# Claude Code Quick Reference Card # Squad: claude-code-mastery # Last updated: 2026-03-02 version: "1.0.0" claude_code_version: "1.0.x" # --------------------------------------------------------------------------- # INTERNAL TOOLS (16+) # --------------------------------------------------------------------------- tools: file_operations: - name: Read description: "Read file contents with optional line range" supports: "text, images, PDFs (page ranges), Jupyter notebooks" - name: Write description: "Create new files or complete rewrites" note: "Must Read file first if it exists" - name: Edit description: "Exact string replacement in existing files" note: "old_string must be unique; use replace_all for bulk" - name: MultiEdit description: "Multiple edits in a single file atomically" search: - name: Glob description: "Fast file pattern matching (e.g., **/*.ts)" returns: "File paths sorted by modification time" - name: Grep description: "Content search using ripgrep regex" modes: [content, files_with_matches, count] execution: - name: Bash description: "Execute shell commands with timeout control" max_timeout_ms: 600000 supports: "background execution via run_in_background" - name: Agent description: "Spawn subagent for complex multi-step tasks" types: [Explore, Plan, general-purpose, custom] web: - name: WebSearch description: "Search the web for current information" - name: WebFetch description: "Fetch content from a specific URL" context: - name: TodoRead description: "Read current todo/task list" - name: TodoWrite description: "Update todo/task list items" - name: TaskOutput description: "Read output from background tasks" mcp_tools: description: "Tools from connected MCP servers (mcp__server__tool)" pattern: "mcp__{server_name}__{tool_name}" # --------------------------------------------------------------------------- # PERMISSION MODES # --------------------------------------------------------------------------- permission_modes: - name: askAlways description: "Prompt before every tool use" badge: "[Ask]" use_case: "Maximum control, learning, auditing" - name: acceptEdits description: "Auto-approve reads/searches, ask for writes/executions" badge: "[Edits]" use_case: "Balanced safety for daily development" - name: autoApprove description: "Auto-approve all tool uses" badge: "[Auto]" use_case: "YOLO mode, trusted automated workflows" cli_flag: "--dangerously-skip-permissions" - name: plan description: "Read-only exploration, no modifications allowed" badge: "[Plan]" use_case: "Architecture review, codebase exploration" # --------------------------------------------------------------------------- # HOOK EVENTS (17) # --------------------------------------------------------------------------- hook_events: tool_lifecycle: - event: PreToolUse description: "Before any tool execution" use_cases: [damage-control, security-gate, logging] matcher_fields: [tool_name, tool_input] - event: PostToolUse description: "After tool execution completes" use_cases: [auto-lint, metrics, notification] matcher_fields: [tool_name, tool_input, tool_output] session_lifecycle: - event: Stop description: "When the agent turn ends" use_cases: [cost-tracking, summary-generation, cleanup] - event: SubagentStart description: "When a subagent is spawned" use_cases: [logging, resource-tracking] - event: SubagentStop description: "When a subagent completes" use_cases: [result-aggregation, cleanup] context_management: - event: PreCompact description: "Before context window compaction" use_cases: [state-preservation, important-context-save] - event: UserPromptSubmit description: "When user submits a prompt" use_cases: [input-validation, routing, logging] - event: Notification description: "When a notification is triggered" use_cases: [slack-integration, email-alerts, desktop-notify] advanced_events: - event: PreToolUse (Bash) description: "Matcher for Bash tool specifically" example_matcher: "tool_name == 'Bash'" - event: PreToolUse (Write) description: "Matcher for Write tool specifically" example_matcher: "tool_name == 'Write'" - event: PostToolUse (Edit) description: "After Edit tool completes" example_matcher: "tool_name == 'Edit'" hook_config_format: location: "~/.claude/settings.json or .claude/settings.json" structure: | { "hooks": { "PreToolUse": [{ "matcher": "...", "hooks": [{ "type": "command", "command": "..." }] }], "PostToolUse": [{ "matcher": "", "hooks": [{ "type": "command", "command": "..." }] }], "Stop": [{ "matcher": "", "hooks": [{ "type": "command", "command": "..." }] }] } } # --------------------------------------------------------------------------- # SUBAGENT TYPES # --------------------------------------------------------------------------- subagent_types: builtin: - name: Explore description: "Read-only exploration subagent" permissions: "Can read files, search, but cannot write" use_case: "Safe codebase exploration and analysis" - name: Plan description: "Planning subagent that creates execution plans" permissions: "Read-only, outputs structured plan" use_case: "Architecture decisions, implementation planning" custom: - name: general-purpose description: "Full-capability subagent" permissions: "Inherits parent permissions" use_case: "Delegated implementation tasks" - name: custom-agents location: ".claude/agents/*.md" description: "User-defined agent personas with custom instructions" activation: "claude --agent agents/my-agent.md" # --------------------------------------------------------------------------- # SETTINGS HIERARCHY (highest to lowest priority) # --------------------------------------------------------------------------- settings_hierarchy: - level: 1 name: "Managed Policy" file: "~/.claude/policies/*.md" description: "Enterprise/org-level policies, cannot be overridden" scope: global - level: 2 name: "CLI Flags" description: "Runtime flags (--dangerously-skip-permissions, --model)" scope: session - level: 3 name: "Project Local" file: ".claude/settings.local.json" description: "Per-project settings, gitignored" scope: project - level: 4 name: "Project Shared" file: ".claude/settings.json" description: "Shared project settings, committed to git" scope: project - level: 5 name: "User Global" file: "~/.claude/settings.json" description: "User-level defaults across all projects" scope: global # --------------------------------------------------------------------------- # MCP TRANSPORTS # --------------------------------------------------------------------------- mcp_transports: - name: stdio description: "Standard I/O communication (most common)" config_key: "command" example: '{"command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem"]}' use_case: "Local MCP servers, npm packages" - name: http_streamable description: "HTTP Streamable transport (newer, recommended for remote)" config_key: "url" example: '{"url": "https://mcp.example.com/sse"}' use_case: "Remote MCP servers, cloud services" - name: sse description: "Server-Sent Events (legacy, being replaced by HTTP Streamable)" config_key: "url" example: '{"url": "http://localhost:3001/sse"}' use_case: "Legacy remote MCP servers" # --------------------------------------------------------------------------- # MEMORY SYSTEM # --------------------------------------------------------------------------- memory_system: project_instructions: - file: "CLAUDE.md" location: "project root" description: "Primary project instructions, always loaded" priority: highest - file: ".claude/CLAUDE.md" location: ".claude directory" description: "Additional project instructions" priority: high - file: "~/.claude/CLAUDE.md" location: "user home" description: "Global user instructions across all projects" priority: medium rules: location: ".claude/rules/*.md" description: "Contextual rules loaded based on file path patterns" frontmatter_key: "paths" example: "paths: ['src/**/*.ts'] loads only when editing TypeScript files" auto_memory: description: "Claude Code can auto-generate memory from conversations" storage: ".claude/memory/" behavior: "Persists learnings across sessions" subagent_memory: description: "Subagents inherit parent context but have isolated memory" pattern: "Parent CLAUDE.md + agent-specific .md file" # --------------------------------------------------------------------------- # KEYBOARD SHORTCUTS # --------------------------------------------------------------------------- keyboard_shortcuts: - key: "Ctrl+C" action: "Cancel current operation / clear input" - key: "Ctrl+R" action: "Resume last conversation" - key: "Escape" action: "Clear current input line" - key: "Up/Down" action: "Navigate command history" - key: "Tab" action: "Accept autocomplete suggestion" - key: "Ctrl+L" action: "Clear terminal screen" - key: "/" action: "Open slash command menu" - key: "#" action: "File reference autocomplete" - key: "@" action: "Agent/mention autocomplete" # --------------------------------------------------------------------------- # CLI FLAGS (common) # --------------------------------------------------------------------------- cli_flags: - flag: "--model" description: "Override model (e.g., claude-sonnet-4-20250514)" - flag: "--agent" description: "Load custom agent from .md file" - flag: "--resume" description: "Resume a previous conversation" - flag: "--print / -p" description: "Headless mode, output to stdout (for CI/CD)" - flag: "--output-format" values: [text, json, stream-json] description: "Output format for headless mode" - flag: "--dangerously-skip-permissions" description: "Auto-approve all tool uses (YOLO)" - flag: "--allowedTools" description: "Restrict which tools the agent can use" - flag: "--max-turns" description: "Limit agent turns in headless mode" - flag: "--add-dir" description: "Add additional directory to context" - flag: "--permission-mode" values: [default, plan, autoApprove] description: "Set permission mode for session"