新项目

文档 社区

技能解读

当前原文的自动摘译,尚未人工审校。

1. 面试(如需要)

如果没有给出语言/框架,询问用户:什么语言/框架?什么类型的项目? (API、CLI、库、Web 应用程序等)有任何特定的工具首选项吗? (测试框架、linter等)

2.初始化项目

在当前工作目录中创建项目目录并初始化git:项目在$PWD/<project-name>/创建。如果您想在其他地方使用,请先将cd移动到所需的父目录。

3. 创建工作流程结构

创建这些目录: 注意:技能全局安装在~/.claude/skills/。不要创建项目级技能副本——它们会重复并偏离全球版本。全局技能(/architecture、/tdd、/security、/adr、/rfc、/spec、/roadmap、/feature、/fix、/review、/autopilot)已经可用。对于堆栈感知代码审查,请安装来自 claude-code-plugins 的 Anthropic 官方 code-review 技能。仅创建项目级代理和设置。

4.创建CLAUDE.md

在项目根目录创建一个适合语言/框架的CLAUDE.md:使构建命令和样式部分适应所选的语言/框架。

5. 创建项目级代理

将安全审核器和架构审核器复制到.claude/agents/,根据需要根据项目的语言进行定制。

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
文档
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@tomevault-io · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
即装即用
是否需要外部接口密钥
不需要
兼容的系统
未声明(默认跨平台)
运行环境要求
Python
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
  • 读取环境变量
网络行为
仅限本地
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md
Scaffold a new project with the AI-assisted development workflow.

**Argument format:** `<project-name> [language/framework]`

- `<project-name>` (required) — name of the project directory to create
- `[language/framework]` (optional) — the tech stack, used to tailor Makefile targets, linter config, .gitignore, and pre-commit hooks. Examples: `python`, `python/fastapi`, `go`, `typescript/nextjs`, `rust`. If omitted, you will be asked.

**Examples:**
- `/new-project my-api python/fastapi` — creates `./my-api/` with Python/FastAPI tooling
- `/new-project my-cli go` — creates `./my-cli/` with Go tooling
- `/new-project my-app` — creates `./my-app/`, asks what stack to use

## Process

### 1. Interview (if needed)

If no language/framework was given, ask the user:
- What language/framework?
- What type of project? (API, CLI, library, web app, etc.)
- Any specific tooling preferences? (test framework, linter, etc.)

### 2. Initialize the project

Create the project directory **inside the current working directory** and initialize git:

```bash
mkdir -p <project-name>
cd <project-name>
git init
```

The project is created at `$PWD/<project-name>/`. If you want it elsewhere, `cd` to the desired parent directory first.

### 3. Create workflow structure

Create these directories:

```
.claude/
.claude/agents/
docs/
docs/specs/
docs/roadmap/
docs/adr/
docs/rfc/
```

**Note:** Skills are installed globally at `~/.claude/skills/`. Do NOT create project-level skill copies — they'd duplicate and drift from the global versions. The global skills (`/architecture`, `/tdd`, `/security`, `/adr`, `/rfc`, `/spec`, `/roadmap`, `/feature`, `/fix`, `/review`, `/autopilot`) are already available. For stack-aware code review, install Anthropic's official `code-review` skill from `claude-code-plugins`. Only create project-level agents and settings.

### 4. Create CLAUDE.md

Create a `CLAUDE.md` at the project root tailored to the language/framework:

```markdown
# CLAUDE.md

## Build & Test
- `make test` — run test suite
- `make lint` — run linters
- `make typecheck` — type checking (if applicable)
- `make build` — build the project (if applicable)

## Code Style
- [Language-specific conventions that Claude might get wrong]

## Architecture
- [Brief description of project structure]

## Workflow
- All PRs require passing CI + human review
- Commit messages use conventional commits (feat:, fix:, refactor:, chore:)
- Security-sensitive changes require /sec-review before PR
- Features start with a spec in docs/specs/
```

Adapt the build commands and style section to the chosen language/framework.

### 5. Create project-level agents

Copy the security-reviewer and architecture-reviewer into `.claude/agents/`, tailored to the project's language if needed.

### 6. Create project-level settings

Create `.claude/settings.json` with hooks appropriate to the language:

```json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "command": "make lint-changed 2>&1 | head -20",
        "description": "Lint after every file edit"
      }
    ]
  }
}
```

### 7. Create Makefile

Create a `Makefile` with standard targets for the chosen language/framework:
- `test` — run tests
- `lint` — run linter
- `typecheck` — run type checker (if applicable)
- `build` — build (if applicable)
- `lint-changed` — lint only changed files
- `security-scan` — run security scanner (gitleaks, bandit, gosec, etc.)

### 8. Create .pre-commit-config.yaml

Set up pre-commit hooks:
- Language-appropriate linter
- Type checker (if applicable)
- Test runner
- gitleaks for secret detection

### 9. Create .gitignore

Appropriate for the language/framework. Always include:
- `.claude/` local files
- `.env`
- Language-specific build artifacts

### 10. Create initial docs

Create `.gitkeep` files to preserve directory structure:
- `docs/specs/.gitkeep`
- `docs/roadmap/.gitkeep`
- `docs/adr/.gitkeep`
- `docs/rfc/.gitkeep`

### 11. Summary

Tell the user what was created and suggest next steps:
1. `cd <project-name>`
2. `pre-commit install`
3. `/prd <project-name>` — define what we're building
4. `/architecture` — define system structure
5. `/tdd` — define testing strategy, dev environment, CI/CD
6. `/security` — define the threat model (if applicable)
7. `/spec <first-feature>` — write your first feature spec
8. `/feature docs/specs/<first-feature>.md` — implement it

---
> Source: [0xrafasec/ai-workflow](https://github.com/0xrafasec/ai-workflow) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。