Claude Code Cheat Sheet

⭐ Most Used Slash Commands

Command	Explanation	Typical Usage
/context	Shows current context window usage — tokens consumed vs. remaining in this session	Check before a long task to see if you need to compact or clear first
/clear	Wipes the entire conversation history — Claude has no memory of anything said before	Starting a brand-new, unrelated task; getting a clean slate
/compact	Compresses conversation history into a summary, freeing context while keeping the working thread	Conversation is getting long but you want to continue the same task without losing context
/plan	Enters plan mode — Claude outlines its approach and waits for approval before taking any action	Before a risky or multi-step task; when you want to review the approach before Claude touches any files

↩️ After Every Reply — What You Can Do

Continue

Send "continue" or "keep going" if Claude stopped mid-task. It will pick up exactly where it left off.

Rewind

Edit a previous message in the thread to re-run the conversation from that point — useful to try a different approach without re-typing context.

Clear /clear

Wipe history and start completely fresh. Use when moving to a new, unrelated task.

Compact /compact

Compress the conversation into a summary. Use when context is running low but the task isn't done.

Sub-agent

Ask Claude to spawn a sub-agent for a parallel or isolated subtask — e.g. "use an agent to research X while we continue". Results return as tool output.

💬 Other Slash Commands

/bug Report a bug to Anthropic with session context attached

/btw Add a silent note to Claude's context — no response generated; useful for passing context mid-task

/config View or change model, theme, and other settings

/cost Show token usage and estimated cost for this session

/doctor Run diagnostics — checks API key, config, connectivity

/help Show available commands and shortcuts

/init Generate a CLAUDE.md for the current codebase

/mcp List connected MCP servers and their status

/memory View and manage persistent memory files

/model Switch model mid-session (e.g. opus, sonnet)

/review Review a pull request — opens PR review skill

/simplify Review changed code for reuse, quality, and efficiency — then fix any issues found

/status Show current model, session ID, and working directory

📄 CLAUDE.md

Place a CLAUDE.md at the project root to give Claude persistent context — conventions, stack, commands to know, what to avoid.

/init Auto-generate a CLAUDE.md from the codebase

Claude reads CLAUDE.md at session start. Checked-in files inform all collaborators; .claude/settings.local.json is for personal overrides.

Global instructions: ~/.claude/CLAUDE.md — applies to all projects on this machine.

⚠️ Aim to keep CLAUDE.md under 200 lines. Shorter files load faster and stay focused — link out to docs rather than pasting them in.

⌨️ Keyboard Shortcuts

Enter Submit message

Shift+Enter New line without submitting

Ctrl+C Cancel current response or interrupt running tool

Ctrl+C Ctrl+C Exit Claude Code (double press)

↑ / ↓ Navigate prompt history

Tab Autocomplete file paths and slash commands

Ctrl+L Clear terminal output (doesn't reset conversation)

Esc Cancel pending input / close autocomplete

🚩 CLI Flags

--print / -p Non-interactive: print response and exit (pipe-friendly)

--continue / -c Resume the most recent conversation

--resume <id> Resume a specific session by ID

--model / -m Set model: claude-opus-4-7, claude-sonnet-4-6, etc.

--add-dir <path> Add an extra directory to Claude's context

--allowedTools Comma-separated list of tools to permit

--output-format text (default), json, or stream-json

--max-turns Limit agentic turns in non-interactive mode

--verbose Show full tool call details and timing

--version Print installed version and exit

🔒 Permissions & Trust

--dangerously-skip-permissions Skip all permission prompts — use in trusted, isolated environments only

Permissions are stored in .claude/settings.json (project) or ~/.claude/settings.json (global). Use /update-config skill to manage them.

Trust levels: allow (auto-approve), deny (auto-reject), or prompt (default). Applies per tool, per path prefix, or per command pattern.

🧠 Memory

Claude Code has a file-based memory system at ~/.claude/projects/<project>/memory/. Memories persist across sessions.

/memory View, search, and manage memory files

Types: user (preferences), feedback (corrections + confirmations), project (context, deadlines), reference (where things live).

Say "remember that…" to save something. Say "forget that…" to remove it. Claude saves proactively when it learns something non-obvious.

🪝 Hooks

Hooks are shell commands that run automatically in response to events — configured in settings.json under a hooks key.

PreToolUse Runs before a tool call — can block it by exiting non-zero

PostToolUse Runs after a tool completes — good for linting, logging

Notification Fires when Claude sends a desktop notification

Stop Runs when Claude finishes a response

Use the /update-config skill to add hooks ("run X after every file edit", "whenever Claude stops…").

🔌 MCP Servers

Model Context Protocol — extend Claude with external tools (databases, APIs, services). Configured in settings.json under mcpServers.

/mcp List connected MCP servers and available tools

Scopes: project (.claude/settings.json), user (~/.claude/settings.json), or via --mcp-config <file> flag.

Server types: stdio (local process) or http (remote). Each server exposes tools Claude can call like any built-in tool.

🤖 Agentic / Headless Use

Run one-shot tasks from the shell — useful for CI, scripts, and automation.

-p "prompt" Print mode: run prompt, output result, exit

Example: claude -p "summarise changes in this PR" --output-format json

Pipe input: cat file.py | claude -p "review this" — stdin is included in context.

Sub-agents: Use the SDK's Agent tool to spawn parallel sub-agents. Each starts cold — write self-contained prompts. Results return as tool output.

💡 Tips & Patterns

Run shell commands yourself: prefix with ! in the prompt — e.g. ! git log --oneline -10 — output lands directly in the conversation.

Context management: use /compact when the conversation gets long; use /clear for a clean slate on a new task.

Exploratory questions ("what could we do about X?") get a 2-3 sentence recommendation. Claude won't implement until you say go.

Images & screenshots: drag a file into the prompt or reference a path — Claude can read PNG/JPG/PDF and Jupyter notebooks.

Parallel agents: send multiple Agent tool calls in a single message to run them concurrently and cut wall-clock time on independent tasks.

Worktrees: use isolation: "worktree" on agents to give them an isolated git branch — auto-cleaned if no changes are made.

Models (2025): Opus 4.7 (claude-opus-4-7) · Sonnet 4.6 (claude-sonnet-4-6) · Haiku 4.5 (claude-haiku-4-5-20251001). Default to Sonnet for cost/quality balance; Opus for complex reasoning.