FAQ & Troubleshooting

• Handles routine coding tasks (boilerplate, scaffolding)
• Executes your vision — it doesn't create the vision
• You maintain full control and oversight
• You review all changes before committing
• Architecture, decisions, and quality gates remain your responsibility
• Developers now focus on higher-level work: design, planning, orchestration

• You review all code before it's committed
• Your testing and CI/CD pipelines still apply
• Code is yours — you control all quality gates
• Security reviews remain your responsibility
• Integration with existing workflows is seamless

• React Native, Flutter, Android (Kotlin), iOS (Swift/SwiftUI)
• TypeScript, JavaScript, Kotlin, Swift, Java, Python, Go, Rust, and more
• Custom tooling and internal libraries

• Claude Pro Subscription — $20/month, straightforward, unlimited within plan limits
• Anthropic Console API Key — pay per token, more transparent, suitable for organizations

• Works with any standard internet connection
• No VPN or special firewall setup typically needed
• Can work with corporate proxies
• Bandwidth usage is minimal

• Builds context gradually from file structure and contents
• Understands interdependencies between modules
• Focuses on files relevant to the current task
• CLAUDE.md helps guide decision-making for your specific patterns

• Can understand older patterns and paradigms
• Helps gradually modernize — incrementally, not all-at-once
• Maintains consistency with existing code style
• Can add tests to untested legacy code
• Can refactor incrementally without breaking working functionality

• Windows "token '&&' not valid": You're in CMD, not PowerShell. Open Windows PowerShell instead.
• macOS "Permission denied": Use nvm to install Node.js, or use Homebrew (handles permissions correctly).
• Linux "command not found": Installation may not have completed. Restart terminal and check PATH variable.
• "Connection refused": Check internet connection and verify your firewall isn't blocking the install.

brew upgrade claude-code         # Homebrew (macOS)
winget upgrade Anthropic.ClaudeCode  # Windows WinGet
apt upgrade claude-code          # Debian/Ubuntu
dnf upgrade claude-code           # Fedora/RHEL

1. Run: claude
2. Browser opens automatically
3. Log in with your Anthropic account
4. Return to terminal — you're logged in
5. This only needs to happen once per machine

• Simple tasks (vague is OK): "Create a button component" · "Add a logout button" · "Fix this error [with error message]"
• Complex tasks (be specific):

"Create a settings screen with:
- Dark mode toggle with Redux persistence
- Notification preferences (push, email, SMS)
- Language selection with i18n
- Save to backend API
- Handle all error cases
- Include accessibility labels"

• Ask for changes: "Use TypeScript interfaces instead of types" · "Make the styling consistent with our theme"
• Iterate: "Simplify this implementation" · "Add comments to explain the logic"
• Start over: Delete generated files, ask again with a better/different prompt
• Manual edit: Edit the generated code yourself — you have full control

1. You make a request
2. Claude Code displays "Here's my plan:"
3. You review the approach
4. You approve, reject, or request changes
5. Only then does Claude Code execute

• ✓ Creating new branches
• ✓ Making new commits
• ✓ Pushing to remote
• ✓ Opening PRs

• ✗ Rewrite existing git history
• ✗ Force push
• ✗ Delete remote branches

Code Generation → Code Review → Testing → Linting → CI/CD → Deploy

• Manual code review (now also AI-assisted)
• Automated tests (now potentially more comprehensive)
• Linting and formatting (checked before commit)
• CI/CD pipelines (completely unchanged)
• Security scans (unchanged)

# Code Standards
- Components: PascalCase
- Functions: camelCase
- Always use TypeScript
- Redux for global state
- Jest for testing
- ESLint config at root

• Faster learning: See working, pattern-correct code generated
• More independence: Can tackle tasks without waiting for senior review on every decision
• Fewer blockers: Don't need expert help for every implementation detail
• Knowledge transfer: Generated code teaches architectural patterns

• Generated code is yours to use however you want
• Modify or delete as needed — no restrictions
• No licensing conflicts with your project
• No sharing requirements with Anthropic
• Deploy to production with full confidence

• Velocity: Features per week, story points completed, time per task
• Quality: Test coverage %, bugs per release, code review time
• Experience: Developer satisfaction, onboarding time, knowledge sharing

• Week 1: 10–15% gain (learning curve)
• Week 2–4: 40–60% gain (integration phase)
• Month 2+: 60–80% gain (mastery phase)

• Your natural language prompt
• Code context (only files Claude needs to understand the task)
• Error messages (for debugging)

• ✓ Your entire codebase (only relevant files)
• ✓ API keys or secrets (you use environment variables)
• ✓ Private or sensitive data

// ✓ Good
const API_KEY = process.env.STRIPE_API_KEY;
const DB_PASSWORD = process.env.DB_PASSWORD;

// ✗ Never do this
const API_KEY = "sk_live_abc123xyz";
const DB_PASSWORD = "myPassword123";

• ✓ Not intentionally create security vulnerabilities
• ✓ Follow proper authentication patterns
• ✓ Recommend environment variables for secrets
• ✓ Validate input and handle errors gracefully

• Review all generated code before committing
• Run your security scanning tools
• Implement your security review gates
• Never blind-trust any code — AI-generated or otherwise

• Your Local Machine: Code stays on your machine. Claude Code executes locally with full visibility.
• Anthropic Servers: Only receives prompts and code context needed. Not used for training without permission.
• Enterprise: Can use Anthropic Console with private deployments where code never leaves your infrastructure.

• Google Drive (access design docs, specifications)
• Slack (retrieve context, post updates)
• Jira (manage issues and requirements)
• Custom MCP servers for your internal tools

# Guidelines
- Use TypeScript always
- Redux for state, not Context
- Tests for all business logic
- ESLint strict mode
- No console.log in production
- Components: PascalCase

# Project Structure
/src/screens    - React Native screens
/src/components - Reusable components
/src/services   - API calls and utilities
/src/redux      - Redux slices

# Stack
React Native + TypeScript + Redux + Jest

# Patterns
- Use Redux selectors (not direct state access)
- Custom hooks for logic extraction
- API calls in services layer only

# Code Style
- 2-space indentation (not 4)
- Always use const (not let)
- Semicolons required
- Trailing commas in objects
- Max line length: 100 characters

# Component Pattern
- Functional components only
- Props destructured in signature
- TypeScript interfaces required

"Fix the failing tests that were broken by these changes"
OR
"Run tests and show me what's failing, then fix them"

• API is temporarily down (rare, usually brief)
• Your authentication expired
• No internet connection
• Rate limit reached

1. Check internet connection
2. Try logging in again: claude --logout then claude
3. Wait a moment and retry
4. Check status at: anthropicstatus.com

• Check internet speed — Claude Code is API-dependent
• Break large tasks into smaller, focused requests
• Work on feature branches with focused scope
• Let it build context gradually — first request in a project takes longer
• Close other heavy applications consuming bandwidth