Claude Code: Interactive Mode and Operation Guide
Last updated: 2026-08-31
Claude Code offers multiple interaction modes, from terminal conversations to pipeline automation, each suited for different scenarios.
💡 Tip: Interactive mode is for conversing with Claude Code; pipeline mode is for scripts calling Claude Code; headless mode is for UI-free automation.
📋 Prerequisites: Chapter 6 - Project Initialization and Structure
1. What You'll Learn
- Three interaction modes and use cases
- Interactive session operation tips
- Pipeline mode input/output
- Headless mode automation scenarios
- Interrupt, resume, and multi-session management
2. Three Interaction Modes
(1) Mode Overview
| Mode | Launch Method | Use Case | Human Interaction |
|---|---|---|---|
| Interactive | claude |
Daily dev, exploratory coding | Full interaction |
| Pipeline | echo "..." | claude |
Quick one-off tasks | Semi-interactive |
| Headless | claude --headless |
CI/CD, automation scripts | No interaction |
(2) Mode Selection Decision
graph TD
A[Need Claude Code] --> B{Need interaction?}
B -->|Yes| C{Single or multi-turn?}
B -->|No| D[Headless mode]
C -->|Multi-turn| E[Interactive]
C -->|Single| F[Pipeline mode]
D --> D1[CI/CD]
D --> D2[Batch scripts]
E --> E1[Daily development]
E --> E2[Complex refactoring]
F --> F1[Quick query]
F --> F2[Single file modification]
3. Interactive Mode
(1) Launch and Operations
BASH
# Standard launch
claude
# Launch with initial instruction
claude "Analyze project structure"
# Resume last session
claude --resume
# Specify model
claude --model claude-opus-4-20250514
(2) In-Session Operation Control
| Operation | Key/Command | Description |
|---|---|---|
| Submit instruction | Enter | Send current input |
| New line | Shift+Enter | Multi-line input |
| Interrupt | Ctrl+C | Stop current operation |
| Exit | Ctrl+D or /exit |
Exit Claude Code |
| Clear context | /clear |
Clear conversation history |
| View help | /help |
Show available commands |
▶ Example 1: Interactive Session Flow
TEXT
📖 Display only
$ claude
> Show me the utility functions under src/utils/
Claude: Reading src/utils/...
- date.ts formatDate(), relativeTime()
- string.ts capitalize(), truncate()
- validation.ts isValidEmail(), isValidPhone()
> Add a slugify function to string.ts
Claude: Adding slugify() to src/utils/string.ts...
→ Reading src/utils/string.ts
→ Adding slugify function with tests
→ Running: npm test -- string.test.ts
✅ All tests passed
> Also add a camelToSnake conversion function
Claude: Adding camelToSnake() to src/utils/string.ts...
→ Reading src/utils/string.ts
→ Adding camelToSnake with tests
→ Running: npm test -- string.test.ts
✅ All tests passed
4. Pipeline Mode
(1) Basic Usage
BASH
# Pass instruction from stdin
echo "Explain this function's purpose" | claude
# Pass file content to Claude Code
cat src/auth.ts | claude "Review this code's security"
# Combine usage
git diff | claude "Review these changes, identify potential issues"
(2) Output Control
BASH
# Plain text output (no interactive confirmation)
claude -p "List all TODO comments" --output text
# JSON format output
claude -p "Analyze project dependencies" --output json
# Quiet mode (output results only)
echo "Run tests" | claude --quiet
▶ Example 2: Pipeline Mode Practical Applications
BASH
# Code review automation
git diff main..feature | claude "Review changes:
1. Security vulnerabilities
2. Code style compliance
3. Missing tests"
# Batch documentation generation
for file in src/*.ts; do
echo "Generate JSDoc comments for $file" | claude --quiet >> docs/api.md
done
# Quick query
echo "What outdated dependencies are in package.json?" | claude
5. Headless Mode
(1) Launch Method
BASH
# No UI (auto-approve operations)
claude --headless "Run full tests and fix all failing cases"
# Specify output format
claude --headless --output json "Analyze code quality"
# With CI/CD
claude --headless --allow-full-auto "Code review and auto-fix lint issues"
(2) Security Considerations
| Parameter | Risk Level | Description |
|---|---|---|
--headless |
Medium | Auto-runs but needs pre-approved permissions |
--allow-full-auto |
High | Fully automatic, all operations without confirmation |
--allowed-tools |
Low | Only allows specified tools |
▶ Example 3: Headless Mode in CI/CD
YAML
# GitHub Actions config
name: AI Code Review
on: [pull_request]
jobs:
review:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Install Claude Code
run: npm install -g @anthropic-ai/claude-code
- name: Run AI Review
env:
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
run: |
claude --headless --output json \
"Review this PR's changes, check:
1. Security vulnerabilities
2. Performance issues
3. Code style
Output JSON format report"
6. Multi-Session Management
(1) Session Resume
BASH
# List historical sessions
claude --list-sessions
# Resume specific session
claude --resume <session-id>
# Resume most recent session
claude --resume
(2) Switching Between Sessions
TEXT
📖 Display only
# Alice handling multiple tasks simultaneously
# Task 1: Frontend refactoring
cd frontend && claude
> Refactor component structure...
# Ctrl+D to exit
# Task 2: Backend API
cd backend && claude
> Add new endpoint...
# Resume task 1
cd frontend && claude --resume
7. Comprehensive Example: Multi-Mode Collaboration
BASH
# Alice's daily development workflow
# Morning: Interactive mode exploration
claude "Check yesterday's TODOs, help me handle each one"
# Noon: Pipeline mode quick fix
git diff | claude "Fix these lint errors" | git apply
# Afternoon: Interactive mode deep development
claude "Implement RBAC model for user permission system"
# Evening: Headless mode full check
claude --headless --allowed-tools Read,Write,Bash \
"Run full tests, fix all failing cases, don't modify test assertions"
# CI: Headless mode auto review
claude --headless --output json "Code review"
❓ FAQ
Q Can I switch between interactive and pipeline modes?
A No. Mode is selected at launch and cannot be switched during runtime. Exit and relaunch for a different mode.
Q Is headless mode safe?
A
--headless only removes the UI; operations still require permission control. But --allow-full-auto auto-approves all operations — high risk, only recommended in CI environments.Q How to interrupt Claude Code execution?
A Press Ctrl+C to interrupt current operation without losing existing modifications. Ctrl+D exits the entire session.
Q How long is session history retained?
A Session history is saved locally, retaining the last 30 days by default. View with
--list-sessions.Q Does pipeline mode support multi-turn conversation?
A No. Pipeline mode is single input/output. Use interactive mode for multi-turn.
Q How to determine if Claude Code executed successfully in a script?
A Check exit codes: 0=success, 1=failure, 2=interrupted.
📖 Summary
- Three modes: Interactive (daily), Pipeline (quick one-off), Headless (automation)
- Interactive mode supports multi-turn conversation, Ctrl+C interrupt, Ctrl+D exit
- Pipeline mode suits single tasks, combinable with git diff and other commands
- Headless mode for CI/CD;
--allow-full-autorequires caution - Sessions can be resumed, supporting multi-task parallelism
📝 Exercises
- Basic (⭐): Complete the same task using both interactive and pipeline modes, compare operation differences.
- Intermediate (⭐⭐): Implement automated code review flow with
git diff | claude "code review". - Advanced (⭐⭐⭐): Configure a GitHub Actions workflow to auto-run Claude Code code review on PRs.
