安装命令

/plugin install claude-code-setup@claude-plugins-official

介绍

claude-code-setup 是 Anthropic 官方维护的一个 Claude Code 插件,作用不是直接帮你写代码,而是分析当前代码库,然后推荐应该给 Claude Code 安装哪些自动化能力。可以把它理解为:

“Claude Code 的架构顾问 / 配置顾问”。 (DeepWiki)

安装方式

/plugin install claude-code-setup@claude-plugins-official

官方插件市场仓库中的安装方式也是如此。 (GitHub)


它具体做什么?

安装后,它提供了一个 Skill:

  • claude-automation-recommender

这个 Skill 会扫描你的项目,然后推荐:

  • MCP Servers
  • Skills
  • Hooks
  • Subagents
  • 其他 Claude Code Plugins

并根据项目技术栈给出配置建议。 (DeepWiki)

例如它会分析:

  • package.json
  • requirements.txt
  • pyproject.toml
  • Dockerfile
  • GitHub Actions
  • 项目目录结构

然后判断:

这是一个 Python AI Agent 项目

或者:

这是一个 React + TypeScript 前端项目

再推荐适合的 Claude Code 生态组件。 (DeepWiki)


举个例子

假设你的 JobCopilot 仓库包含:

backend/
frontend/
docker-compose.yml
.github/workflows/

并且:

FastAPI
PostgreSQL
pgvector
LangGraph

那么它可能建议:

MCP

安装:

  • Context7(文档检索)
  • GitHub MCP
  • PostgreSQL MCP

Skills

安装:

  • code-review
  • architecture-review
  • database-design

Hooks

增加:

提交前自动运行测试
提交前自动执行 lint

Subagents

创建:

backend-engineer
frontend-engineer
devops-engineer

这样 Claude 可以自动委派任务。 (DeepWiki)


它怎么工作的?

这个插件本身是只读的。

官方说明显示它只使用:

  • Read
  • Glob
  • Grep
  • Bash(用于检测)

来分析项目结构。 (DeepWiki)

也就是说:

读取代码
分析项目
生成建议

不会自动修改你的项目。


适合谁?

非常适合

  • 刚接触 Claude Code
  • 不知道 MCP 怎么配
  • 不知道 Skill 怎么写
  • 不知道 Hooks 有什么玩法
  • 想把 Claude Code 调教成 Agent 系统

不太适合

已经深度定制 Claude Code 的用户。

因为:

你已经知道自己需要什么

它给出的很多建议你可能已经配置过了。


对你有没有价值?

结合你之前的情况:

  • 使用 Claude Code
  • 做 JobCopilot
  • 用 LangGraph
  • 用 Docker
  • 开始接触 CI/CD
  • 对 Claude Code 的高级能力还在探索

我觉得这是一个值得安装的官方插件。

它最大的价值不是功能本身,而是:

帮你发现 Claude Code 生态里有哪些你还不知道的能力。

很多开发者甚至不知道 Claude Code 有:

  • Hooks
  • Subagents
  • Skills
  • MCP Marketplace

这些扩展机制,而 claude-code-setup 就像一个自动顾问,会根据项目告诉你哪些值得用。 (DeepWiki)

简单总结:

项目 说明
类型 官方插件
核心功能 扫描代码库并推荐 Claude Code 自动化配置
会修改代码吗 不会
适合新手吗 非常适合
适合你的 JobCopilot 项目吗
我建议安装吗 建议安装一次看看推荐结果

安装后你可以直接让 Claude:

Analyze this repository and recommend Claude Code automations.

它就会调用这个 Skill,对整个仓库做一次自动化能力审查。 (DeepWiki)

使用方法

有两种方式。

方式1:Slash Command(显式调用,最可靠)

根据插件定义,这个 Skill 的 Slash Command 是:

/claude-code-setup:claude-automation-recommender

例如:

/claude-code-setup:claude-automation-recommender

或者:

/claude-code-setup:claude-automation-recommender
Analyze this repository

这是官方定义的显式调用方式。(Claude Plugin Hub)


方式2:自然语言触发

这个 Skill 同时支持 Claude 自动调用,所以你直接说:

recommend automations for this project

或者:

help me set up Claude Code

或者:

what hooks should I use?

Claude 通常会自动加载这个 Skill。(Claude)


如何确认是否调用成功?

在 Claude Code 中一般会看到类似:

Using skill:
claude-code-setup:claude-automation-recommender

或者任务日志中出现:

Loaded skill claude-automation-recommender

查看所有已安装 Skills

你可以执行:

/skills

然后搜索:

claude-automation-recommender

如果安装成功,应该能看到:

claude-code-setup
└── claude-automation-recommender

Claude Code 官方文档说明,Skill 默认既支持自动调用,也支持用户通过 /skill-name 直接调用。(Claude Code)


对于你这个 JobCopilot 项目,我建议直接这样用:

/claude-code-setup:claude-automation-recommender

Analyze this repository and recommend:
1. MCP servers
2. Hooks
3. Subagents
4. Custom skills

Focus on a LangGraph + FastAPI + PostgreSQL + Docker project.

这样得到的结果通常比简单一句“recommend automations”更有针对性。

示例

❯ /claude-code-setup:claude-automation-recommender Analyse this repository

● Skill(claude-code-setup:claude-automation-recommender)
  ⎿  Successfully loaded skill

● 让我分析一下仓库结构。

  Searched for 2 patterns, read 1 file, listed 3 directories (ctrl+o to expand)

● ---
  Claude Code Automation Recommendations

  Codebase Profile

  - Type: Python 3.11 (5 FastAPI microservices) + TypeScript (Next.js 15)
  - Key Libraries: LangGraph, Playwright, SQLAlchemy+asyncpg, Temporal, Keycloak, Vercel AI SDK, TanStack Query
  - CI/CD: GitHub Actions (ci.yml + cd.yml)
  - Security surface: AES-256 cookie encryption, multi-tenant JWT, per-service Alembic migrations

  ---
  🔌 MCP Servers

  1. Playwright MCP

  Why: Playwright is already a dependency for LinkedIn scraping. A Playwright MCP server lets you drive the
  browser directly from Claude — useful for the manual E2E verification you're doing right now, and for future UI
  debugging without writing test scripts by hand.
  Install:
  claude mcp add playwright npx @playwright/mcp@latest

  2. GitHub MCP

  Why: You have two GitHub Actions workflows (ci.yml + cd.yml). GitHub MCP lets Claude read CI run results, open
  issues, and manage PRs without leaving the conversation — especially useful once you hit the K8s milestone and
  PRs get more complex.
  Install:
  claude mcp add github -- npx -y @modelcontextprotocol/server-github
  # Requires: export GITHUB_PERSONAL_ACCESS_TOKEN=...

  ---
  ⚡  Hooks

  1. PostToolUse: auto-run Ruff after every Python edit

  Why: Ruff is already required in CI and in CLAUDE.md's pre-push checklist. Running it automatically after every
  file edit surfaces lint errors immediately instead of at push time.
  Where: .claude/settings.json
  {
    "hooks": {
      "PostToolUse": [{
        "matcher": "Edit|Write",
        "hooks": [{
          "type": "command",
          "command": "cd /home/user/projects/JobCopilot && ~/.local/bin/uv run ruff check --fix
  \"$CLAUDE_TOOL_INPUT_FILE_PATH\" 2>/dev/null || true"
        }]
      }]
    }
  }

  2. PreToolUse: block edits to uv.lock

  Why: uv.lock must only be updated via uv sync, never hand-edited. Blocking accidental edits prevents silent
  dependency corruption.
  Where: .claude/settings.json
  {
    "hooks": {
      "PreToolUse": [{
        "matcher": "Edit|Write",
        "hooks": [{
          "type": "command",
          "command": "if echo \"$CLAUDE_TOOL_INPUT_FILE_PATH\" | grep -q 'uv\\.lock'; then echo 'uv.lock is
  managed by uv sync — do not edit directly' >&2; exit 1; fi"
        }]
      }]
    }
  }

  ---
  🎯 Skills

  1. create-migration — Alembic migration generator

  Why: You have 5 services each with their own Alembic setup. A skill that scaffolds a correctly-named,
  tenant-aware migration (with the WHERE tenant_id reminder baked in) saves repeated setup work.
  Create: .claude/skills/create-migration/SKILL.md
  Invocation: User-only (disable-model-invocation: true)

  2. service-test — pytest test generator following project conventions

  Why: 138 Python files across 5 services; test coverage is likely uneven. A skill that generates pytest stubs
  using the existing asyncio_mode=auto config and marks integration tests correctly reduces the friction of
  writing tests.
  Create: .claude/skills/service-test/SKILL.md
  Invocation: Both

  ---
  🤖 Subagents

  1. security-reviewer

  Why: This codebase has a large security surface — AES-256-GCM cookie encryption, JWT RS256 validation,
  mandatory WHERE tenant_id isolation on every DB query, and a strict "no credentials in logs" policy. A
  dedicated security subagent that knows these invariants can run in parallel during code review and catch
  cross-tenant leaks or encryption misuse that a general reviewer would miss.
  Where: .claude/agents/security-reviewer.md

  ---
  Want more? Ask for additional recommendations for any specific category (e.g., "show me more hook ideas" or
  "give me the full skill template for create-migration").

  Want help implementing any of these? Say the word and I'll set it up.

Logo

汇聚全球AI编程工具,助力开发者即刻编程。

更多推荐