Section 1: Terminal Workflows & Command Patterns

CLI Basics

Interactive Mode

# Start interactive session
claude

# In session:
> How does authentication work in this project?
> Fix the linting errors in src/
> Generate tests for the UserService

Command Mode (One-shot)

# Single command
claude "Explain the database schema"

# With file context
claude "@src/models/User.js explain this model"

# Extended thinking
claude --thinking=5000 "Design microservices architecture"

Anti-Pattern: Using extended thinking for simple tasks (formatting, quick questions)
Better: Reserve --thinking for complex reasoning tasks. Use default mode for simple queries.
Why: Extended thinking adds 2-4x latency and cost. Simple tasks like “fix linting” or “format code” don’t benefit from deep reasoning.

Piping (Unix Composition)

# Pipe command output
git diff | claude "Review these changes"
npm run lint 2>&1 | claude "Fix these lint errors"
cat error.log | claude "Analyze this error"

# Chain operations
git log --oneline -10 | claude "Summarize these commits" > summary.txt

Anti-Pattern: Not using piping - copying/pasting command output manually
Better: Pipe directly to Claude: git diff | claude "review"
Why: Piping is faster, preserves formatting, and enables automation in scripts.

Command Patterns

Code Analysis

claude "Analyze code quality in @src/services/"
claude "Find security vulnerabilities in @src/"
claude "List all TODO comments with context"

Code Generation

claude "Generate API endpoint for user CRUD operations"
claude "Create React component for user profile"
claude "Write database migration to add email_verified column"

Anti-Pattern: Vague prompts like “fix my code” or “improve this”
Better: Specific requests with context: “Fix authentication timeout - 401 after OAuth refresh. Error: [paste]”
Why: Vague prompts get generic advice. Specific prompts with error messages, file context, and desired outcome get targeted fixes.

Refactoring

claude "Refactor @src/legacy.js to use async/await"
claude "Extract reusable functions from @src/utils.js"
claude "Convert class component to hooks in @components/UserList.jsx"

Session Management

Context Persistence

Session maintains conversation history
Files referenced stay in context
Can reference previous responses

Multi-turn Conversations

claude
> Add user authentication to the API
# Claude generates auth code
> Now add rate limiting to those endpoints
# Claude adds rate limiting, aware of previous auth code
> Generate tests for both features
# Claude generates tests for auth + rate limiting

Anti-Pattern: Starting new session for each question instead of continuing conversation
Better: Use multi-turn conversations for related tasks - context accumulates and responses improve
Why: Each new session loses context. Multi-turn conversations let Claude build on previous understanding, saving tokens (via caching).

CLI Configuration

Claude Code uses JSON-based settings files at multiple scopes:

User Settings (~/.claude/settings.json)

{
  "model": "claude-sonnet-4-5-20250929",
  "thinking": {
    "enabled": false,
    "budgetTokens": 5000
  },
  "output": {
    "color": true,
    "verbose": false
  }
}

Project Settings (.claude/settings.json)

{
  "model": "claude-sonnet-4-5-20250929",
  "contextFiles": [
    "CLAUDE.md",
    "README.md",
    "package.json"
  ]
}

User Preferences & MCP (~/.claude.json)

{
  "mcpServers": {
    "google-drive": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-gdrive"]
    }
  }
}

← Back to Part 2 Index

Next: CLI Automation →