init-project

A Claude Skill that initializes a Vibcoding (AI-driven development) project environment. It does not write any source code or install any dependencies — instead, it scaffolds the documentation and specification layer that an AI agent needs in subsequent development phases.

Overview

init-project runs a short interview with the user to collect basic project information (name, language, tech stack, project type, etc.), then automatically generates a standardized set of project documents and an AI memory system, including:

• CLAUDE.md — AI behavior rules (auto-loaded into context every session; the most critical file)
• README.md — Project entry document
• ROADMAP.md — Development roadmap
• .gitignore — Universal ignore rules
• memory/ — AI memory system (MEMORY.md, user-preferences.md, decisions.md)

When It Triggers

Claude invokes this skill automatically when the user expresses any of the following intents:

• "init project" / "initialize project"
• "scaffold" / "set up project"
• "Phase 1"
• Starting a brand-new project from scratch

Directory Structure

init-project/
├── SKILL.md                    # Main instruction file (read by AI)
├── README.md                   # This file (English documentation, default)
├── README_CN.md                # Chinese documentation
├── scripts/
│   └── init-memory.sh          # Creates the AI memory directory and files
├── references/
│   ├── claude-template.md      # CLAUDE.md template (with placeholders)
│   ├── readme-template.md      # README.md template (with placeholders)
│   ├── roadmap-template.md     # ROADMAP.md template (with placeholders)
│   └── vibecoding-template.md  # VIBECODING.md template
└── assets/
    └── gitignore_template      # Universal .gitignore template

Workflow

1. Interview the user — Use AskUserQuestion to gather the project name, one-line description, language, tech stack, and project type. Optional fields include Git strategy, development mode, and emoji preference.
2. Generate core documents — Read templates from references/, replace {{PLACEHOLDER}} values with user input, then write to the project root.
3. Create the memory system — Run bash scripts/init-memory.sh to initialize the memory/ directory.
4. Optional Git initialization — If version control is enabled, run git init and make the first commit; if a remote URL is provided, also link and push.

Placeholder Mapping

Placeholder	Source
{{PROJECT_NAME}}	User-provided project name
{{PROJECT_DESCRIPTION}}	One-line description
{{LANGUAGE}}	Programming language
{{TECH_STACK}}	Full tech stack
{{PROJECT_TYPE}}	CLI tool / Web app / Library / Mobile app / etc.
{{CODE_STANDARDS}}	Code standards mapped by language
{{TEST_STANDARDS}}	Test standards mapped by language
{{PROJECT_ROOT}}	Project directory name

Language → Standards Mapping (Excerpt)

Language	Code Standards	Test Standards
Python	Google-style docstring + PEP 484 type hints	pytest, coverage > 80%
TypeScript / JavaScript	ESLint + Prettier, strict mode	Vitest / Jest, coverage > 80%
Java	Javadoc + Spring Boot conventions	JUnit 5, coverage > 80%
Go	gofmt + standard project layout	go test, coverage > 80%
Rust	rustfmt + clippy linting	cargo test
Other	Community standard conventions	Standard test framework for the language

For the complete mapping, see SKILL.md.

Required Tools

This skill depends on the following AI tool capabilities:

• Write — Create documents
• Edit — Adjust content
• Read — Load templates
• AskUserQuestion — User interview
• Bash — Execute scripts and Git commands

What This Skill Does NOT Do

To keep responsibilities focused, this skill explicitly avoids:

• Creating any source code files
• Installing dependencies or toolchains
• Configuring build systems
• Creating framework-specific files
• Configuring IDEs or editors

These belong to later development phases (Phase 2 and beyond). This skill only scaffolds the specification and documentation layer that an AI agent needs to work effectively.

Important Rules

1. Do not use emoji unless the user explicitly requests it
2. Template content lives in references/, never hardcoded in the main instruction file
3. User preferences are recorded in memory/user-preferences.md
4. Key architectural decisions (with dates) are recorded in memory/decisions.md
5. CLAUDE.md is the core file — it governs all subsequent AI behavior
6. Vibcoding configuration placement: All vibcoding-related configurations (Claude command whitelists, hooks, skills, subagent definitions, etc.) MUST be placed inside the .claude/ directory at the project root. Global configuration should ONLY be used when the user explicitly requests it.

---

For the Chinese version, see README_CN.md.