Most People Use 5% of Claude Code
They type. They hit enter. They hope. Then they complain it's slow or expensive or dumb.
The power users are doing something different. They've set up a project CLAUDE.md so they never re-explain themselves. They run plan mode before risky edits. They fan out parallel subagents. They've wired hooks that auto-lint on every file write. They have MCP servers connecting Claude to Slack, GitHub, Stripe, the browser, and their own internal tools.
None of this is hidden. It's all built in. It's just not in the welcome screen. Here's every move, organized so you can actually use it.
STEP 1 / THE FOUR MODES
The 4 Modes Most Users Don't Know Exist
Press Shift + Tab to cycle. Each mode changes how Claude behaves when it wants to use a tool or edit a file.
- Default. Claude asks before each tool. Safe. Slow. Best for unfamiliar repos.
- Auto-accept. Claude runs tools without asking. Best for grinding through obvious work.
- Plan mode. Claude makes a full plan before touching a single file. You approve. Then it executes. Use this for anything risky.
- YOLO mode. Launch with
claude --dangerously-skip-permissions. No prompts ever. Best for sandboxes and worktrees. Not for production repos.
The move: default for new repos, plan mode for migrations and refactors, auto-accept once you trust the plan.
22 KEYBOARD SHORTCUTS
The Power-User Hands
The 5 starred ones change everything. Memorize those first.
Shift + Tab★ Cycle modes (default → auto-accept → plan).
Esc Esc★ Rewind to a previous message and edit it. Time travel.
#★ Prefix anything with # and Claude writes it to memory / CLAUDE.md.
!★ Prefix to run a bash command directly without going through Claude.
@★ Reference any file with autocomplete. Loads the file into context.
EscInterrupt Claude mid-response.
Ctrl + RToggle verbose mode. See every tool input and output.
Ctrl + CCancel current input.
Ctrl + C ×2Quit Claude Code.
Ctrl + DExit session.
Ctrl + LClear screen, keep context.
Ctrl + JNewline inside the prompt.
Shift + EnterNewline (after /terminal-setup).
Option + EnterMac alternative newline.
↑ / ↓Cycle prompt history.
TabAutocomplete file paths and commands.
Ctrl + VPaste image from clipboard. Claude reads it.
Drag + dropDrop a file or image into the terminal.
Ctrl + BRun current task in background.
Ctrl + TView background tasks and todos.
Ctrl + OExpand truncated tool output.
Ctrl + _Undo the last edit inside your prompt.
32 BUILT-IN SLASH COMMANDS
Most People Know 4 of These
Type a slash, get autocomplete. The ones worth their weight in gold: /init, /compact, /agents, /hooks, /mcp, /resume.
/helpList all available commands.
/clearWipe conversation context. Fresh start.
/compactSummarize and shrink the context window without losing the thread.
/compact [focus]Compact while telling Claude what to keep. Example: /compact keep the auth refactor notes.
/initGenerate a starter CLAUDE.md for the current project. Run this first thing in any new repo.
/memoryEdit your memory files directly.
/modelSwitch model (Opus / Sonnet / Haiku) mid-session.
/configOpen settings UI.
/permissionsManage allowed tools and Bash commands.
/loginSwitch Anthropic account.
/logoutSign out.
/statusAccount, model, session info.
/costShow token usage and dollar spend for the current session.
/doctorDiagnose your install health.
/bugFile a bug report straight to Anthropic.
/reviewCode-review the current changes.
/pr-commentsRead GitHub PR comments inline.
/install-github-appInstall the Claude Code GitHub app.
/agentsManage subagents. Create, edit, list.
/mcpManage MCP server connections.
/hooksConfigure event hooks (pre-tool, post-tool, stop, prompt-submit).
/output-stylePick an output style (terse, teach-y, pair-programmer, etc).
/output-style:newMake your own output style.
/statuslineCustomize the bottom status line. Show git branch, cost, model live.
/terminal-setupInstall Shift+Enter newline binding for your terminal.
/vimToggle vim editing mode in the prompt.
/ideConnect Claude Code to VS Code or JetBrains.
/exportExport the current conversation.
/resumeResume a previous session. Pick from a list.
/upgradeUpgrade your Anthropic plan.
/release-notesWhat changed in this version.
/privacy-settingsManage data sharing.
46 WORKFLOW HACKS · PART 1 OF 6
CLI Launchers (Headless Claude)
Most people think Claude Code only works interactively. It also runs as a one-shot CLI. Perfect for scripts, cron jobs, and piping.
claude -cContinue your last session. Best shortcut on the list.
claude -rResume a specific past session by id.
claude -p "prompt"One-shot non-interactive mode. Returns the answer and exits.
--dangerously-skip-permissionsYOLO mode. No prompts. Use in sandboxes only.
--model sonnetPick the model on launch.
--append-system-promptExtend the system prompt with extra rules for this session.
--add-dir ../otherGive Claude access to another folder outside the working directory.
The pipe trick: stream any text into Claude from the shell.
cat error.log | claude -p "summarize the root cause in 3 bullets"
That one line replaces an entire log-analysis script.
WORKFLOW HACKS · PART 2 OF 6
Memory & Context (Stop Re-Explaining Yourself)
CLAUDE.md is the single highest-leverage move in this entire cheat sheet. Write one. Every session inherits it.
Project CLAUDE.mdAt repo root. Goes into git. Team-wide context.
~/.claude/CLAUDE.mdGlobal user rules. Your tools, your voice, your style.
Nested CLAUDE.mdDrop one in any subdirectory. Context scoped to that folder only.
@file.mdInject any file into the current message. Specs, schemas, examples.
# remember thisType # mid-chat and Claude writes the line to your memory file automatically.
Paste screenshotsMatch a UI, debug an error dialog, parse a chart. Claude reads images natively.
Drop logs / stack tracesPaste raw output. No formatting needed. Claude parses it.
WORKFLOW HACKS · PART 3 OF 6
Custom Commands, Agents, and Skills
Every repeating workflow you do should live in a slash command. Every specialist role should be a subagent. Every domain you work in should be a skill.
.claude/commands/Project-level slash commands. Lives in git. Shareable with the team.
~/.claude/commands/Personal slash commands. Available in every project.
.claude/agents/Custom subagents. Each one is a focused worker with its own prompt and tools.
~/.claude/agents/Global subagents. Available everywhere.
.claude/skills/Project skills folder. Domain workflows triggered by keywords.
~/.claude/skills/Global skills folder. The /onboard, /free-guide, /script types live here.
Command frontmatterAdd description and argument-hint to your custom commands so they autocomplete properly.
$ARGUMENTSPass args from the user into your custom slash command body.
WORKFLOW HACKS · PART 4 OF 6
Hooks (Automate Around Every Tool Call)
Configure in .claude/settings.json. Hooks run shell commands at specific lifecycle events. This is how you bake quality control into every session.
PreToolUseFires before a tool runs. Block dangerous commands. Confirm destructive ops.
PostToolUseFires after a tool runs. Auto-format every file Claude edits. Auto-run tests.
StopFires when Claude finishes. Send yourself a Slack ping. Play a sound.
UserPromptSubmitFires when you hit enter. Preprocess your input. Inject context. Block PII.
The killer use: a PostToolUse hook that runs prettier --write on any file Claude edits. You never see a poorly formatted file again.
WORKFLOW HACKS · PART 5 OF 6
MCP Servers (Connect Claude to Everything)
MCP is the protocol that lets Claude talk to outside services. Slack, GitHub, Stripe, browsers, your own internal tools, anything with an API.
Local MCP (stdio)Runs as a local process. Best for custom tools you build.
Remote MCP (HTTP/SSE)Hosted server. No install. OAuth in the browser.
claude mcp addAdd a server from the CLI in one line.
claude mcp listSee all configured servers.
.mcp.jsonCommit this file to git and your team gets the same MCP servers wired up automatically.
Start with two: the Slack MCP (so Claude can send messages, read channels, post into project rooms) and the GitHub MCP (so Claude can open PRs, read issues, review code). Those two alone replace a bunch of manual context-switching.
WORKFLOW HACKS · PART 6 OF 6
The Thinking Ladder & The Last 10
Claude has a hidden reasoning budget. You unlock more of it with magic words. This is real, not folklore.
- think — standard reasoning
- think hard — more budget
- think harder — more again
- ultrathink — max thinking budget
Plan → approve → buildThe gold-standard loop. Ask for a plan first, approve, then say go.
Parallel subagentsSpawn multiple Task agents in one message. They run concurrently.
Git worktreesMultiple Claude sessions on the same repo, each on its own branch. No conflicts.
Background bashLong-running commands keep going while you do other work.
Output stylesTerse, teach-y, pair-programmer. Change Claude's personality per project.
Custom status lineShow git branch, cost, model, anything you want, in real time.
claude config set theme darkFlip theme from the CLI without opening config.
/resume pickerJump back into any past session. Full context restored.
IF YOU ONLY DO 5 THINGS
The 5 Moves That Will 10x Your Output This Week
100 commands is a lot. If you only adopt 5, make it these.
- Write a CLAUDE.md at the root of every active project. Who you are, what tools you use, how you write. Run
/init to get started. - Use plan mode (Shift+Tab twice) before any non-trivial change. Approve the plan, then watch it execute.
- Spawn parallel subagents when you have independent work. Tell Claude "run these in parallel" and it dispatches multiple Tasks at once.
- Set up a PostToolUse hook that auto-formats on every edit. You stop fixing whitespace forever.
- Add 2 MCP servers. Start with Slack and GitHub. Everything else can wait.
Those 5 cover roughly 80% of the gap between casual user and power user.
Want the Printable Cheat Sheet?
All 100 commands fit on two pages. Designed to print, pin above your monitor, and screenshot for your phone.
DM me "CLAUDE"on Instagram and I'll send you the PDF.
The Next Step
Guides show you what. AI Builders shows you how.
Inside the community, I walk through every build live — including the stuff that doesn't make it into guides. Regular people (not developers) figuring out AI together, shipping real projects, asking me anything. No fluff, no theory, just the actual work.
skool.com/ten-fold