Claude Code: Tips & Power Features

Project Configuration: CLAUDE.md

What: A markdown file in your project root that gives Claude persistent context.

Location: /project-root/CLAUDE.md (or .claude/CLAUDE.md)

What to Include

# Project: [Name]

## Overview
Brief description of what this project does.

## Tech Stack
- Frontend: React 18, TypeScript, Tailwind
- Backend: Node.js, Express, PostgreSQL
- Testing: Jest, Playwright

## Architecture
- /src/components - React components
- /src/api - API routes
- /src/utils - Shared utilities

## Conventions
- Use functional components with hooks
- Prefer named exports
- Tests go in __tests__ folders
- Use kebab-case for file names

## Common Commands
- `npm run dev` - Start development server
- `npm test` - Run tests
- `npm run build` - Production build

## Important Notes
- Never commit to main directly
- All API keys are in .env (not committed)
- Database migrations are in /migrations

Why It Matters

• Claude reads this automatically at session start
• Reduces repeated context in prompts
• Ensures consistent code style
• Documents tribal knowledge

Memory: memory.md

What: A file where Claude stores learnings that persist across sessions.

Location: .claude/memory.md

Example Content

# Memory

## Learned Preferences
- User prefers verbose error messages
- Always run tests before committing
- Use conventional commits format

## Project Discoveries
- The auth module has a bug with special characters (ticket #123)
- Performance issues traced to N+1 queries in /api/users
- Design system colors are in /src/theme/colors.ts

## Session Notes
- 2025-02-10: Refactored payment module, tests passing
- 2025-02-11: Started work on dashboard feature

Subagents: Parallel Execution

What: Spawn separate Claude instances to work on tasks in parallel.

When to Use

• Independent tasks that don't block each other
• Research while coding
• Testing while refactoring

How to Invoke

You: "Work on the login page while a subagent writes tests for the API"

Best Practices

Do	Don't
Use for truly independent tasks	Use for sequential dependencies
Merge results intentionally	Let subagents conflict
Give clear, scoped instructions	Give vague, overlapping tasks

Hooks: Automation Triggers

What: Shell commands that trigger automatically at specific events.

Hook Events

Event	When It Fires	Example Use
PreToolUse	Before Claude uses a tool	Validate commands
PostToolUse	After Claude uses a tool	Run linter after edits
Notification	When Claude wants to notify	Send Slack message
Stop	When Claude finishes	Run test suite

Example: Auto-run Tests After File Changes

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "command": "npm test --watchAll=false"
      }
    ]
  }
}

Example: Log Failures for Learning

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Bash|Execute",
        "command": "if [ $EXIT_CODE -ne 0 ]; then echo \"$(date -Iseconds) | $TOOL | $COMMAND | Exit: $EXIT_CODE\" >> .claude/failures.log; fi"
      }
    ]
  }
}

Slash Commands

Essential Built-in Commands:

Command	What It Does
/help	Show available commands
/clear	Clear conversation context
/compact	Summarize and compress context
/config	View/edit configuration
/cost	Show token usage and costs
/doctor	Diagnose setup issues
/init	Initialize CLAUDE.md for project
/mcp	Manage MCP servers
/memory	View/edit memory
/review	Review recent changes
/revert	Undo recent file changes

Plan Mode vs. Auto Mode

Plan Mode (Recommended for Learning)

claude --plan

• Claude proposes actions before executing
• You approve or modify the plan
• Safer for complex or risky changes

Auto Mode (For Trusted Tasks)

claude --auto

• Claude executes without asking
• Faster for routine tasks
• Use with good test coverage

Hybrid Approach

• Start projects in plan mode
• Switch to auto for familiar patterns
• Return to plan for critical sections

Tips from Power Users

1. Start with the end in mind — "Create a working login page" is better than "Help me code"
2. Give Claude the full picture — Share error messages completely, include relevant file context, explain the why not just the what
3. Iterate, don't restart — Build on what Claude produces; refine rather than re-prompt
4. Let Claude read first — "Read the codebase and understand the patterns before making changes"
5. Use Claude for code review — "Review this PR for bugs, security issues, and style"
6. Automate the boring stuff — Tests, documentation, boilerplate. Claude is tireless for repetitive tasks
7. Trust but verify — Run the tests, review the diff, understand the changes