Claude Code Ultimate Guide
Learn Claude Code, build reliable agents, and scale their use safely. The website is the primary reading and discovery interface. This repository contains the canonical Markdown sources, reusable files, machine-readable indexes, and contribution history.
Start here: complete a first task online · browse the guide portal · open the complete sitemap · read the Markdown source
Choose your next step
The table below is generated from machine-readable/navigation.json. The same contract feeds the public sitemap, so the repository and website expose the same intent model.
Why this guide exists
Claude Code documentation explains the product. This guide connects product behavior to engineering decisions: what belongs in context, when to use an agent instead of a skill, how to verify generated work, and which controls matter when usage moves beyond one developer.
The guide favors explicit trade-offs and verifiable procedures. Where evidence is incomplete, the relevant page should preserve that limit instead of presenting one workflow as universal.
Start
Install Claude Code
Choose one installation method:
# npm, macOS, Linux, or Windows
npm install -g @anthropic-ai/claude-code
# macOS with Homebrew
brew install claude-code
# macOS or Linux native installer
curl -fsSL https://claude.ai/install.sh | sh
Windows PowerShell also supports:
irm https://claude.ai/install.ps1 | iex
Verify the installation and authenticate:
claude --version
claude doctor
claude auth login
The Quick Start chapter documents installation alternatives, authentication, updates, permission modes, and common first-day failures.
Complete a first task
Open a small repository with a clean or understood Git state, then start Claude Code:
cd your-project
claude
Give Claude a bounded request that includes the expected result and verification:
Explain how this project runs its tests. Do not modify files.
Name the relevant commands and cite the files that define them.
Before asking Claude to edit code, add a project-level CLAUDE.md that records the commands and constraints Claude must follow:
# Project instructions
## Commands
- Test: `npm test`
- Lint: `npm run lint`
## Boundaries
- Do not edit generated files.
- Do not change dependencies without approval.
- Run the relevant tests before claiming completion.
Continue with the first workflow or use the starter CLAUDE.md templates.
Choose a learning path
For a personalized route, use the repository onboarding prompt:
claude "Fetch and follow the onboarding instructions from: https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/tools/onboarding-prompt.md"
Build
Agent engineering
An agent is one component of a larger system. The surrounding harness controls context, tools, permissions, state, stopping conditions, recovery, and evaluation.
Context and memory
Context quality affects every tool call and decision. Start with project instructions, add specialized context only when a recurring task needs it, and test whether the extra material changes behavior.
Workflows
Browse every workflow for task management, GitHub Actions, production reliability, event-driven agents, design-to-code, PDF generation, search, and team instructions.
Choose the smallest interface
Claude Code can call local commands, skills, agents, hooks, plugins, and MCP servers. More infrastructure adds setup, permissions, failure modes, and maintenance.
| If the task needs |
Prefer |
| A deterministic local command |
CLI or script |
| Reusable instructions and supporting files |
Skill |
| A separate context and role |
Agent |
| A response to a lifecycle event |
Hook |
| A packaged collection of capabilities |
Plugin |
| A typed interface to a remote service |
MCP server |
Use the MCP or CLI decision guide and the trade-off framework before introducing another integration.
Scale
Reliability and security
Security claims and threat counts change as sources are added or corrected. Use the linked database and guides as the current source instead of copying their counts into project documentation.
Evaluation and operations
A green structural check proves only what it inspected. Runtime behavior, task acceptance, security boundaries, and business outcomes require separate evidence.
Organization and economics
The economics pages separate observed costs from estimates and scenarios. Recalculate them with your workload, acceptance criteria, review effort, and risk constraints.
MCP Server: Use the guide from any coding client
Install the guide as a stdio MCP server and query it from Claude Code, Codex, Cursor, VS Code, or another MCP client.
claude mcp add --scope user claude-code-guide -- npx -y [email protected]
codex mcp add claude-code-guide -- npx -y [email protected]
Project-scoped Claude Code configuration belongs in .mcp.json:
{
"mcpServers": {
"claude-code-guide": {
"type": "stdio",
"command": "npx",
"args": ["-y", "[email protected]"]
}
}
}
| Capability |
Count |
Names |
| Tools |
17 |
compare_versions, diff_official_docs, get_changelog, get_cheatsheet, get_digest, get_example, get_release, get_threat, init_official_docs, list_examples, list_threats, list_topics, read_section, refresh_official_docs, search_examples, search_guide, search_official_docs |
| Resources |
6 |
claude-code-guide://agent-harnesses, claude-code-guide://distribution-channels, claude-code-guide://llms, claude-code-guide://reference, claude-code-guide://releases, claude-code-guide://translations |
| Prompts |
1 |
claude-code-expert |
| Companion commands |
5 |
/ccguide:daily, /ccguide:diff-docs, /ccguide:init-docs, /ccguide:refresh-docs, /ccguide:search-docs |
The list operations and search index use bundled content. Section, example, cheatsheet, changelog, digest (get_digest), and threat tools can fetch GitHub content and write a 24-hour local cache. The official-doc initialization and refresh tools fetch Anthropic documentation and write separate local snapshots.
Canonical technical guide, installation, privacy, limitations, and dated statistics
Package README and diagnostics
Companion commands rendered from the repository: /ccguide:daily, /ccguide:diff-docs, /ccguide:init-docs, /ccguide:refresh-docs, /ccguide:search-docs.
Load the guide into another assistant
Use the small index for discovery:
curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/machine-readable/llms.txt
Use reference.yaml when the assistant needs structured topic routes and links into the full guide. Maintenanc
这是一份深入的 Claude Code 指南，超过 43 万行，涵盖智能体工作流、Hook、Skill、MCP 服务器、安全和生产实践，并配有可运行的模板和测验供你学习。