Step‑by‑Step Guide to Install Claude Code and Build Custom AI Skills

1. Installing Claude Code

Claude Code is an AI‑powered coding assistant that runs in a terminal, IDE, or GitHub. Install it with one of the following commands:

macOS/Linux (recommended) : curl -fsSL https://claude.ai/install.sh | bash Homebrew (macOS/Linux) : brew install --cask claude-code Windows (recommended) : irm https://claude.ai/install.ps1 | iex WinGet (Windows) : winget install Anthropic.ClaudeCode NPM (deprecated) : npm install -g @anthropic-ai/claude-code After installation, run claude in your project directory. On Windows you may need to add the executable directory (e.g., C:\Users\HP\.local\bin) to the PATH environment variable.

2. Configuring the Large‑Model Backend

To use Claude’s models, install Antigravity Tools version 4.0.10 (download from

https://github.com/lbjlaq/Antigravity-Manager/releases/tag/v4.0.10

) and place the following JSON in ~/.claude/config.json (or C:\Users\HP\.claude\config.json on Windows):

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "sk-f2e1c87e0c50465c95feeabddbe7538b",
    "ANTHROPIC_BASE_URL": "http://127.0.0.1:8045",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-5-20251001",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-4-5-20251101",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-5",
    "ANTHROPIC_MODEL": "claude-opus-4-5-20251101",
    "ANTHROPIC_REASONING_MODEL": "claude-sonnet-4-5"
  },
  "includeCoAuthoredBy": false,
  "enabledPlugins": { "webapp-testing@awesome-claude-skills": true },
  "model": "opus"
}

Setting includeCoAuthoredBy = false skips the login step.

3. Skills – Modular Extensions for Claude

A Skill is a folder that extends Claude’s capabilities. After a fresh install the /skills command shows no skills. Community and official Skills can be downloaded from the Anthropic Marketplace or GitHub (e.g., https://github.com/anthropics/skills/tree/main/skills) and placed under ~/.claude/skills (or C:\Users\HP\.claude\skills on Windows).

3.1 Directory layout

skill-name/
├── SKILL.md          # required, contains metadata and workflow
├── scripts/          # optional executable scripts (Python, Bash, …)
├── references/       # optional docs loaded on demand
└── assets/           # optional output assets (templates, images, fonts)

3.2 SKILL.md – the Skill’s soul

SKILL.md consists of two parts:

YAML front‑matter (metadata). Example:

---
name: deep-research
description: Generate structured research reports, auto‑outline, multi‑source integration.
---

The description field is the sole trigger for Claude; a clear description ensures correct activation.

3.3 Markdown body – command logic

The body defines the workflow, for example:

# Deep Research Skill
## 工作流程
1. 确认研究主题和范围
2. 生成研究大纲
3. 分章节收集资料
4. 整合撰写报告
5. 生成参考文献
## 输出格式
使用 scripts/generate_report.py 生成最终 PDF…

3.4 Progressive disclosure design

Skills load in three layers to conserve Claude’s context window:

┌─────────────────────────────────
│  第 1 层：元信息（始终在上下文中）│ name + description (~100 words)
├────────────────────────────────────
│  第 2 层：SKILL.md 正文（触发后加载）│ ≤ 500 行
├─────────────────────────────────────
│  第 3 层：bundled resources（按需加载）│ scripts/、references/、assets/（无上限）
└────────────────────────────────────────┘

This prevents token waste and keeps the most relevant information visible.

4. Building a Real‑World Pre‑Sales Proposal Skill

4.1 Business planning

The Skill generates pre‑sales proposals and therefore needs a knowledge base containing market information, technical specifications, case studies, system architecture diagrams, etc.

Market pre‑sales

Selection guidance

Site survey guidance

Business quotation

Tendering

Installation & debugging

Customer case studies

System functions

R&D team information

Construction plan

4.2 Skill architecture

Directory layout (excerpt):

assets/
├── logo.png
├── template.docx
├── fonts/
└── 技术架构.png
references/
├── INDEX.md   # index of all docs with description, scenario, and path
└── xxx档案库   # actual documents
SKILL.md          # workflow definition

INDEX.md lists each file with a brief description, applicable scenario, and path, enabling the Skill to load only the needed resources.

4.3 SKILL.md definition

---
name: dcim-proposal-architect
description: 当用户要求生成数据中心售前方案、解决方案或标书时触发此技能。
metadata:
  author: wengl
  version: "1.0"
---
# 数据中心售前方案生成专家（dcim-proposal-architect）
## 角色定义 (Role)
你是一名拥有10年经验的资深数据中心售前架构师，能够结合公司文档输出逻辑严密、说服力强的中文方案。
## 策略 (Strategy)
遵循 **"Index‑First"** 策略，仅在需要时读取文件全文。
## 功能
- 读取技术要求文件（./references/技术要求）
- 根据模板（./assets/动环监控系统技术方案格式模板.docx）生成结构化文档
- 支持 Markdown/Word/PDF 输出
## 触发条件 (Triggers)
- 用户说 “帮我写个方案” 或 “生成售前文档”
- 用户上传需求文档并要求中文回复
## 工作流 (Process)
1. **分析需求 & 查阅索引**：读取 ./references/INDEX.md，决定相关文件。
2. **精准读取**：仅读取选中的 docx/pdf/xlsx 文件的全文。
3. **内容构建**：使用模板生成 Word，嵌入 ./assets 中的图片、图表。
4. **特殊处理**：英文资料自动翻译，专有名词保留英文。
## 约束 (Constraints)
- 禁止一次性读取全部 ./references 文件。
- 必须基于具体文件引用提供证据。
## 输出格式 (Output)
- 生成图文并茂的 Word 文件（如 售前方案_客户名称_日期.docx），返回文件链接。

4.4 Running the Skill

Execute the Skill from the Claude CLI:

/dcim-proposal-architect 根据 C:\Users\HP\.claude\skills\dcim-proposal-architect\references\技术要求 中的技术要求生成对应的售前方案

The Skill automatically loads auxiliary Skills for PDF, DOCX, and XLSX handling, reads the indexed files, assembles the proposal, and outputs a Word document. Sample output excerpt:

● P好的，我来按照工作流程为您生成售前方案。首先读取索引文件和技术要求。
● Reading 1 file…
● Skill(pdf) … successfully loaded
● 第一步：分析需求与查阅索引
● 根据用户需求，我需要读取以下相关文件：
   1. 动环监控系统功能模块描述.docx
   2. 施工方案.docx
   3. 研发团队人员介绍.docx
● 第二步：精准读取相关文件 …
● 第三步：生成 Word 文档并嵌入图片
● B✅ 售前方案已生成完成！ 文档路径： C:\Users\wengl\.claude\skills\dcim-proposal-architect\数据中心平台技术方案_2025.docx

5. Key Takeaways

Install Claude Code via curl, Homebrew, PowerShell, WinGet, or npm.

Configure the large‑model backend with Antigravity Tools and a JSON token file.

Skills are modular folders; the essential file is SKILL.md, which contains a YAML front‑matter (trigger) and a markdown body (workflow).

Progressive disclosure (metadata → SKILL.md body → bundled resources) conserves Claude’s context window.

When building a Skill, define clear triggers, a concise description, and an index‑first strategy to load only needed documents.

Running a Skill via the CLI demonstrates automatic loading of auxiliary Skills (pdf, docx, xlsx) and produces the final artifact.