Lesson 12: Tips, Tricks, and Best Practices

Learning Objectives

After completing this lesson, you will be able to:

• Communicate effectively with Claude for optimal results
• Recognize and avoid common mistakes
• Optimize performance and speed
• Manage costs effectively with /stats and /cost
• Collaborate efficiently in team environments
• Troubleshoot common issues independently

Prerequisites

• Completed Lessons 1-11 - Full Claude Code knowledge
• Real project experience - Applied workflows to your code
• Encountered challenges - Hit some roadblocks to overcome

Estimated Time: 20 minutes

---

Communication Tips for Better Results

How you talk to Claude significantly impacts the quality of assistance.

Be Specific and Clear

❌ Vague:

You> Fix the code

✅ Specific:

You> The login endpoint in src/api/auth.js line 45 is throwing
a 500 error when the password contains special characters.
Please fix the validation regex.

Provide Context

❌ No context:

You> Why doesn't this work?

✅ With context:

You> I'm trying to implement OAuth2 login using the library in
package.json. I followed the docs at example.com/oauth but
I'm getting a "invalid_grant" error. Here's my implementation:
[shows code]

Use Examples

❌ Abstract:

You> Make it faster

✅ With example:

You> Optimize this query. Currently it takes 3 seconds on a
table with 10k rows. I'd like it under 100ms.

[current query shown]

State Your Goal

❌ Task-focused:

You> Change the function

✅ Goal-focused:

You> I need to validate user input before saving to database
to prevent SQL injection. Please update the saveUser function.

Break Down Complex Requests

❌ Monolithic:

You> Refactor the entire application to use TypeScript,
add tests, fix bugs, improve performance, and document everything

✅ Incremental:

You> Let's start with the authentication module. Convert it to
TypeScript first, then we'll move to other modules.

---

Common Mistakes to Avoid

Learn from others' mistakes to avoid frustration.

Mistake 1: Not Reading File Before Editing

Problem:

You> Add error handling to the process function

Claude guesses wrong about file structure.

Solution:

You> Read @src/process.py first, then add error handling

Mistake 2: Ignoring Claude's Questions

Problem: Claude asks clarifying questions, you ignore them, get wrong result.

Solution:

Claude: Do you want this validation on client or server side?
You: Server side, using the existing validation middleware

Mistake 3: Too Large Changes at Once

Problem:

You> Refactor everything to use microservices

Overwhelming, prone to errors.

Solution:

You> Let's start by extracting the user service into a separate module

Mistake 4: Not Using Plan Mode

Problem: Jumping into complex tasks without planning.

Solution:

You> /plan
You> I want to migrate from PostgreSQL to MongoDB

Mistake 5: Blindly Accepting Changes

Problem: Always pressing Y without reviewing diffs.

Solution:

• Always review diffs
• Understand what's changing
• Ask questions if unsure

Mistake 6: Not Providing Enough Context

Problem:

You> It's broken

Solution:

You> The payment flow fails at step 3 with error code E5001.
Here's the error log: [paste log]

---

Performance Optimization

Make Claude Code faster and more efficient.

Reduce Token Usage

Use targeted reading:

❌ You> Read the entire src directory
✅ You> Read @src/auth.js and @src/user.js

Use @ for files:

❌ You> Can you look at the file src/api/users.js and tell me...
✅ You> @src/api/users.js - how does this work?

Be concise:

❌ You> I was wondering if you could perhaps help me understand
✅ You> Explain the auth flow

Leverage Context

Don't repeat information:

❌ You> Read the file
   [Claude reads]
   You> Now that you've read it, add error handling
   [Don't need to re-explain]

✅ Just continue: "Add error handling" (Claude knows context)

Use CLAUDE.md:

Put repeated context in CLAUDE.md instead of explaining
every session.

Use Background Tasks Wisely

Good for background:

• Long test runs
• Large file searches
• Build processes

Not for background:

• Quick questions
• Simple edits
• Urgent fixes

Choose Right Model

Haiku for:

• Quick questions
• Simple code generation
• Cost-sensitive tasks

Sonnet for:

• Most development work (default)
• Balanced speed and quality

Opus for:

• Complex reasoning
• Architecture decisions
• Critical features

---

Cost Management

Track and control your API usage.

Using /stats

Check session stats:

You> /stats

SESSION STATISTICS:
Duration: 45 minutes
Total tokens: 125,000
Input tokens: 80,000
Output tokens: 45,000
Estimated cost: $0.75

Operations:
- Read: 45 files
- Edit: 12 files
- Bash: 8 commands

Using /cost

Detailed cost breakdown:

You> /cost

COST BREAKDOWN:
Model: claude-sonnet-4-5
Input: $0.003/1K tokens × 80K = $0.24
Output: $0.015/1K tokens × 45K = $0.675
Total: $0.915

Most expensive operations:
1. Long explanation of architecture ($0.30)
2. Code generation for feature ($0.25)
3. Refactoring discussion ($0.20)

Cost Optimization Tips

Reduce input tokens:

• Use targeted file reading
• Avoid repeating context
• Keep CLAUDE.md concise

Reduce output tokens:

• Ask for concise answers
• Request code without explanations when you understand
• Use Haiku for simple tasks

Monitor regularly:

# Check stats periodically
You> /stats

# Set budget alerts in settings
{
  "budget": {
    "alertThreshold": 10.00,
    "hardLimit": 50.00
  }
}

---

Team Collaboration Tips

Use Claude Code effectively in teams.

Shared Configuration

Project-level settings:

.claude/settings.json - Commit to git

Shared by entire team.

Team Skills

Create shared skills:

.claude/skills/team-review/SKILL.md

Standardized code review checklist.

CLAUDE.md Convention

Document in CLAUDE.md:

# Team conventions
- We use PEP 8 for Python
- Test file name: test_<module>.py
- PR requires 1 approval

Git Attribution

Keep attribution enabled:

{
  "git": {
    "attribution": {
      "enabled": true
    }
  }
}

Transparent about AI assistance.

Communication

Document AI usage:

• In PR descriptions: "Assisted by Claude Code"
• In commit messages: Co-Authored-By
• In documentation: Note AI-generated sections

---

Troubleshooting Common Issues

Solve problems independently.

Issue: "Permission Denied" Errors

Symptoms: Claude can't read/write files.

Solutions:

1. Check file permissions:

ls -la file.py
chmod +r file.py

2. Run with appropriate user:

sudo claude  # Be careful with sudo!

3. Check permission mode:

You> What permission mode am I in?

Issue: Slow Responses

Symptoms: Claude takes long to respond.

Solutions:

1. Check internet connection
2. Try smaller model (Haiku)
3. Reduce context:

You> /compact  # Compact conversation

4. Check API status:

curl https://status.anthropic.com

Issue: "File Not Found"

Symptoms: Claude can't find files.

Solutions:

1. Use absolute paths:

You> @/full/path/to/file.py

2. Check current directory:

You> Show current directory

3. List files:

You> List all Python files

Issue: Wrong Model Behavior

Symptoms: Claude gives unexpected responses.

Solutions:

1. Check model:

You> What model are you using?

2. Switch model:

claude --model claude-opus-4-5

3. Clear and rephrase:

You> Let me clarify: [rephrase question]

Issue: Configuration Not Applied

Symptoms: Settings changes don't take effect.

Solutions:

1. Check settings location:

# User vs project vs local
claude --show-config

2. Restart session:

You> /exit
claude

3. Validate JSON:

cat ~/.config/claude-code/settings.json | jq .

Issue: Memory/Context Loss

Symptoms: Claude forgets earlier conversation.

Solutions:

1. Use CLAUDE.md for persistent context
2. Reference earlier parts:

You> Remember when we discussed the auth flow? Apply that here.

3. Manual compaction:

You> /compact

---

Pro Tips

Keyboard Shortcuts

Essential shortcuts:

• Ctrl+C - Cancel current operation
• Ctrl+D - Exit session
• Ctrl+P - Cycle permission modes
• Ctrl+L - Clear screen (keep history)

Alias Creation

Create convenient aliases:

# In ~/.bashrc or ~/.zshrc
alias cc='claude'
alias ccd='claude --develop'
alias ccr='claude --resume'
alias ccc='claude --commit'

Session Automation

Start with project:

# Function to start in project with context
ccd() {
  cd ~/projects/$1
  claude
}

Quick Commands

One-liners:

# Quick question
claude -p "What's the syntax for Python list comprehension?"

# Quick file check
claude -r @src/auth.py -p "Review this file"

# Quick commit
claude -c "git status" -c "git add ." -c "git commit -m 'WIP'"

---

✅ Check Your Understanding

1. What's the best way to ask about a specific file?

   • "Look at the file"
   • Use @filename
   • "Read the thing"
   • "Check file.py"

2. What command shows session statistics?

   • /info
   • /stats
   • /session
   • /show

3. Why should you use plan mode for complex tasks?

   • It's faster
   • It creates a roadmap before implementation
   • It uses less tokens
   • It's required

4. True or False: You should always accept Claude's changes without reviewing.**

   • True
   • False - always review diffs

5. What's a common mistake?

   • Asking questions
   • Being specific
   • Not providing enough context
   • Using @ for files

Answers: 1-b, 2-b, 3-b, 4-False, 5-c

---

Summary

In this lesson, you learned:

• Communication - Be specific, provide context, use examples
• Common mistakes - Not reading files, ignoring questions, large changes
• Performance - Reduce tokens, leverage context, choose right model
• Cost management - Use /stats, monitor usage, optimize prompts
• Team collaboration - Shared config, team skills, attribution
• Troubleshooting - Solve common issues independently

Next Steps

In Lesson 13: Reference Materials, you'll find:

• Complete slash command reference
• All CLI flags documentation
• Built-in tools reference
• Keyboard shortcuts
• Settings reference
• File locations cheat sheet

Further Reading

• Best Practices Guide
• Performance Tips
• Troubleshooting

---

Continue to Lesson 13: Reference Materials →