AI 原生开放教材

外观

Claude How-To工程手册·英文原文 ↗原件 ↓仓库 ↗

Lesson Quiz — Question Bank ​

10 questions per lesson. Each question has: category, question text, options (3-4), correct answer, explanation, and review section.

Lesson 01: Slash Commands ​

Q1 ​

  • Category: conceptual
  • Question: What are the four types of slash commands in Claude Code?
  • Options: A) Built-in, skills, plugin commands, MCP prompts | B) Built-in, custom, hook commands, API prompts | C) System, user, plugin, terminal commands | D) Core, extension, macro, script commands
  • Correct: A
  • Explanation: Claude Code has built-in commands (like /help, /compact), skills (SKILL.md files), plugin commands (namespaced plugin-name:command), and MCP prompts (/mcp__server__prompt).
  • Review: Overview section

Q2 ​

  • Category: practical
  • Question: How do you pass all user-provided arguments to a skill?
  • Options: A) Use ${args} | B) Use $ARGUMENTS | C) Use $@ | D) Use $INPUT
  • Correct: B
  • Explanation: $ARGUMENTS captures all text after the command name. For positional args, use $0, $1, etc.
  • Review: Arguments

Q3 ​

  • Category: conceptual
  • Question: When both a skill (.claude/skills/name/SKILL.md) and a legacy command (.claude/commands/name.md) exist with the same name, which takes priority?
  • Options: A) The legacy command | B) The skill | C) Whichever was created first | D) Claude asks the user to choose
  • Correct: B
  • Explanation: Skills take precedence over legacy commands with the same name. The skill system supersedes the older command system.
  • Review: Skill vs Command Conflict section

Q4 ​

  • Category: practical
  • Question: How do you inject live shell output into a skill's prompt?
  • Options: A) Use $(command) syntax | B) Use !command`` (backtick with !) syntax | C) Use @shell:command syntax | D) Use \{command\} syntax
  • Correct: B
  • Explanation: The !command`` syntax runs a shell command and injects its output into the skill prompt before Claude sees it.
  • Review: Dynamic Context with Shell Commands

Q5 ​

  • Category: conceptual
  • Question: What does disable-model-invocation: true do in a skill's frontmatter?
  • Options: A) Prevents the skill from running entirely | B) Allows only the user to invoke it (Claude cannot auto-invoke) | C) Hides it from the /help menu | D) Disables the skill's AI processing
  • Correct: B
  • Explanation: disable-model-invocation: true means only the user can trigger the command via /command-name. Claude will never auto-invoke it, useful for skills with side effects like deployments.
  • Review: Frontmatter Reference section

Q6 ​

  • Category: practical
  • Question: You want to create a skill that only Claude can invoke automatically (hidden from the user's / menu). Which frontmatter field do you set?
  • Options: A) disable-model-invocation: true | B) user-invocable: false | C) hidden: true | D) auto-only: true
  • Correct: B
  • Explanation: user-invocable: false hides the skill from the user's slash menu but allows Claude to invoke it automatically based on context.
  • Review: Frontmatter Reference section

Q7 ​

  • Category: practical
  • Question: What is the correct directory structure for a new custom skill called "deploy"?
  • Options: A) .claude/commands/deploy.md | B) .claude/skills/deploy/SKILL.md | C) .claude/skills/deploy.md | D) .claude/deploy/SKILL.md
  • Correct: B
  • Explanation: Skills live in a directory under .claude/skills/ with a SKILL.md file inside. The directory name matches the command name.
  • Review: Creating a Custom Command as a Skill section

Q8 ​

  • Category: conceptual
  • Question: How do plugin commands avoid name conflicts with user commands?
  • Options: A) They use a plugin-name:command-name namespace | B) They have a special .plugin extension | C) They are prefixed with p/ | D) They override user commands automatically
  • Correct: A
  • Explanation: Plugin commands use a namespace like pr-review:check-security to avoid conflicts with standalone user commands.
  • Review: Plugin commands section

Q9 ​

  • Category: practical
  • Question: You want to restrict which tools a skill can use. Which frontmatter field do you add?
  • Options: A) tools: [Read, Grep] | B) allowed-tools: [Read, Grep] | C) permissions: [Read, Grep] | D) restrict-tools: [Read, Grep]
  • Correct: B
  • Explanation: The allowed-tools field in SKILL.md frontmatter scopes which tools the command can invoke.
  • Review: Frontmatter Reference

Q10 ​

  • Category: conceptual
  • Question: What is the @file syntax used for in a skill?
  • Options: A) Importing another skill | B) Referencing a file to include its content in the prompt | C) Creating a symlink | D) Setting file permissions
  • Correct: B
  • Explanation: The @path/to/file syntax in a skill includes the referenced file's content into the prompt, allowing skills to pull in templates or context files.
  • Review: File references section

Lesson 02: Memory ​

Q1 ​

  • Category: conceptual
  • Question: How do CLAUDE.md memory files relate to each other, and what has the highest authority?
  • Options: A) Higher tiers override lower tiers entirely | B) They are concatenated into context; Managed Policy loads first and cannot be excluded | C) The most recently modified file wins | D) Project Memory has the highest priority
  • Correct: B
  • Explanation: CLAUDE.md files are concatenated into context rather than overriding each other — load order sets position in context, not precedence. Managed Policy (admin-managed) loads first and cannot be excluded by individual settings. Auto memory is a separate mechanism.
  • Review: Memory hierarchy section

Q2 ​

  • Category: practical
  • Question: How do you quickly add a new rule to memory during a conversation?
  • Options: A) Use the /memory slash command or ask conversationally | B) Prefix your message with # (e.g., # always use TypeScript) | C) Type /rule "rule text" | D) Use @add-memory "rule text"
  • Correct: A
  • Explanation: The recommended ways to add memory are the /memory command (opens memory files in your editor) or asking Claude conversationally (e.g., "remember that we always use TypeScript strict mode"). The # prefix was discontinued and no longer works.
  • Review: Quick memory updates section in README

Q3 ​

  • Category: conceptual
  • Question: What is the maximum depth for @path/to/file imports in CLAUDE.md?
  • Options: A) 3 levels deep | B) 4 levels deep | C) 10 levels deep | D) Unlimited
  • Correct: B
  • Explanation: The @path import syntax supports recursive imports up to a maximum depth of 4 hops to prevent infinite loops.
  • Review: Memory Architecture section

Q4 ​

  • Category: practical
  • Question: How do you scope a rule file to only apply to files in src/api/?
  • Options: A) Put the rule in src/api/CLAUDE.md | B) Add paths: src/api/** YAML frontmatter to a .claude/rules/*.md file | C) Name the file .claude/rules/api.md | D) Use @scope: src/api in the rule file
  • Correct: B
  • Explanation: Files in .claude/rules/ support a paths: frontmatter field with glob patterns to scope rules to specific directories.
  • Review: Path-specific rules section

Q5 ​

  • Category: conceptual
  • Question: How many lines of Auto Memory's MEMORY.md are loaded at session start?
  • Options: A) All lines | B) First 100 lines | C) First 200 lines | D) First 500 lines
  • Correct: C
  • Explanation: The first 200 lines of MEMORY.md — or the first 25 KB, whichever comes first — are auto-loaded into context at session start. Topic files referenced from MEMORY.md are loaded on demand.
  • Review: Auto Memory section

Q6 ​

  • Category: practical
  • Question: You want personal project preferences that are NOT committed to git. Which file should you use?
  • Options: A) ~/.claude/CLAUDE.md | B) CLAUDE.local.md | C) .claude/rules/personal.md | D) .claude/memory/personal.md
  • Correct: B
  • Explanation: CLAUDE.local.md in the project root is for personal project-specific preferences. It should be git-ignored.
  • Review: Memory Locations Table

Q7 ​

  • Category: conceptual
  • Question: What does the /init command do?
  • Options: A) Initializes a new Claude Code project from scratch | B) Generates a template CLAUDE.md based on your project structure | C) Resets all memory to defaults | D) Creates a new session
  • Correct: B
  • Explanation: /init analyzes your project and generates a template CLAUDE.md with suggested rules and standards. It's a one-time bootstrapping tool.
  • Review: /init command section

Q8 ​

  • Category: practical
  • Question: How do you disable Auto Memory completely?
  • Options: A) Delete the ~/.claude/projects directory | B) Set CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 | C) Add auto-memory: false to CLAUDE.md | D) Use /memory disable auto
  • Correct: B
  • Explanation: Setting CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 disables auto memory. Value 0 forces it on. Unset = default on.
  • Review: Auto Memory configuration section

Q9 ​

  • Category: conceptual
  • Question: Can a lower-priority memory tier override rules from a higher-priority tier?
  • Options: A) Yes, the most recent rule always wins | B) No — CLAUDE.md files are concatenated into context, not overridden; load order sets position, not precedence | C) Yes, if the lower tier uses the !important flag | D) It depends on the rule type
  • Correct: B
  • Explanation: CLAUDE.md files (user, project, local, rules) are concatenated into context rather than overriding each other — later files appear later in context, not "instead of" earlier ones. Only Managed Policy is special: it loads first and cannot be excluded by individual settings. Auto memory is a separate mechanism, not part of the concatenation order.
  • Review: Memory hierarchy section

Q10 ​

  • Category: practical
  • Question: You work across two repositories and want Claude to load CLAUDE.md from both. What flag do you use?
  • Options: A) --multi-repo | B) --add-dir /path/to/other | C) --include /path/to/other | D) --merge-context /path/to/other
  • Correct: B
  • Explanation: The --add-dir flag loads CLAUDE.md from additional directories, allowing multi-repo context.
  • Review: Additional directories section

Lesson 03: Skills ​

Q1 ​

  • Category: conceptual
  • Question: What are the 3 levels of progressive disclosure in the skill system?
  • Options: A) Metadata, instructions, resources | B) Name, body, attachments | C) Header, content, scripts | D) Summary, details, data
  • Correct: A
  • Explanation: Level 1: Metadata (~100 tokens, always loaded), Level 2: SKILL.md body (<5k tokens, loaded on trigger), Level 3: Bundled resources (scripts/references/assets, loaded on demand).
  • Review: How Skills Work: Progressive Disclosure

Q2 ​

  • Category: practical
  • Question: What is the most important factor for a skill to be auto-invoked by Claude?
  • Options: A) The skill's file name | B) The description field in frontmatter with when-to-use keywords | C) The skill's directory location | D) The auto-invoke: true frontmatter field
  • Correct: B
  • Explanation: Claude decides whether to auto-invoke a skill based solely on its description field. It must include specific trigger phrases and scenarios.
  • Review: Controlling Skill Invocation section

Q3 ​

  • Category: conceptual
  • Question: What is the maximum recommended length for a SKILL.md file?
  • Options: A) 100 lines | B) 250 lines | C) 500 lines | D) 1000 lines
  • Correct: C
  • Explanation: SKILL.md should be kept under 500 lines. Larger reference material belongs in references/ subdirectory files.
  • Review: Keep SKILL.md Under 500 Lines section

Q4 ​

  • Category: practical
  • Question: How do you make a skill run in an isolated subagent with its own context?
  • Options: A) Set isolation: true in frontmatter | B) Set context: fork with an agent field in frontmatter | C) Set subagent: true in frontmatter | D) Put the skill in .claude/agents/
  • Correct: B
  • Explanation: context: fork runs the skill in a separate context, and the agent field specifies which agent type (e.g., Explore, Plan, custom agent) to use.
  • Review: Running skills in subagents section

Q5 ​

  • Category: conceptual
  • Question: What is the approximate context budget allocated to skill metadata (Level 1)?
  • Options: A) 0.5% of context window | B) 1% of context window | C) 5% of context window | D) 10% of context window
  • Correct: B
  • Explanation: Skill metadata occupies about 1% of the context window (fallback: 8,000 characters). This is configurable with SLASH_COMMAND_TOOL_CHAR_BUDGET.
  • Review: the "Description budget" paragraph under Automatic Discovery

Q6 ​

  • Category: practical
  • Question: A skill needs to reference a large API specification. Where should you put it?
  • Options: A) Inline in SKILL.md | B) In a references/api-spec.md file inside the skill directory | C) In the project's CLAUDE.md | D) In a separate .claude/rules/ file
  • Correct: B
  • Explanation: Large reference material belongs in the references/ subdirectory. Claude loads Level 3 resources on demand, keeping SKILL.md lean.
  • Review: Supporting files structure section

Q7 ​

  • Category: conceptual
  • Question: What is the difference between Reference Content and Task Content in a skill?
  • Options: A) Reference is read-only, Task is read-write | B) Reference adds knowledge to context, Task provides step-by-step instructions | C) Reference is for documentation, Task is for code | D) There is no difference
  • Correct: B
  • Explanation: Reference Content adds domain knowledge to Claude's context (e.g., brand guidelines). Task Content provides actionable step-by-step instructions for a workflow.
  • Review: Skill content types section

Q8 ​

  • Category: practical
  • Question: What characters are allowed in the name field of a skill's frontmatter?
  • Options: A) Any characters | B) Lowercase letters, numbers, and hyphens only (max 64 chars) | C) Letters and underscores | D) Alphanumeric only
  • Correct: B
  • Explanation: The name must be kebab-case (lowercase, hyphens), max 64 characters, and cannot contain "anthropic" or "claude". name is itself optional — if omitted it defaults to the skill's directory name — but when supplied it must follow these rules.
  • Review: SKILL.md format section

Q9 ​

  • Category: conceptual
  • Question: In what order does Claude search for skills?
  • Options: A) User > Project > Enterprise | B) Enterprise > Personal > Project (plugin uses namespace) | C) Project > User > Enterprise | D) Alphabetical order
  • Correct: B
  • Explanation: Priority order is: Enterprise > Personal > Project. Plugin skills use a namespace (plugin-name:skill) so they don't conflict.
  • Review: Skill Types & Locations

Q10 ​

  • Category: practical
  • Question: How do you prevent Claude from automatically invoking a skill while still allowing users to use it manually?
  • Options: A) Set user-invocable: false | B) Set disable-model-invocation: true | C) Remove the description field | D) Set auto-invoke: false
  • Correct: B
  • Explanation: disable-model-invocation: true prevents Claude from auto-invoking but keeps the skill available in the user's / menu for manual use.
  • Review: Controlling Skill Invocation

Lesson 04: Subagents ​

Q1 ​

  • Category: conceptual
  • Question: What is the main advantage of subagents over inline conversation?
  • Options: A) They are faster | B) They operate in a separate, clean context window preventing context pollution | C) They can use more tools | D) They have better error handling
  • Correct: B
  • Explanation: Subagents get a fresh context window, receiving only what the main agent passes. This prevents the main conversation from being polluted with task-specific details.
  • Review: Overview section

Q2 ​

  • Category: practical
  • Question: What is the priority order for agent definitions?
  • Options: A) Project > User > CLI | B) CLI > Project > User | C) User > Project > CLI | D) They all have equal priority
  • Correct: B
  • Explanation: Full order is managed settings > --agents flag > project (.claude/agents/) > user (~/.claude/agents/) > plugin. Of the options listed, CLI > Project > User is the correct relative ordering.
  • Review: File locations section

Q3 ​

  • Category: conceptual
  • Question: Which built-in subagent is optimized for read-only codebase exploration?
  • Options: A) general-purpose | B) Plan | C) Explore | D) Bash
  • Correct: C
  • Explanation: The Explore subagent is the read-only codebase search agent. Since v2.1.198 it inherits the session model, capped at Opus — model: haiku is opt-in, not the default. It supports three thoroughness levels: quick, medium, very thorough.
  • Review: Built-in subagents section

Q4 ​

  • Category: practical
  • Question: How do you restrict which subagents a coordinator agent can spawn?
  • Options: A) Use allowed-agents: field | B) Use Task(agent_name) syntax in the tools field | C) Set spawn-limit: 2 | D) Use restrict-agents: [name1, name2]
  • Correct: B
  • Explanation: Adding Task(worker, researcher) in the tools field creates an allowlist — the agent can only spawn subagents named "worker" or "researcher".
  • Review: Restrict spawnable subagents section

Q5 ​

  • Category: conceptual
  • Question: What does isolation: worktree do for a subagent?
  • Options: A) Runs the agent in a Docker container | B) Gives the agent its own git worktree so changes don't affect the main tree | C) Prevents the agent from reading any files | D) Runs the agent in a sandbox
  • Correct: B
  • Explanation: Worktree isolation creates a separate git worktree. If the agent makes no changes, it auto-cleans up. If changes are made, the worktree path and branch are returned.
  • Review: Worktree isolation section

Q6 ​

  • Category: practical
  • Question: How do you force a subagent to always run in the background (never inline)?
  • Options: A) Set background: true in the agent config | B) Use async: true in the agent config | C) Press Ctrl+D after starting it | D) Use --background CLI flag
  • Correct: A
  • Explanation: background: true in the agent configuration makes the subagent always run as a background task and prevents it from running inline. Subagents already run in the background by default since v2.1.198; this field removes the inline option. Users can also use Ctrl+B to send a foreground task to background.
  • Review: Background subagents section

Q7 ​

  • Category: conceptual
  • Question: What does the memory field with scope project do for a subagent?
  • Options: A) Gives read access to the project CLAUDE.md | B) Creates a persistent memory directory scoped to the current project | C) Shares the main agent's conversation history | D) Loads the project's git history
  • Correct: B
  • Explanation: The memory field creates a persistent directory for the subagent. Scope project means the memory is tied to the current project. The first 200 lines of the agent's MEMORY.md auto-load.
  • Review: Persistent memory section

Q8 ​

  • Category: practical
  • Question: How do you include a phrase in a subagent's description to encourage Claude to automatically delegate tasks to it?
  • Options: A) Add "priority: high" | B) Include "use PROACTIVELY" or "MUST BE USED" in the description | C) Set auto-delegate: true | D) Add "trigger: always"
  • Correct: B
  • Explanation: Including phrases like "use PROACTIVELY" or "MUST BE USED" in the description strongly encourages Claude to automatically delegate matching tasks.
  • Review: Automatic delegation section

Q9 ​

  • Category: conceptual
  • Question: What are the valid permissionMode values for a subagent?
  • Options: A) read, write, admin | B) manual, acceptEdits, bypassPermissions, plan, dontAsk, auto | C) safe, normal, dangerous | D) restricted, standard, elevated
  • Correct: B
  • Explanation: Subagents support six permission modes: manual (prompts for everything — renamed from default in v2.1.200, which is still accepted as an alias), acceptEdits (auto-accepts file edits), bypassPermissions (skips all), plan (read-only), dontAsk (auto-denies unless pre-approved), auto (background classifier decides).
  • Review: Configuration fields section

Q10 ​

  • Category: practical
  • Question: How do you resume a subagent that returned an agentId from a previous run?
  • Options: A) Use /resume agent-id | B) Pass the resume parameter with the agentId when calling Task tool | C) Use claude -r agent-id | D) Subagents cannot be resumed
  • Correct: B
  • Explanation: Subagents can be resumed by passing the resume parameter with the previously returned agentId, continuing with full context preserved.
  • Review: Resumable agents section

Lesson 05: MCP ​

Q1 ​

  • Category: conceptual
  • Question: What are the four MCP transport protocols, and which is recommended?
  • Options: A) HTTP (recommended), Stdio, SSE (deprecated), WebSocket (ws) | B) REST (recommended), GraphQL, gRPC, Thrift | C) TCP, UDP, HTTP | D) Stdio (recommended), HTTP, SSE
  • Correct: A
  • Explanation: HTTP is recommended for remote servers. Stdio is for local processes (most common currently). SSE is deprecated but still supported. WebSocket (type: "ws") holds a persistent bidirectional connection for servers that push events unprompted; it is added with claude mcp add-json or .mcp.json, not with --transport, and supports header-only auth.
  • Review: MCP Installation Methods section

Q2 ​

  • Category: practical
  • Question: How do you add a GitHub MCP server via CLI?
  • Options: A) claude mcp install github | B) claude mcp add --transport http github https://api.github.com/mcp | C) claude plugin add github-mcp | D) claude connect github
  • Correct: B
  • Explanation: Use claude mcp add with --transport flag, a name, and the server URL. For stdio: claude mcp add github -- npx -y @modelcontextprotocol/server-github.
  • Review: MCP configuration management section

Q3 ​

  • Category: conceptual
  • Question: What happens when MCP tool descriptions exceed 10% of the context window?
  • Options: A) They are truncated | B) Tool Search auto-enables to dynamically select relevant tools | C) Claude shows an error | D) Extra tools are disabled
  • Correct: B
  • Explanation: MCP Tool Search auto-enables when tools exceed 10% of context. It requires Sonnet 4 or Opus 4 minimum (Haiku not supported).
  • Review: MCP Tool Search section

Q4 ​

  • Category: practical
  • Question: How do you use environment variable fallbacks in MCP config?
  • Options: A) ${VAR || "default"} | B) ${VAR:-default\} | C) ${VAR:default} | D) ${VAR ? "default"\}
  • Correct: B
  • Explanation: ${VAR:-default} provides a fallback value if the environment variable is not set. ${VAR\} without fallback will error if not set.
  • Review: Environment variable expansion section

Q5 ​

  • Category: conceptual
  • Question: What is the difference between MCP and Memory for data access?
  • Options: A) MCP is faster, Memory is slower | B) MCP is for live/changing external data, Memory is for persistent/static preferences | C) MCP is for code, Memory is for text | D) They are interchangeable
  • Correct: B
  • Explanation: MCP connects to live, changing external data sources (APIs, databases). Memory stores persistent, static project context and preferences.
  • Review: MCP vs Memory section

Q6 ​

  • Category: practical
  • Question: What happens when a team member first encounters a project-scoped .mcp.json?
  • Options: A) It loads automatically | B) They get an approval prompt to trust the project's MCP servers | C) It's ignored unless they opt in via settings | D) Claude asks the admin to approve
  • Correct: B
  • Explanation: Project-scoped .mcp.json triggers a security approval prompt on each team member's first use. This is intentional — it prevents untrusted MCP servers.
  • Review: MCP Scopes section

Q7 ​

  • Category: conceptual
  • Question: What does claude mcp serve do?
  • Options: A) Starts an MCP server dashboard | B) Makes Claude Code itself act as an MCP server for other applications | C) Serves MCP documentation | D) Tests MCP server connections
  • Correct: B
  • Explanation: claude mcp serve turns Claude Code into an MCP server, enabling multi-agent orchestration where one Claude instance can be controlled by another.
  • Review: Claude as MCP Server section

Q8 ​

  • Category: practical
  • Question: What is the default maximum output size for MCP tools?
  • Options: A) 5,000 tokens | B) 10,000 tokens | C) 25,000 tokens | D) 50,000 tokens
  • Correct: C
  • Explanation: Default max is 25,000 tokens (MAX_MCP_OUTPUT_TOKENS). A warning appears at 10k tokens. Disk persistence caps at 50k characters.
  • Review: MCP Output Limits section

Q9 ​

  • Category: conceptual
  • Question: When both allowedMcpServers and deniedMcpServers match a server in managed config, which wins?
  • Options: A) Allowed wins | B) Denied wins | C) The last one configured wins | D) Both are applied independently
  • Correct: B
  • Explanation: In managed MCP configuration, deny rules always take precedence over allow rules.
  • Review: Managed MCP Configuration section

Q10 ​

  • Category: practical
  • Question: How do you reference an MCP resource in a conversation?
  • Options: A) Use /mcp resource-name | B) Use @server-name:protocol://resource/path mention syntax | C) Use mcp.get("resource") | D) Resources are auto-loaded
  • Correct: B
  • Explanation: MCP resources are accessed via @server-name:protocol://resource/path mention syntax in conversation.
  • Review: MCP Resources section

Lesson 06: Hooks ​

Q1 ​

  • Category: conceptual
  • Question: What are the five types of hooks in Claude Code?
  • Options: A) Pre, Post, Error, Filter, and Transform hooks | B) Command, HTTP, MCP Tool, Prompt, and Agent hooks | C) Before, After, Around, Through, and Beside hooks | D) Input, Output, Filter, Transform, and Validate hooks
  • Correct: B
  • Explanation: Command hooks run shell scripts, HTTP hooks call webhook endpoints, MCP Tool hooks invoke an MCP tool, Prompt hooks use single-turn LLM evaluation, and Agent hooks use subagent-based verification. Agent hooks are experimental and may change.
  • Review: Hook types section

Q2 ​

  • Category: practical
  • Question: A hook script exits with code 2. What happens?
  • Options: A) Non-blocking warning shown | B) Blocking error — stderr is shown as an error to Claude, tool use is prevented | C) Hook is retried | D) Session ends
  • Correct: B
  • Explanation: Exit code 0 = success/continue, exit code 2 = blocking error (stderr shown as error), any other non-zero = non-blocking (stderr in verbose only).
  • Review: Exit codes section

Q3 ​

  • Category: conceptual
  • Question: What JSON fields does a PreToolUse hook receive on stdin?
  • Options: A) tool_name and tool_output | B) session_id, tool_name, tool_input, hook_event_name, cwd, and more | C) Only tool_name | D) The full conversation history
  • Correct: B
  • Explanation: Hooks receive a JSON object on stdin with: session_id, transcript_path, hook_event_name, tool_name, tool_input, tool_use_id, cwd, and permission_mode.
  • Review: JSON Input (via stdin)

Q4 ​

  • Category: practical
  • Question: How can a PreToolUse hook modify the tool's input parameters before execution?
  • Options: A) Return modified JSON on stderr | B) Return JSON with updatedInput field on stdout (exit code 0) | C) Write to a temp file | D) Hooks cannot modify inputs
  • Correct: B
  • Explanation: A PreToolUse hook can output JSON with "updatedInput": \{...\} on stdout (with exit 0) to modify the tool's parameters before Claude uses them.
  • Review: PreToolUse output section

Q5 ​

  • Category: conceptual
  • Question: Which hook event supports CLAUDE_ENV_FILE for persisting environment variables into the session?
  • Options: A) PreToolUse | B) UserPromptSubmit | C) SessionStart | D) All events
  • Correct: C
  • Explanation: CLAUDE_ENV_FILE is available to SessionStart, CwdChanged, and FileChanged hooks, which use it to persist environment variables into the session. Of the options listed, SessionStart is the one that supports it — PreToolUse and UserPromptSubmit do not, and it is not available to all events.
  • Review: SessionStart section

Q6 ​

  • Category: practical
  • Question: You want a hook that only runs once when a skill is first loaded, not on every tool call. What field do you add?
  • Options: A) run-once: true | B) once: true in the component hook definition | C) single: true | D) max-runs: 1
  • Correct: B
  • Explanation: Component-scoped hooks (defined in SKILL.md or agent frontmatter) support once: true to run only on first activation.
  • Review: Component-scoped hooks section

Q7 ​

  • Category: conceptual
  • Question: A Stop hook is defined in a subagent's frontmatter. What does it automatically convert to?
  • Options: A) A PostToolUse hook | B) A SubagentStop hook | C) A SessionEnd hook | D) It stays as a Stop hook
  • Correct: B
  • Explanation: When a Stop hook is placed in a subagent's frontmatter, it auto-converts to SubagentStop so it runs when that specific subagent finishes.
  • Review: Component-scoped hooks section

Q8 ​

  • Category: practical
  • Question: How do you match a hook to all MCP tools from a specific server?
  • Options: A) matcher: "mcp_github" | B) matcher: "mcp__github__.*" (regex pattern) | C) matcher: "mcp:github:*" | D) matcher: "github-mcp"
  • Correct: B
  • Explanation: Use regex patterns for matchers. MCP tools follow the mcp__server__tool naming convention, so mcp__github__.* matches all GitHub MCP tools.
  • Review: Matcher patterns section

Q9 ​

  • Category: conceptual
  • Question: How many hook events does Claude Code support in total?
  • Options: A) 16 | B) 25 | C) 33 | D) 40
  • Correct: C
  • Explanation: Claude Code supports 33 hook events. They fall into four groups: session lifecycle (SessionStart, Setup, SessionEnd), prompt and message handling (UserPromptSubmit, UserPromptExpansion, MessageDisplay), tool use and permissions (PreToolUse, PostToolUse, PermissionRequest, …), and agent, task, and environment coordination (SubagentStart, TaskCreated, CwdChanged, WorktreeCreate, PreModelSwitch, PostModelSwitch, …).
  • Review: Hook events table

Q10 ​

  • Category: practical
  • Question: You want to debug why a hook isn't firing. What's the best approach?
  • Options: A) Add print statements to the hook script | B) Use --debug flag and Ctrl+O for verbose mode | C) Check the system log | D) Hooks don't have debugging tools
  • Correct: B
  • Explanation: The --debug flag and Ctrl+O verbose mode show hook execution details including which hooks fire, their inputs, and outputs.
  • Review: Debugging section

Lesson 07: Plugins ​

Q1 ​

  • Category: conceptual
  • Question: What is the core manifest file for a plugin and where does it live?
  • Options: A) plugin.yaml in the root directory | B) .claude-plugin/plugin.json | C) package.json with a "claude" key | D) .claude/plugin.md
  • Correct: B
  • Explanation: The plugin manifest lives at .claude-plugin/plugin.json. Only name (kebab-case identifier) is required; description, version, author, homepage, repository and license are optional metadata.
  • Review: Plugin definition structure section

Q2 ​

  • Category: practical
  • Question: How do you test a plugin locally before publishing?
  • Options: A) Use /plugin test ./my-plugin | B) Use claude --plugin-dir ./my-plugin | C) Use claude plugin validate ./my-plugin | D) Copy it to ~/.claude/plugins/
  • Correct: B
  • Explanation: The --plugin-dir flag loads a plugin from a local directory for testing. It's repeatable for loading multiple plugins.
  • Review: Testing section

Q3 ​

  • Category: conceptual
  • Question: What environment variable is available inside plugin hooks and MCP configs to reference the plugin's installation directory?
  • Options: A) $PLUGIN_HOME | B) ${CLAUDE_PLUGIN_ROOT\} | C) $PLUGIN_DIR | D) ${CLAUDE_PLUGIN_PATH\}
  • Correct: B
  • Explanation: ${CLAUDE_PLUGIN_ROOT} resolves to the plugin's installed directory, enabling portable path references in hooks and MCP configs.
  • Review: Plugin Structure Example section

Q4 ​

  • Category: practical
  • Question: A plugin has a command called "check-security" in the "pr-review" plugin. How does a user invoke it?
  • Options: A) /check-security | B) /pr-review:check-security | C) /plugin pr-review check-security | D) /pr-review/check-security
  • Correct: B
  • Explanation: Plugin commands use a plugin-name:command-name namespace to avoid conflicts with user commands and other plugins.
  • Review: Plugin Commands section in Lesson 01 (Slash Commands)

Q5 ​

  • Category: conceptual
  • Question: Which components can a plugin bundle?
  • Options: A) Only commands and settings | B) Skills, agents, commands, hooks, MCP servers, LSP config, settings, templates, scripts | C) Only commands, hooks, and MCP servers | D) Only skills and agents
  • Correct: B
  • Explanation: Plugins can bundle: skills/, agents/, commands/, hooks/hooks.json, .mcp.json, .lsp.json, settings.json, templates/, scripts/, docs/, tests/. Note that commands/ is legacy — official guidance is to use skills/ for new plugins, though existing commands/ directories keep working.
  • Review: Plugin Structure Example section

Q6 ​

  • Category: practical
  • Question: How do you install a plugin from GitHub?
  • Options: A) claude plugin add github:username/repo | B) /plugin install github:username/repo | C) npm install @claude/username-repo | D) git clone then claude plugin register
  • Correct: B
  • Explanation: Use /plugin install github:username/repo to install directly from a GitHub repository.
  • Review: Installation methods section

Q7 ​

  • Category: conceptual
  • Question: What does the settings.json agent key do in a plugin?
  • Options: A) Specifies authentication credentials | B) Sets the main thread agent for the plugin | C) Lists available subagents | D) Configures agent permissions
  • Correct: B
  • Explanation: The agent key in a plugin's settings.json specifies which agent definition to use as the main thread agent when the plugin is active.
  • Review: Plugin Settings section

Q8 ​

  • Category: practical
  • Question: How do you manage plugin lifecycle (enable/disable/update)?
  • Options: A) Edit a config file manually | B) Use /plugin enable, /plugin disable, claude plugin update <name> | C) Use claude plugin-manager | D) Reinstall the plugin
  • Correct: B
  • Explanation: /plugin enable and /plugin disable are real slash commands. For updating an installed plugin, claude plugin update <name> is the form documented in the plugins reference, and the one Claude Code points you to when an update is available.
  • Review: Plugin CLI Commands section

Q9 ​

  • Category: conceptual
  • Question: What is the main advantage of a plugin over standalone skills/hooks/MCP?
  • Options: A) Plugins are faster | B) Single-command install, versioned, marketplace distribution, bundles everything together | C) Plugins have more permissions | D) Plugins work offline
  • Correct: B
  • Explanation: Plugins package multiple components into one installable unit with versioning, marketplace distribution, and automatic updates — vs. manual setup of standalone components.
  • Review: Standalone vs Plugin Approach

Q10 ​

  • Category: practical
  • Question: Where do plugin hooks configuration live within the plugin directory?
  • Options: A) .claude-plugin/hooks.json | B) hooks/hooks.json | C) plugin.json hooks section | D) .claude/settings.json
  • Correct: B
  • Explanation: Plugin hooks are configured in hooks/hooks.json within the plugin directory structure.
  • Review: Plugin Structure Example

Lesson 08: Checkpoints ​

Q1 ​

  • Category: conceptual
  • Question: What four things do checkpoints capture?
  • Options: A) Git commits, branches, tags, stashes | B) Messages, file modifications, tool usage history, session context | C) Code, tests, logs, configs | D) Inputs, outputs, errors, timing
  • Correct: B
  • Explanation: Checkpoints capture conversation messages, file modifications made by Claude's tools, tool usage history, and session context.
  • Review: Overview section

Q2 ​

  • Category: practical
  • Question: How do you access the checkpoint browser?
  • Options: A) Use /restore command | B) Press Esc + Esc (double-escape) or use /rewind | C) Use /history command | D) Press Ctrl+Z
  • Correct: B
  • Explanation: Double-escape (Esc+Esc) or the /rewind command opens the checkpoint browser to select a restore point.
  • Review: Accessing checkpoints section

Q3 ​

  • Category: conceptual
  • Question: How many rewind options are available, and what are they?
  • Options: A) 3: Undo, Redo, Reset | B) 6: Restore code+conversation, Restore conversation, Restore code, Summarize from here, Summarize up to here, Never mind | C) 2: Full restore, Partial restore | D) 4: Code, Messages, Both, Cancel
  • Correct: B
  • Explanation: The 6 options are: Restore code and conversation (full rollback), Restore conversation only, Restore code only, Summarize from here (compress forward), Summarize up to here (compress backward), Never mind (cancel). The two summary options give bidirectional context compaction.
  • Review: Rewind options section

Q4 ​

  • Category: practical
  • Question: You used rm -rf temp/ via Bash in Claude Code, then want to rewind. Will the checkpoint restore those files?
  • Options: A) Yes, checkpoints capture everything | B) No, Bash filesystem operations (rm, mv, cp) are not tracked by checkpoints | C) Only if you used the Edit tool instead | D) Only if autoCheckpoint was enabled
  • Correct: B
  • Explanation: Checkpoints track file changes made by Claude's own Write/Edit tools in the main thread. Bash commands (rm, mv, cp), subagent edits (except a foreground context: fork skill), edits made outside Claude Code, and symlinked/hardlinked paths are all outside checkpoint tracking.
  • Review: Limitations section

Q5 ​

  • Category: conceptual
  • Question: How long are checkpoints retained?
  • Options: A) Until session ends | B) 7 days | C) 30 days | D) Indefinitely
  • Correct: C
  • Explanation: Checkpoints persist across sessions for up to 30 days, after which they are automatically cleaned up. Retention is also capped by count: only the 100 most recent checkpoints per session are kept, so older ones are dropped even inside the 30-day window. The window is governed by cleanupPeriodDays.
  • Review: Checkpoint Retention section

Q6 ​

  • Category: practical
  • Question: What does "Summarize from here" do when rewinding?
  • Options: A) Deletes the conversation from that point | B) Compresses the conversation into an AI-generated summary while preserving the original in the transcript | C) Creates a bullet-point list of changes | D) Exports the conversation to a file
  • Correct: B
  • Explanation: Summarize compresses the conversation into a shorter AI-generated summary. The original full text is preserved in the transcript file.
  • Review: Rewind Options section

Q7 ​

  • Category: conceptual
  • Question: When are checkpoints created automatically?
  • Options: A) Every 5 minutes | B) On every user prompt | C) Only when you manually save | D) After every tool use
  • Correct: B
  • Explanation: Automatic checkpoints are created with every user prompt, capturing the state before Claude processes the request.
  • Review: Automatic checkpoints section

Q8 ​

  • Category: practical
  • Question: How do you disable automatic checkpoint creation?
  • Options: A) Use --no-checkpoints flag | B) Set fileCheckpointingEnabled: false in settings | C) Delete the checkpoints directory | D) Checkpoints cannot be disabled
  • Correct: B
  • Explanation: Set fileCheckpointingEnabled: false in your configuration to disable automatic checkpoint creation (default is true; requires v2.1.119+). Env equivalent: CLAUDE_CODE_DISABLE_FILE_CHECKPOINTING.
  • Review: Configuration section

Q9 ​

  • Category: conceptual
  • Question: Are checkpoints a replacement for git commits?
  • Options: A) Yes, they're more powerful | B) No, they are complementary — checkpoints are session-scoped and expire, git is permanent and shareable | C) Yes, for small projects | D) Only in solo development
  • Correct: B
  • Explanation: Checkpoints are temporary (30-day retention), session-scoped, and cannot be shared. Git commits are permanent, auditable, and shareable. Use both together.
  • Review: Integration with git section

Q10 ​

  • Category: practical
  • Question: You want to compare two different approaches. What's the recommended checkpoint workflow?
  • Options: A) Create two separate sessions | B) Checkpoint before approach A, try it, rewind to checkpoint, try approach B, compare results | C) Use git branches instead | D) There's no good way to compare approaches
  • Correct: B
  • Explanation: The branching strategy: checkpoint at clean state, try approach A, note results, rewind to the same checkpoint, try approach B. Compare both outcomes.
  • Review: Workflow patterns section

Lesson 09: Advanced Features ​

Q1 ​

  • Category: conceptual
  • Question: What are the six permission modes in Claude Code?
  • Options: A) read, write, execute, admin, root, sudo | B) manual, acceptEdits, plan, auto, dontAsk, bypassPermissions | C) safe, normal, elevated, admin, unrestricted, god | D) view, edit, run, deploy, full, bypass
  • Correct: B
  • Explanation: The six modes are: manual (prompts for everything — renamed from default in v2.1.200, which is still accepted as an alias), acceptEdits (auto-accepts file edits), plan (read-only analysis), auto (background classifier decides), dontAsk (auto-denies unless pre-approved), bypassPermissions (skips all checks).
  • Review: Permission Modes section

Q2 ​

  • Category: practical
  • Question: How do you activate planning mode?
  • Options: A) Only via /plan command | B) Via /plan, Shift+Tab/Alt+M, --permission-mode plan flag, or default config | C) Via --planning flag only | D) Planning is always on
  • Correct: B
  • Explanation: Planning mode can be activated multiple ways: /plan command, Shift+Tab/Alt+M keyboard shortcut, --permission-mode plan CLI flag, or as a default in config.
  • Review: Planning Mode section

Q3 ​

  • Category: conceptual
  • Question: What does the opusplan model alias do?
  • Options: A) Uses only Opus for everything | B) Uses Opus for planning phase and Sonnet for implementation | C) Uses a special planning-optimized model | D) Enables plan mode automatically
  • Correct: B
  • Explanation: opusplan is a model alias that uses Opus for the planning phase (higher quality analysis) and Sonnet for the execution phase (faster implementation).
  • Review: Planning Mode section

Q4 ​

  • Category: practical
  • Question: How do you toggle extended thinking on or off during a session?
  • Options: A) Type /effort max | B) Press Option+T (macOS) or Alt+T | C) Include "ultrathink" in prompt | D) It's always enabled and cannot be toggled
  • Correct: B
  • Explanation: Option+T (macOS) or Alt+T toggles extended thinking on/off for the session. (Ctrl+O toggles verbose mode to show/hide the reasoning text.) For one-off deep reasoning, include "ultrathink" in your prompt; for session-level control, use /effort command.
  • Review: Extended Thinking section

Q5 ​

  • Category: conceptual
  • Question: Does the "ultrathink" keyword trigger deep reasoning?
  • Options: A) Yes, it triggers deep reasoning for one response without changing session settings | B) No, it's treated as regular prompt text | C) Yes, but only on Opus 4.6 | D) Yes, and it permanently changes the effort level
  • Correct: A
  • Explanation: Including "ultrathink" in your prompt adds an in-context instruction for the model to reason more on that turn. It does not change the effort level sent to the API—use /effort max for session-level deep reasoning.
  • Review: Extended Thinking section

Q6 ​

  • Category: practical
  • Question: How do you run Claude in a CI/CD pipeline with structured JSON output and a turn limit?
  • Options: A) claude --ci --json --limit 3 | B) claude -p --output-format json --max-turns 3 "review code" | C) claude --pipeline --format json | D) claude run --json --turns 3
  • Correct: B
  • Explanation: Print mode (-p) with --output-format json and --max-turns is the standard CI/CD integration pattern.
  • Review: Headless Mode

Q7 ​

  • Category: conceptual
  • Question: What does the Task List feature (Ctrl+T) provide?
  • Options: A) A list of running background processes | B) A persistent to-do list that survives context compaction, shareable via CLAUDE_CODE_TASK_LIST_ID | C) A history of past sessions | D) A queue of pending tool calls
  • Correct: B
  • Explanation: The Task List (Ctrl+T) is persistent across context compactions and can be shared across sessions via named task directories using CLAUDE_CODE_TASK_LIST_ID.
  • Review: Task List section

Q8 ​

  • Category: practical
  • Question: How do you edit a plan externally (in your preferred editor) during planning mode?
  • Options: A) Copy-paste from the terminal | B) Press Ctrl+G to open the plan in an external editor | C) Use /export-plan command | D) Plans can't be edited externally
  • Correct: B
  • Explanation: Ctrl+G opens the current plan in your configured external editor for modification.
  • Review: Planning Mode section

Q9 ​

  • Category: conceptual
  • Question: What is the difference between dontAsk and bypassPermissions modes?
  • Options: A) They are the same | B) dontAsk auto-denies unless pre-approved; bypassPermissions skips all checks entirely | C) dontAsk is for files; bypassPermissions is for commands | D) bypassPermissions is safer
  • Correct: B
  • Explanation: dontAsk auto-denies permission requests unless they match pre-approved patterns. bypassPermissions skips all safety checks entirely — it's dangerous for routine use.
  • Review: Permission Modes section

Q10 ​

  • Category: practical
  • Question: How do you hand off a CLI session to the desktop app?
  • Options: A) Use /export command | B) Use /desktop command | C) Copy the session ID and paste in the app | D) Sessions can't transfer between CLI and desktop
  • Correct: B
  • Explanation: The /desktop command hands off the current CLI session to the native desktop application for visual diff review and multi-session management.
  • Review: Desktop App section

Lesson 10: CLI Reference ​

Q1 ​

  • Category: conceptual
  • Question: What are the two primary modes of the Claude CLI?
  • Options: A) Online and offline mode | B) Interactive REPL (claude) and Print mode (claude -p) | C) GUI and terminal mode | D) Single and batch mode
  • Correct: B
  • Explanation: Interactive REPL is the default conversational mode. Print mode (-p) is non-interactive, scriptable, pipeable — it exits after one response.
  • Review: CLI architecture section

Q2 ​

  • Category: practical
  • Question: How do you pipe a file into Claude and get JSON output?
  • Options: A) claude --file error.log --json | B) cat error.log | claude -p --output-format json "explain this" | C) claude < error.log --format json | D) claude -p --input error.log --json
  • Correct: B
  • Explanation: Pipe content via stdin to print mode (-p) and use --output-format json for structured output.
  • Review: Interactive vs Print Mode section

Q3 ​

  • Category: conceptual
  • Question: What is the difference between -c and -r flags?
  • Options: A) Both do the same thing | B) -c continues the most recent session; -r resumes by name or ID | C) -c creates a new session; -r resumes | D) -c is for code; -r is for review
  • Correct: B
  • Explanation: -c/--continue resumes the most recent conversation. -r/--resume "name" resumes a specific session by name or session ID.
  • Review: Session management section

Q4 ​

  • Category: practical
  • Question: How do you guarantee schema-valid JSON output from Claude?
  • Options: A) Just use --output-format json | B) Use --output-format json --json-schema '{"type":"object",...}' | C) Use --strict-json flag | D) JSON output is always schema-valid
  • Correct: B
  • Explanation: --output-format json alone produces best-effort JSON. Adding --json-schema with a JSON Schema definition guarantees the output matches the schema.
  • Review: Output & Format

Q5 ​

  • Category: conceptual
  • Question: Which flag only works in print mode (-p) and has no effect in interactive mode?
  • Options: A) --model | B) --system-prompt-file | C) --verbose | D) --max-turns
  • Correct: B
  • Explanation: --system-prompt-file loads a system prompt from a file but only works in print mode. Use --system-prompt (inline string) for interactive sessions.
  • Review: System prompt flags comparison table

Q6 ​

  • Category: practical
  • Question: How do you restrict Claude to only use read-only tools for a security audit?
  • Options: A) claude --read-only "audit code" | B) claude --permission-mode plan --tools "Read,Grep,Glob" "audit code" | C) claude --safe-mode "audit code" | D) claude --no-write "audit code"
  • Correct: B
  • Explanation: Combine --permission-mode plan (read-only analysis) with --tools (allowlist of specific tools) to restrict Claude to only read operations.
  • Review: Tool & Permission Management

Q7 ​

  • Category: conceptual
  • Question: What is the agent definition priority order?
  • Options: A) Project > User > CLI | B) CLI > Project > User | C) User > CLI > Project | D) All are equal priority
  • Correct: B
  • Explanation: Full order is managed settings > --agents flag > project (.claude/agents/) > user (~/.claude/agents/) > plugin. Of the options listed, CLI > Project > User is the correct relative ordering.
  • Review: Agents configuration section

Q8 ​

  • Category: practical
  • Question: How do you fork an existing session to try a different approach without losing the original?
  • Options: A) Use /fork command | B) Use --resume session-name --fork-session "branch name" | C) Use --clone session-name | D) Use /branch session-name
  • Correct: B
  • Explanation: --resume with --fork-session creates a new independent branch from the resumed session, preserving the original conversation.
  • Review: Session management section

Q9 ​

  • Category: conceptual
  • Question: What exit code does claude auth status return when the user is logged in?
  • Options: A) 1 | B) 0 | C) 200 | D) It doesn't return an exit code
  • Correct: B
  • Explanation: claude auth status exits with code 0 when logged in, 1 when not. This makes it scriptable for CI/CD authentication checks.
  • Review: CLI commands table

Q10 ​

  • Category: practical
  • Question: How do you process multiple files in a batch with Claude?
  • Options: A) claude --batch *.md | B) Use a for loop: for file in *.md; do claude -p "summarize: $(cat $file)" > ${file%.md\}.json; done | C) claude -p --files *.md "summarize all" | D) Batch processing is not supported
  • Correct: B
  • Explanation: Use shell for-loops with print mode to process files one at a time. Each invocation is independent and can produce structured output.
  • Review: Batch processing section

Last Updated: September 2, 2026 Claude Code Version: 2.1.257 Sources: