priority

You: Query the database for user registration data from the past 7 days and generate a trend analysis

With MCP, Claude Code is no longer limited to the local filesystem.

7. Effective Prompting Techniques

7.1 Task Decomposition

Bad prompt:

Build me a blog search feature

Good prompt:

Add full-text search to the blog:
1. Use Pagefind for static search indexing
2. Integrate search UI into the SearchBar component
3. Highlight matching keywords in search results
4. Support both Chinese and English search

First, look at the existing SearchBar.tsx and build process, then give me an implementation plan.

7.2 Provide Constraints

Refactor the Navigation component with these requirements:
- Keep the existing API unchanged
- Don't introduce new dependencies
- Use a hamburger menu on mobile
- All changes must be type-safe

7.3 Leverage Pipe Input

# Have Claude write new code matching existing style
cat src/components/PostCard.tsx | claude "write a TagBadge component in the same style"
 
# Have Claude explain complex errors
npm run build 2>&1 | claude "explain these errors and suggest fixes"
 
# Have Claude review a PR
gh pr diff 42 | claude "review this PR"

7.4 Iterative Development

Don't try to do everything in one shot:

Round 1: Implement core functionality, get the happy path working
Round 2: Add error handling and edge cases
Round 3: Optimize performance and UX
Round 4: Write tests

Use /compact between rounds to compress context and keep Claude focused.

8. Keyboard Shortcuts Cheat Sheet

9. Real-World Workflow Templates

9.1 Bug Fix Workflow

1. Feed the error message to Claude
2. Claude locates the problem file and root cause
3. Enter Plan Mode to confirm the fix approach
4. Execute the fix
5. /test to run tests and verify
6. /review to self-check changes
7. /commit to commit

9.2 New Feature Workflow

1. Describe requirements and constraints in natural language
2. Plan Mode to design the approach
3. Confirm, then execute implementation
4. Iterate by module (core → edge cases → optimization → tests)
5. /review + /commit

9.3 Code Refactoring Workflow

1. Have Claude analyze the existing code structure
2. Plan Mode to outline refactoring steps
3. Run tests after each step to ensure nothing breaks
4. /compact to keep context clean
5. Final /review for a full check

10. Common Pitfalls

1. Cramming too many tasks at once — Claude works best with one clear task at a time
2. Not clearing context — Context pollution degrades output quality; use /clear when needed
3. Skipping CLAUDE.md — Repeating project background every time wastes tokens and time
4. Not using Plan Mode — Diving straight into complex tasks often leads to dead ends
5. Ignoring Hooks — Manually running lint/format/test is less efficient than automation
6. Not breaking down tasks — "Build me a complete e-commerce site" is far less effective than 20 small, focused tasks

Further Reading

• The Complete History of LLMs: From RNN to GPT — Understand the AI foundations behind Claude
• Claude Code Performance & Cost Optimization Guide — Get the most out of Claude Code while controlling costs
• Claude Code Security Best Practices — Use Claude Code safely
• Claude Code Official Hooks Guide
• How the Creator of Claude Code Actually Uses It
• Claude Code Context Management Manual
• 50 Claude Code Tips
• Claude Code Best Practices Checklist
• CLAUDE.md Optimization Tips
• Claude Code Complete Command Reference