Free Guide

Claude Code Power User Guide

Sessions, permissions, skills, MCP, hooks, subagents, and workflows. No fluff.

By Joon | AI Agent Architect, Founder of AI Topia · April 2026

12 Sections
Complete Reference
MCP & Skills Guide
100% Free

1Session Management

CommandWhat It Does
claudeNew session in current directory
claude -cResume most recent session
claude -rBrowse and pick from recent sessions
claude -n "name"Name a session at start
/resumeSwitch sessions without exiting
/compactSummarize history to free context
/clearWipe context and start fresh

Pro Tip

Name sessions for active features (claude -n "auth-refactor"). Resume with claude -r the next day.

2Permission Modes

Cycle with Shift+Tab

ModeBehavior
defaultPrompts for every dangerous operation
acceptEditsAuto-accepts file edits, prompts for commands
planRead-only. No edits, no execution
autoAI classifiers block danger, safe ops run freely

When to use Plan Mode

Any task that touches 5+ files or crosses system boundaries. Available tools: Read, Glob, Grep, WebFetch, WebSearch.

3Extended Thinking (Adaptive)

Claude decides dynamically when and how deeply to think. Toggle: Alt+T

Trigger WordDepth
"think"Light reasoning
"think harder"Deeper analysis
"ultrathink"Maximum depth

Ultraplan (Research Preview)

Planning runs on Anthropic's cloud with Opus 4.6 + 1M context for up to 30 minutes. Terminal stays free while planning runs remotely.

4CLAUDE.md & Memory

Persistent memory files. More specific overrides on conflict. Keep each under 200 lines.

LevelPathScope
Global~/.claude/CLAUDE.mdAll projects
Project/root/CLAUDE.mdShared (committed)
Directory/root/frontend/CLAUDE.mdSubdirectory-scoped
Personal/root/CLAUDE.local.mdGitignored, private

What to include:

Tech stackConventionsBuild commandsArchitectureTesting rulesGotchas

Auto Memory

Claude writes notes to ~/.claude/memory/MEMORY.md based on your corrections and preferences.

5Anatomy of the .claude Folder

your_project/
├── CLAUDE.md              # Project instructions
├── CLAUDE.local.md        # Personal, gitignored
├── .claude/
│   ├── settings.json       # Permissions & env
│   ├── settings.local.json
│   ├── skills/
│   │   ├── code-review/
│   │   │   └── SKILL.md
│   │   └── testing/
│   │       └── SKILL.md
│   ├── commands/
│   │   └── deploy.md
│   ├── agents/
│   │   └── security-reviewer.md
│   └── arc/                # Session archives
├── .mcp.json              # Shared MCP config
└── .gitignore

settings.json

{
  "permissions": {
    "allow": ["Read:*", "Bash:git:*"],
    "deny": ["Read:env:*", "Bash:sudo:*"]
  },
  "hooks": {
    "PostToolUse": [{
      "matcher": "Write",
      "hooks": [{
        "type": "command",
        "command": "prettier --write $FILE"
      }]
    }]
  },
  "env": {
    "MAX_THINKING_TOKENS": "10000"
  }
}

6Skills (The Superpower)

Markdown guides Claude auto-invokes via natural language. Description field is critical for auto-activation.

Project Skill

.claude/skills/<name>/SKILL.md

Personal Skill

~/.claude/skills/<name>/SKILL.md
---
name: testing-patterns
description: Jest testing patterns
allowed tools: Read, Grep, Glob
---
Use describe + it + AAA pattern.
Use factory mocks. No snapshots.

Skill Ideas for Engineers:

code-reviewtestingdeployapi-designsecurity-auditcommit-messages

7Hooks System

Run shell commands, prompts, or sub-agents at lifecycle events. Browse with /hooks.

CommandPromptAgent
EventUse Case
PreToolUseBlock edits to sensitive files, validate inputs
PostToolUseAuto-format on every file write
SessionStartLoad context on launch
PreCommitSecret detection
NotificationSlack/webhook alerts

Exit code 2 = Block

Return exit code 2 from a hook to block the action. Use this for guardrails (prevent edits to .env, block sudo, etc).

8Subagents & Worktrees

Independent Claude instances with own context, system prompt, and scoped tool access.

Research

Read-only, parallel exploration across the codebase

Implementation

Full edit access, scoped to specific files or directories

Worktree-Isolated

claude -w feature-name creates isolated git branch

Agent Teams

Peer-to-peer coordination via mailbox system (experimental)

# Enable Agent Teams (experimental)
{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

9MCP Servers (Model Context Protocol)

Connect Claude Code to external tools and data. 20,000+ public servers. Scopes: local user project

Setup Commands:

# Add a server
claude mcp add <name> <command> [args]

# Import from Claude Desktop
claude mcp add-from-claude-desktop

# List / Remove
claude mcp list
claude mcp remove <name>

Quick Install:

# Context7 - current library docs
claude mcp add context7 \
  npx -y @anthropic-ai/context7-mcp

# Brave Search
claude mcp add brave-search \
  npx -y @anthropic-ai/brave-search-mcp

# Filesystem
claude mcp add filesystem \
  npx -y @anthropic-ai/filesystem-mcp /dir

Top MCP Servers:

GitHub (PRs, issues, repos)Supabase (DB, migrations)Context7 (1000+ lib docs)Firecrawl (web scrape)Brave Search (web search)Playwright (browser)Sequential ThinkingPostgreSQL (direct DB)NotionLinearSlackGoogle Drive

10Keyboard Shortcuts

ActionShortcut
Interrupt generationEscape
Rewind modeEscape x2
Cycle permission modesShift+Tab
Stash current promptCtrl+S
Search command historyCtrl+R
External editor inputCtrl+G
Kill all bg agentsCtrl+F x2
Toggle extended thinkingAlt+T
Model pickerAlt+P
Multi-line modeAlt+M
Side question/btw
Show all shortcuts?

Custom keybindings: ~/.claude/keybindings.json

11Rewind & Plugins

Rewind / Checkpointing

Press Escape twice or type /rewind

ConversationRoll back chat, keep file changes
FilesRestore files, keep conversation
BothFull rewind to checkpoint

Plugins Marketplace

CommandAction
/pluginOpen plugin manager (Discover/Installed/Updates)
/plugin install <name>Install from official marketplace
/reload-pluginsActivate after install

101 official plugins (33 by Anthropic). Community at claudemarketplaces.com and buildwithclaude.com (497+).

12Daily Workflow Pattern

cd project && claudeShift+Tab → PlanDescribe intentShift+Tab → Auto/compact when slowCommit frequently
1

Morning Standup

claude -c then "What did we work on yesterday? Summarize changes and what's left."

2

Parallel Research

Launch multiple subagents to explore different parts of the codebase. Findings return without polluting main context.

3

Context Optimization

Use /compact when Claude gets slow. /btw for side questions. Ctrl+S to stash prompts mid-flow.

4

MCP-Powered Dev

Connect Supabase + GitHub + Brave Search. Claude queries your DB, checks PRs, and researches without leaving the terminal.

Quick Reference Card

Start

claudeNew session
claude -cResume
claude -wWorktree
/initGen CLAUDE.md

Navigate

/compactFree space
/clearFresh start
Esc x2Rewind
/btwSide Q

Modes

Shift+TabCycle modes
Alt+TThinking
Alt+PModel picker
?All shortcuts

MCP

mcp addAdd server
mcp listList servers
/pluginPlugin mgr
/hooksBrowse hooks

Want to see this in action?

Join 1,000+ builders automating their GTM with AI. Get templates, workflows, and live support.