AgentSkillsCN

claude-code-guide

当用户就以下主题向您提问时，请使用此代理：“Claude…能否…”、“Claude…是否…”、“我该如何…”——（1）Claude Code（CLI 工具）：功能特性、钩子、斜杠命令、MCP 服务器、设置、IDE 集成、键盘快捷键；（2）Claude Agent SDK：构建自定义代理；（3）Claude API（原 Anthropic API）：API 使用、工具调用、Anthropic SDK 使用。**重要提示：** 在启动新代理之前，请先检查是否已有正在运行或刚刚完成的 claude-code-guide 代理，您可以使用“resume”参数继续沿用该代理。

SKILL.md
--- frontmatter
name: claude-code-guide
description: |
  Use this agent when the user asks questions ("Can Claude...", "Does Claude...", "How do I...") about: (1) Claude Code (the CLI tool) - features, hooks, slash commands, MCP servers, settings, IDE integrations, keyboard shortcuts; (2) Claude Agent SDK - building custom agents; (3) Claude API (formerly Anthropic API) - API usage, tool use, Anthropic SDK usage. **IMPORTANT:** Before spawning a new agent, check if there is already a running or recently completed claude-code-guide agent that you can resume using the "resume" parameter.
allowed-tools:
  - Glob
  - Grep
  - Read
  - WebFetch
  - WebSearch

Claude Code Guide

This skill provides comprehensive guidance for Claude Code - the agentic coding tool from Anthropic.

What is Claude Code?

Claude Code is an agentic coding assistant that operates directly in your terminal. Unlike chat-based AI tools, Claude Code:

  • •Sees your codebase: Reads files, understands project structure, follows imports
  • •Takes action: Edits files, runs commands, creates commits, manages git workflows
  • •Verifies results: Runs tests, checks build output, validates changes
  • •Learns context: Remembers project conventions via CLAUDE.md files

The Agentic Loop

Claude Code operates in a continuous gather → act → verify loop:

code
┌─────────────────────────────────────────────────────┐
│                   AGENTIC LOOP                       │
├─────────────────────────────────────────────────────┤
│  1. GATHER CONTEXT                                   │
│     • Read files, search code, explore structure     │
│     • Understand current state and requirements      │
│                                                      │
│  2. TAKE ACTION                                      │
│     • Edit files, run commands, create content       │
│     • May require permission for sensitive ops       │
│                                                      │
│  3. VERIFY RESULTS                                   │
│     • Check command output, run tests                │
│     • Validate changes meet requirements             │
│                                                      │
│  4. REPEAT or COMPLETE                               │
│     • Continue if more work needed                   │
│     • Report results when done                       │
└─────────────────────────────────────────────────────┘

Read more: references/core/agentic-loop.md

Key Features Quick Reference

FeatureDescriptionLearn More
ToolsFile ops, search, bash, web accessreferences/core/tools.md
Context Window~200K tokens, auto-compactionreferences/core/context-management.md
CLAUDE.mdProject-specific instructionsreferences/configuration/memory.md
SkillsReusable prompts & workflowsreferences/configuration/skills.md
HooksShell commands on eventsreferences/configuration/hooks.md
MCP ServersExternal tool integrationsreferences/configuration/mcp.md
SubagentsDelegated specialized tasksreferences/configuration/sub-agents.md
Plan ModeDesign before implementingreferences/workflows/common-workflows.md
Headless ModeNon-interactive automationreferences/workflows/advanced-workflows.md

Extension Points Comparison

Choose the right extension mechanism for your needs:

MechanismBest ForScopeComplexity
CLAUDE.mdProject conventions, coding standards, contextProjectLow
SkillsReusable prompts, workflows, templatesUser/ProjectLow-Medium
HooksAutomation, validation, logging, side effectsUser/ProjectMedium
MCP ServersExternal APIs, databases, custom toolsUser/ProjectMedium-High
SubagentsSpecialized tasks, parallel work, isolationUser/ProjectMedium

When to Use What

Use CLAUDE.md when:

  • •Defining coding standards ("use tabs, not spaces")
  • •Documenting project architecture
  • •Specifying test commands
  • •Setting up file organization rules

Use Skills when:

  • •Creating reusable workflows (/commit, /review)
  • •Building prompt templates with arguments
  • •Packaging documentation for reference
  • •Defining specialized behaviors

Use Hooks when:

  • •Running commands before/after actions
  • •Validating or blocking tool usage
  • •Logging activity
  • •Auto-formatting code on save

Use MCP Servers when:

  • •Connecting to external APIs (GitHub, Jira, databases)
  • •Adding custom tool capabilities
  • •Integrating with existing services
  • •Providing resources from external sources

Use Subagents when:

  • •Delegating specialized investigations
  • •Running parallel analysis tasks
  • •Isolating context for focused work
  • •Creating custom agent personas

Navigation Guide

Understanding How Claude Code Works

  • •How does the agentic loop work? → references/core/agentic-loop.md
  • •What tools does Claude have? → references/core/tools.md
  • •How does context management work? → references/core/context-management.md

Configuration & Setup

  • •How do I set up CLAUDE.md? → references/configuration/memory.md
  • •How do I create skills? → references/configuration/skills.md
  • •How do I configure hooks? → references/configuration/hooks.md
  • •How do I add MCP servers? → references/configuration/mcp.md
  • •How do I create subagents? → references/configuration/sub-agents.md

Workflows & Best Practices

  • •Common tasks (exploration, debugging, PRs) → references/workflows/common-workflows.md
  • •Advanced patterns (parallel, CI/CD, headless) → references/workflows/advanced-workflows.md
  • •Best practices & tips → references/best-practices.md

CLI Reference

  • •Commands, flags, and options → references/cli-reference.md

Quick Tips

Effective Prompting

code
# Be specific about verification
"Add input validation to the signup form. Run the tests to verify."

# Provide context
"The auth system uses JWT tokens stored in httpOnly cookies.
Add a logout endpoint that clears the cookie."

# Request exploration first
"Explore how error handling works in this codebase, then
suggest improvements to the API error responses."

Keyboard Shortcuts

KeyAction
EnterSend message (configurable)
EscCancel current generation
Ctrl+CInterrupt / Exit
TabAccept autocomplete
Up/DownNavigate history

Essential Commands

CommandPurpose
/helpShow all commands
/compactReduce context usage
/clearStart fresh conversation
/rewindUndo to checkpoint
/statusShow current state
/configOpen settings
/reviewReview code changes
/pr-commentsAddress PR feedback

Common Questions

Q: How much context does Claude Code have? A: ~200K tokens. Use /compact if running low. See references/core/context-management.md.

Q: Can Claude Code run commands without asking? A: Depends on permission mode. Safe commands auto-approve; destructive ones ask. See references/core/agentic-loop.md.

Q: How do I share context across sessions? A: Use CLAUDE.md files for persistent context. See references/configuration/memory.md.

Q: Can I use Claude Code in CI/CD? A: Yes, with headless mode (claude -p). See references/workflows/advanced-workflows.md.

Q: How do I add custom tools? A: Via MCP servers. See references/configuration/mcp.md.

IDE Integration

Claude Code integrates with VS Code and JetBrains IDEs:

VS Code: Install "Claude Code" extension

  • •Cmd+Esc (Mac) / Ctrl+Esc (Windows/Linux) to open
  • •Automatic file sync with editor
  • •Inline diff viewing

JetBrains: Install "Claude Code" plugin

  • •Same keyboard shortcut
  • •Project context awareness
  • •Integrated terminal

Getting Help