Claude Code Cheat Sheet 2026: Commands, Keyboard Shortcuts, CLI Flags & Custom Skills

Anthropic’s Claude Code CLI is an autonomous, agentic coding tool that reads your local repository codebase, edits multi-file project structures, runs shell terminal commands, and manages git workflows directly from your terminal. Powered by Claude 3.7 Sonnet, Claude Opus 4.8, and Claude Fable 5, it operates as an interactive pair-programmer—interpreting natural language goals into precise file reads, AST searches, and code modifications.

This Claude Code Cheat Sheet serves as the definitive reference manual for developers, DevOps engineers, and technical architects. Bookmark this page for quick access to every keyboard shortcut, slash command, CLI flag, configuration schema, and Unix piping script.

⚡ 1. Quick Installation & Setup

Claude Code runs natively across macOS, Linux, and Windows PowerShell/WSL2 environments.

# macOS & Linux (Native POSIX Shell Script - Recommended)
curl -fsSL https://claude.ai/install.sh | bash

# macOS via Homebrew Cask
brew install --cask claude-code

# Windows Native PowerShell 7+ (Requires Git for Windows)
irm https://claude.ai/install.ps1 | iex

# Verify Binary & Authenticate
claude --version
claude login

[!WARNING] Deprecated Installation Notice: Anthropic has officially deprecated npm install -g @anthropic-ai/claude-code for general usage because npm global installations cannot support background self-updates. Always use native OS scripts.

⌨️ 2. Essential Keyboard Shortcuts Matrix

These keyboard shortcuts operate inside interactive Claude Code terminal sessions.

Navigation, Interrupts & Execution

Input Formatting & Line Editing

Model & Reasoning Toggles

🛠️ 3. Interactive Slash Commands Reference

Slash commands control session state, token usage, code review suites, and environment settings.

mindmap
  root((Claude Code Slash Commands))
    Session Management
      /clear
      /compact
      /cost
      /context
      /rewind
      /cd
    Project & Code Tools
      /init
      /code-review
      /simplify
      /verify
      /plan
      /branch
      /batch
    System & Diagnostics
      /doctor
      /checkup
      /insights
      /loop
      /teleport
      /remote-control

Context & Session Controls

Code Analysis & Refactoring Tools

System Diagnostics & Advanced Utilities

🤖 4. Non-Interactive CLI Automation & Piping (claude -p)

The -p (print) flag runs Claude Code in headless, non-interactive mode. This enables integration into shell scripts, CI/CD runners, and git hooks.

Core Automation Flags

# Simple non-interactive query
claude -p "Explain the error handling strategy in src/server.ts"

# Read input from standard Unix pipe
cat error.log | claude -p "What is causing this database connection pool exhaustion?"

# Audit Git Diff non-interactively
git diff main | claude -p "Review for OWASP Top 10 security vulnerabilities"

# Enforce JSON output format for downstream shell processing
claude -p "Extract all API route definitions from src/routes/" --output-format json

# Restrict allowed tool calls (Auto-approve specific tools)
claude -p "Fix unit test failures in tests/auth.test.ts" --allowedTools "Read,Edit,Bash(npm test)"

# Explicitly block destructive commands
claude -p "Refactor API layer" --disallowedTools "Bash(rm:*),Bash(sudo:*)"

# Set strict spending and turn limits
claude -p "Refactor database migrations" --max-budget-usd 3.50 --max-turns 5

# Headless CI/CD container mode (Bypasses all confirmation prompts)
claude --dangerously-skip-permissions -p "Run linter and auto-fix formatting issues"

JSON Schema Output Enforcement

To force Claude Code to return structured JSON adhering to a specific schema:

claude -p "Extract exported function signatures from src/utils.ts" \
  --output-format json \
  --json-schema '{
    "type": "object",
    "properties": {
      "functions": {
        "type": "array",
        "items": {
          "type": "object",
          "properties": {
            "name": { "type": "string" },
            "returnType": { "type": "string" }
          },
          "required": ["name", "returnType"]
        }
      }
    },
    "required": ["functions"]
  }'

📄 5. CLAUDE.md Project Configuration Architecture

CLAUDE.md is the primary configuration file for establishing repository-level guidelines, build instructions, and coding standards. It is loaded automatically at the start of every session.

File Hierarchy & Resolution Order

Project Root Configuration:
├── ./CLAUDE.md                  # Project-specific rules (Committed to Git)
├── ./.claude/CLAUDE.md          # Alternative project path (Committed to Git)
├── ./.claude/rules/*.md         # Path-specific conditional rules
└── ~/.claude/CLAUDE.md          # Global user preferences (Personal - Not committed)

Production CLAUDE.md Template

Create CLAUDE.md at your repository root:

# Repository Technical Guidelines — Modern E-Commerce API

## Build, Test & Lint Commands
- Run Unit Tests: `npm test`
- Run Single Test File: `npx jest tests/auth.test.ts`
- Execute Production Build: `npm run build`
- Type Check & Lint: `npm run typecheck && npm run lint`

## Architecture & Framework Stack
- Runtime & Framework: Node.js v20 LTS + Express.js + TypeScript Strict
- Database Engine: PostgreSQL 16 managed via Prisma ORM (`prisma/schema.prisma`)
- Context & Cache Layer: Redis Enterprise (`src/lib/redis.ts`)

## Coding Standards & Conventions
- All database queries must use parameterized statements or Prisma client calls.
- Never write bare `any` types; define explicit Zod validation schemas for API inputs.
- All errors must be handled by `src/middleware/errorHandler.ts`.

## Environment Context References
- Import Database Schema: @prisma/schema.prisma
- Import API Package Manifest: @package.json

Path-Specific Rules (.claude/rules/*.md)

To enforce rules that only apply when Claude touches specific subdirectories, create path-targeted rule files under .claude/rules/.

Example: .claude/rules/api-routes.md:

---
paths:
  - "src/routes/**/*.ts"
  - "src/controllers/**/*.ts"
---

# REST API Security Guidelines
- Every endpoint MUST enforce JWT bearer token validation middleware.
- Input validation MUST be executed using Zod schemas before hitting business logic.
- API endpoints MUST return standard RFC 7807 problem details JSON on error.

🔌 6. MCP Servers Setup (.mcp.json)

Model Context Protocol (MCP) connects Claude Code to external tools, databases, issue trackers, and third-party APIs.

Project Configuration (.mcp.json)

Create .mcp.json at your repository root to share tool integrations across your engineering team:

{
  "mcpServers": {
    "github": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "$GITHUB_TOKEN"
      }
    },
    "postgres": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "DATABASE_URL": "postgresql://postgres:secret@localhost:5432/production_db"
      }
    },
    "sentry": {
      "type": "http",
      "url": "https://mcp.sentry.io/query"
    }
  }
}

Managing MCP via CLI Subcommands

# Add a new stdio MCP server
claude mcp add github -e GITHUB_TOKEN=$GITHUB_TOKEN -- npx @modelcontextprotocol/server-github

# Add an HTTP transport MCP server
claude mcp add --transport http sentry https://mcp.sentry.io/query

# List all active MCP server connections
claude mcp list

# Authenticate OAuth MCP connections
claude mcp login github

🧠 7. Custom Skills & Subagents

Extend Claude Code by authoring reusable slash commands (Skills) and specialized subagents.

Authoring Custom Skills (.claude/skills/)

Create .claude/skills/deploy/SKILL.md:

---
name: deploy
description: Executing automated staging/production deployments
argument-hint: "[environment]"
user-invocable: true
---

Execute a verified deployment to the $0 environment (default: staging).

Execution Workflow:
1. Run `npm run typecheck` and `npm test` to verify codebase integrity.
2. Build static artifacts via `npm run build`.
3. Execute deployment script: `./scripts/deploy.sh $0`.
4. Trigger health check endpoint and verify HTTP 200 OK response.

Invoke your skill inside terminal sessions:

/deploy production

Authoring Custom Subagents (.claude/agents/)

Create .claude/agents/security-auditor/AGENT.md:

---
name: security-auditor
description: Specialized OWASP security auditing agent
tools: Read, Grep, Glob, Bash
model: sonnet
---

You are an expert Application Security Auditor. When invoked:
1. Audit all modified files for SQL injection, XSS, and unauthenticated endpoints.
2. Verify that secrets, private keys, and API tokens are NOT committed.
3. Validate CORS origins and JWT verification middleware.

Invoke your agent in interactive or headless modes:

# Interactive call
@security-auditor Perform security audit on latest changes

# Headless CLI call
claude --agent security-auditor -p "Audit src/auth/ for vulnerabilities"

⚡ 8. Background Agents & Dynamic Workflows

Claude Code supports running full autonomous agent sessions detached in the background.

# Launch a background agent session
claude --bg "Investigate flaky test failures in tests/integration/"

# Launch a background shell command
claude --bg --exec "npm run test:e2e"

# Open interactive TUI background agent manager
claude agents

# Manage background sessions by ID
claude attach 7c5dcf5d    # Attach to terminal
claude logs 7c5dcf5d      # Tail live log output
claude stop 7c5dcf5d      # Stop running background session

Dynamic Multi-Agent Workflows

When handling large tasks under high reasoning effort (/effort max or ultracode), Claude Code automatically spawns parallel subagents:

• /subtask <directive>: Spawns an in-session subagent that executes an isolated task and reports back.
• /batch <task>: Decomposes repository-wide refactoring into 5–30 units running in parallel git worktrees.

⚙️ 9. Environment Variables & TUI Optimization

Fine-tune CLI performance, proxy configuration, and rendering settings via environment variables.

Recommended Shell Profile Exports (~/.zshrc / ~/.bashrc)

# Eliminate terminal screen flicker & enable mouse event support
export CLAUDE_CODE_NO_FLICKER=1

# Set default model & reasoning budget
export ANTHROPIC_MODEL="sonnet"
export CLAUDE_CODE_EFFORT_LEVEL="high"

# Configure corporate HTTP proxy
export HTTPS_PROXY="http://proxy.enterprise.com:8080"

# Safety caps: Limit subagent spawns per session (Default: 200)
export CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION=100

🧪 10. Real-World Shell Piping Examples

Claude Code integrates into Unix pipeline chains.

1. Automated Git Commit Message Generator

git diff --staged | claude -p "Generate a concise Conventional Commit message based on these changes" --model haiku

2. Live Log Error Root Cause Analysis

tail -n 500 /var/log/nginx/error.log | claude -p "Analyze these Nginx log errors and output root cause troubleshooting steps"

3. Kubernetes Pod Failure Audit

kubectl logs deployment/payment-service --tail=200 | claude -p "Identify exception stack traces and suggest fixes"

4. Docker Container Resource Inspection

docker stats --no-stream | claude -p "Which containers are exceeding memory thresholds?"

📊 Summary Quick Reference Card

================================================================================
CLAUDE CODE QUICK REFERENCE CARD
================================================================================
Launch Interactive Session:    claude
Resume Last Session:           claude -c
Initialize CLAUDE.md:          /init
Switch Permission Mode:        Shift + Tab
Rewind Code & History:         Esc + Esc
Compact Context Window:        /compact
Check Token Cost:              /cost
Run Setup Health Check:        /doctor  (press 'f' to auto-fix)
Headless Non-Interactive Run:  claude -p "prompt" --output-format json
Flicker-Free TUI Mode:         export CLAUDE_CODE_NO_FLICKER=1
================================================================================

Related Guides & Workflows

• For native Windows PowerShell & WSL2 setup, read How to Install and Run Claude Code CLI on Windows.
• For cross-platform installation methods, see How to Install and Set Up Claude Code CLI.
• For agent loop mechanics, explore Inside Claude Code Agent: Terminal Loop Architecture.