Claude Code Cheat Sheet

As of 2026-10-24, checked against code.claude.com/docs (CLI reference, commands, settings, permissions, memory, hooks, skills, subagents and MCP pages). Claude Code ships frequently; run /help or claude --help for the version you have. Full lessons: 09 · Claude Code.

1. Starting and resuming

Command	What it does
claude	Start an interactive session in the current directory
claude "explain this project"	Start interactively with a first prompt
claude -p "query"	Print mode: answer once and exit (scripts, CI)
cat log.txt \| claude -p "explain"	Pipe content into print mode
claude -c	Continue the most recent conversation in this directory
claude -r "<id-or-name>"	Resume a specific session (or open the picker)
claude update	Update to the latest version
claude doctor	Read-only install and settings diagnostics
claude auth login / status / logout	Manage sign-in (--console for API billing)
claude mcp ...	Add, list, remove and log in to MCP servers

2. Useful flags

Flag	Purpose
--model <alias-or-id>	Model for this session (opus, sonnet, haiku, fable, or a full ID)
--effort <level>	low, medium, high, xhigh, max (availability depends on model)
--permission-mode <mode>	Start in default, acceptEdits, plan, auto, dontAsk or bypassPermissions
--allowedTools "Bash(npm test *)" "Read"	Pre-approve tools for this run
--disallowedTools "Bash(rm *)"	Deny rules; a bare name removes the tool entirely
--output-format text\|json\|stream-json	Print-mode output format
--max-turns N	Cap agentic turns in print mode
--append-system-prompt "..."	Add text to the default system prompt
--mcp-config <file-or-json>	Load MCP servers for this run
--add-dir <path>	Give file access to another directory
--settings <file-or-json>	One-session settings override (below managed)
--dangerously-skip-permissions	Same as bypassPermissions. Only in isolated, disposable environments

3. In-session slash commands (selection)

Command	Use
/init	Generate a starter CLAUDE.md for the project
/memory	Edit CLAUDE.md files and manage auto memory
/clear	Start a fresh conversation (empty context)
/compact [focus]	Summarise the conversation to free context
/context	Visualise what is using the context window
/model, /effort	Switch model or effort level
/permissions	View and edit allow / ask / deny rules
/hooks	View hook configuration
/mcp	Manage MCP servers, including OAuth sign-in
/agents	Create or manage subagents
/skills	List available skills
/plugin	Manage plugins and marketplaces
/config, /status	Settings dialog; status shows loaded setting sources
/usage (/cost)	Session cost and plan usage
/rewind	Roll back conversation and/or code to a checkpoint
/resume	Pick up an earlier session
/review, /security-review	Review changes; security-focused review of the branch diff
/doctor	Setup checkup
/help	All commands

Type / to see everything, including your own skills and plugin commands.

4. Files Claude Code reads

File	Scope	Purpose
CLAUDE.md or .claude/CLAUDE.md	Project, committed	Build/test commands, conventions, architecture notes
CLAUDE.local.md	Project, personal (gitignore it)	Your sandbox URLs, personal preferences
~/.claude/CLAUDE.md	User, all projects	Personal defaults
Managed CLAUDE.md (system path)	Organisation	Policies from IT/security
.claude/rules/*.md	Project	Topic rules; paths frontmatter scopes them to files
.claude/settings.json	Project, committed	Team permissions, hooks, env, plugins
.claude/settings.local.json	Project, personal	Your overrides; standing approvals land here
~/.claude/settings.json	User	Personal settings
managed-settings.json / MDM / server-managed	Organisation	Enforced policy (highest precedence)
.mcp.json	Project, committed	Shared MCP servers
~/.claude.json	User	Local and user-scope MCP servers, account state
.claude/skills/<name>/SKILL.md	Project	Skills (also ~/.claude/skills/)
.claude/commands/<name>.md	Project	Legacy custom commands (now merged into skills)
.claude/agents/<name>.md	Project	Subagents (also ~/.claude/agents/)

CLAUDE.md imports other files with @path/to/file (up to four hops). Keep each file short (the docs suggest under 200 lines); CLAUDE.md is guidance, not enforcement. To block something, use a deny rule or a hook.

5. Settings precedence (highest first)

1. Managed settings (cannot be overridden)
2. Command line (--settings)
3. Project local (.claude/settings.local.json)
4. Shared project (.claude/settings.json)
5. User (~/.claude/settings.json)

List settings such as permissions.allow merge across levels instead of replacing each other.

6. Permissions

{
  "permissions": {
    "allow": ["Bash(npm run test *)", "Bash(git diff *)", "WebFetch(domain:docs.python.org)"],
    "ask":   ["Bash(git push *)"],
    "deny":  ["Read(./.env)", "Read(./.env.*)", "Bash(curl *)", "mcp__payments__*"]
  }
}

• Evaluation order: deny → ask → allow. A broad deny beats a narrower allow.
• mcp__<server>__<tool> names one MCP tool; mcp__<server>__* names all tools on a server.
• Modes: default (Manual; prompts on first use), acceptEdits (auto-accept edits in the working directory), plan (explore read-only, propose a plan), auto (a classifier reviews actions instead of prompts), dontAsk (auto-deny anything that would prompt), bypassPermissions (skip prompts; isolated environments only).

7. Hooks

Hooks run your code at lifecycle events. Configure them in any settings file:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [{"type": "command", "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/check-bash.sh", "timeout": 30}]
      }
    ],
    "PostToolUse": [
      {"matcher": "Edit|Write", "hooks": [{"type": "command", "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/format.sh"}]}
    ]
  }
}

Common event	Fires	Typical use
SessionStart	Session begins or resumes	Load context, check environment
UserPromptSubmit	Before a prompt is processed	Add context, block secrets in prompts
PreToolUse	Before a tool runs	Block dangerous commands, enforce policy
PermissionRequest	A permission dialog would show	Auto-approve or deny programmatically
PostToolUse	After a tool succeeds	Format, lint, log
Notification	Claude needs attention	Desktop or chat alerts
Stop / SubagentStop	Claude or a subagent finishes	Run tests, enforce "done" criteria
PreCompact / PostCompact	Around compaction	Preserve key context
SessionEnd	Session ends	Clean-up, audit log

A hook receives the event as JSON on stdin (for tool events this includes tool_name and tool_input), so a script such as format.sh reads the edited file path from there. Exit codes for command hooks: 0 success (stdout may be JSON), 2 blocking error on events that can block (stderr becomes the reason), other non-blocking error. Handler types: command, http, mcp_tool, prompt, agent. Matchers: exact names, Edit|Write lists, or a regular expression such as mcp__github__.*.

8. Skills and subagents

---
name: release-notes
description: Draft release notes from merged PRs. Use when the user asks for a changelog or release notes.
disable-model-invocation: true
allowed-tools: Bash(git log *)
---
1. Collect merged PR titles since the last tag.
2. Group them as Features, Fixes, Security.

• Skill: SKILL.md in its own folder; only name and description sit in context until it is invoked (progressive disclosure). Invoke with /name or let Claude load it when relevant.
• Subagent: Markdown file with name, description, optional tools, model, permissionMode; runs in its own context window and returns a summary. Use it to keep noisy searches out of the main context.

9. MCP in Claude Code

claude mcp add --transport http docs https://mcp.example.com/mcp                 # remote (recommended)
claude mcp add --env API_KEY="$API_KEY" --transport stdio local-db -- python server.py   # name must not follow --env
claude mcp add --transport http shared --scope project https://mcp.example.com/mcp
claude mcp list

Scopes: local (default, you only, this project), project (.mcp.json, shared), user (all your projects). Authenticate remote servers with /mcp or claude mcp login <name>. SSE transport is deprecated. Environment knobs: MCP_TIMEOUT (startup), MCP_TOOL_TIMEOUT, MAX_MCP_OUTPUT_TOKENS (default 25,000).

10. Headless and CI

claude -p "Review the diff for security issues and output JSON" \
  --output-format json --permission-mode plan --max-turns 5 \
  --allowedTools "Read" "Grep" "Bash(git diff *)"

Use least-privilege tools, a turn cap and a non-interactive permission mode; never pass secrets in the prompt. For programmatic agents built on the same harness, see the Claude Agent SDK (09 · Headless, CI and the Agent SDK).

Further reading

• CLI reference · Commands
• Settings · Permissions
• Memory (CLAUDE.md) · Hooks
• Skills · Subagents · MCP