Claude Code Workflow Cheatsheet

Setup · Configuration · Skills · Hooks · Agent Teams · Subagents · Scheduled Tasks — 2026 Edition

⚡ Getting Started

# Native install (recommended)
# macOS / Linux / WSL
curl -fsSL https://claude.ai/install.sh | bash
# Windows (PowerShell)
irm https://claude.ai/install.ps1 | iex

cd your-project && claude
/init

Scans your codebase and creates a starter CLAUDE.md file.

Login — claude login — authenticates via browser

IDEs — VS Code and JetBrains extensions available

Verify — claude --version to confirm install

🧠 Understanding CLAUDE.md

Persistent memory loaded every session. Tell Claude about your project:

WHATWHYHOW

○What — tech stack, directory map, architecture

○Why — module purpose, design decisions

○How — build/test commands, workflows, gotchas

# Project: MyApp
FastAPI + React + Postgres

## Commands
npm run dev / test / lint

## Architecture
/app → routes  /lib → utils
/prisma → DB schema

📂 Memory File Hierarchy

~/.claude/CLAUDE.mdGlobal — all projects

CLAUDE.mdProject — shared on git

.claude/CLAUDE.mdProject — alternative location

./src/CLAUDE.mdSubfolder — scoped context

○Keep each file under 200 lines

○Subfolder files append context, never overwrite

○Commit to Git for team sharing

📁 Project File Structure

your-project/
├── CLAUDE.md
├── .claude/
│   ├── settings.json
│   ├── settings.local.json
│   ├── skills/
│   │   ├── code-review/
│   │   │   └── SKILL.md
│   │   └── testing/
│   │       └── SKILL.md
│   ├── commands/
│   │   └── deploy.md
│   └── agents/
│       └── reviewer.md
└── .gitignore

○settings.json — shared team config (commit to git)

○settings.local.json — personal overrides (gitignored)

○commands/ — custom slash commands the team shares

○agents/ — custom subagent definitions

⚙️ Adding Skills (The Superpower)

Skills = markdown guides Claude auto-invokes via natural language.

Project — .claude/skills/name/SKILL.md

Personal — ~/.claude/skills/name/SKILL.md

---
name: testing-patterns
description: Jest testing patterns
allowed_tools: Read, Grep, Glob
---

# Testing Patterns
Use describe + it + AAA pattern
Use factory mocks

○description field is critical for auto-activation

○Type /skill-name to invoke explicitly

🪝 Setting Up Hooks

Hooks = deterministic callbacks at lifecycle events.

PreToolUsePostToolUseNotification

"hooks": {
  "PreToolUse": [{
    "matcher": "Bash",
    "hooks": [{
      "type": "command",
      "command": "scripts/sec.sh",
      "timeout": 5
    }]
  }]
}

Exit codes — 0 → allow, 2 → block

○Input via stdin (command) or POST body (HTTP)

🔒 Permissions & Safety

{
  "permissions": {
    "allow": [
      "Read:*",
      "Bash:git:*",
      "Write:*:*.md"
    ],
    "deny": [
      "Read:env:*",
      "Bash:sudo:*"
    ]
  }
}

○Wildcard patterns: Tool:arg:glob

○allow auto-approves matching tool calls

○deny blocks matching tool calls entirely

○Set defaultMode to plan, auto, or default

🏗️ The 4-Layer Architecture

L1CLAUDE.mdPersistent context and rules

L2SkillsAuto-invoked knowledge packs

L3HooksSafety gates and automation

L4AgentsSubagents with their own context

○Each layer builds on the one below — start with L1, add layers as patterns emerge

🔄 Daily Workflow Pattern

1cd project && claude

2Shift+Tab → Plan Mode

3Describe feature intent

4Shift+Tab → Auto Accept

5! git status → inline bash

6/compact when context grows

7Ctrl+B → background a task

8Esc Esc → rewind if needed

🧠 Modes & Models

InteractivePlan ModeExtended Thinking

claude-opus-4-6 — Most capable — deep reasoning

claude-sonnet-4-6 — Balanced — daily coding

claude-haiku-4-5 — Fastest — quick tasks, lowest cost

Context — Up to 1M tokens with auto-compression

○⌥P to switch models — context preserved

○⌥T to toggle extended thinking

○/rc — remote control from claude.ai web

○/fork — branch conversation to explore alternatives

🤖 Non-Interactive & CI

# Single-shot execution
claude -p "add error handling"

# Pipe content in
cat logs.txt | claude -p \
  "summarize errors"

# JSON output for scripting
claude -p "list TODOs" \
  --output-format json

# Restrict tools in CI
claude -p "review" \
  --allowedTools Read,Grep

CI auth — Set ANTHROPIC_API_KEY env var

Exit code — 0 on success — fits pipelines

○Use in GitHub Actions, pre-commit hooks, build scripts

🔌 MCP Servers

Model Context Protocol — connect external tools via a standardized interface.

// .mcp.json (project root)
{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["server-package"]
    }
  }
}

Project — .mcp.json — shared with team via git

User — ~/.claude.json under mcpServers

○Connect databases, APIs, browsers, cloud services

○Hundreds of community servers available

🤖 Multi-Agent Parallel Work

SubagentsAgent TeamsScheduled

Subagents — Parallel workers via Agent tool, isolated

Agent Teams — Team lead + teammates, shared task list & mailbox

/loop 5m — Repeat a prompt on interval (s/m/h/d)

.claude/agents/reviewer.md
---
name: reviewer
description: Code review agent
model: claude-sonnet-4-6
tools: Read, Grep, Glob
---
Review for OWASP top 10...

○Worktrees — each agent gets isolated git copy

○Shift+Down to cycle teammates in Agent Teams

○/agents to list and manage definitions

🌿 Git & PR Workflows

○Commits — auto-generates conventional commit messages from diff

○Branches — creates feature branches, pushes upstream

○Pull requests — writes title, description, files via gh

○Code review — reads diffs, structured feedback

○Secret scanning blocks commits with API keys or tokens

# Ask Claude naturally:
"commit these changes"
"create a PR for this branch"
"review the last 3 commits"

💰 Context & Cost Management

○/compact — summarize and compress when context gets long

○/clear — full reset when session is confused

○/cost — check session spend at any time

○Auto-compresses as you approach token limits

○Start fresh sessions every ~20 turns for best quality

○Rule of thumb — /compact first; if still off, /clear and start over

📌 Quick Reference

Core Commands

/initGenerate CLAUDE.md

/compactCompress context

/clearReset conversation

/costSession spend

/forkBranch conversation

/rcRemote from claude.ai

/loop 5mRepeat on interval

/btwSide question (ephemeral)

/diffReview pending changes

/agentsList agent definitions

Key Shortcuts

⌥P / Alt+PSwitch model

⌥T / Alt+TExtended thinking

Ctrl+GText editor

Ctrl+BBackground task

Ctrl+TToggle task list

Shift+TabCycle mode

Shift+DownCycle teammates

Esc EscRewind last turn

!Bash mode (inline)

When to Use

Interactive — Exploring, debugging, learning

-p flag — CI/CD, automation, scripts

Plan mode — Complex features, unclear scope

Agent Teams — Multi-file parallel work

/loop — Monitoring, polling, recurring

/rc — Control CLI from web UI

⌥T thinking — Hard bugs, architecture

Claude Code Workflow Cheatsheet — Created 2026
Anthropic's agentic coding CLI & IDE extensions