claude-code-starter

Describe your project idea — Claude sets up the entire development environment for you.

What it does

This template gives Claude Code a bootstrap protocol. You describe your project — as a spec, a few sentences, or a markdown document — and Claude:

1. Decides the optimal tech stack for your project
2. Installs dependencies and initializes the project
3. Creates a CLAUDE.md with domain rules extracted from your description
4. Creates an AGENTS.md for cross-tool AI context
5. Generates project-specific slash commands (skills) for repeated workflows
6. Configures hooks so formatting and typechecking run automatically
7. Makes a first git commit

After bootstrap, the template infrastructure is deleted. What remains is a cleanly configured project tailored to your specific idea.

What you get after bootstrap

• CLAUDE.md — concise session context (under 150 lines) with your stack, commands, and domain rules
• AGENTS.md — cross-platform context for Claude Code, Cursor, Copilot, and others
• .claude/settings.json — hooks for auto-format on edit and typecheck before Claude ends a turn
• .claude/skills/ — slash commands tailored to your project's specific workflows
• agent_docs/ — detailed reference material loaded on demand, not every session
• .claude/agents/code-reviewer.md — isolated code review subagent

Quick start

Option A — GitHub Template (recommended)

1. Click "Use this template" → Create new repository
2. Clone your new repo
3. Open Claude Code in the project root
4. Say: Read INIT.md and set up my project
5. Paste your project description when asked — spec document, bullet points, or a few sentences
6. Claude proposes the setup plan, waits for your approval, then executes

Option B — Clone manually

git clone https://github.com/2zovd/claude-code-starter.git my-project
cd my-project
rm -rf .git && git init
# Open Claude Code and say: Read INIT.md and set up my project

Repository layout

claude-code-starter/
│
├── INIT.md                    # Bootstrap protocol — Claude reads this on first run
│
├── guides/                    # How-to guides Claude reads during bootstrap
│   ├── how-to-claude-md.md    # What makes an effective CLAUDE.md
│   ├── how-to-agents-md.md    # How to write cross-platform AGENTS.md
│   ├── how-to-skills.md       # How to create Claude Code slash commands
│   ├── how-to-hooks.md        # How to configure deterministic hooks
│   └── how-to-agent-docs.md   # When and how to use agent_docs/
│
├── templates/                 # Skeleton files filled in during bootstrap
│   ├── CLAUDE.md.template
│   ├── AGENTS.md.template
│   └── settings.json
│
├── examples/                  # Reference implementation
│   └── node-cli-tool/         # Example of a well-configured TypeScript CLI project
│       ├── CLAUDE.md
│       ├── AGENTS.md
│       └── .claude/skills/new-transformer/SKILL.md
│
└── .claude/
    └── agents/
        └── code-reviewer.md   # Universal code review subagent

All of guides/, templates/, and examples/ are deleted after bootstrap. The result is a clean project with no template overhead.

What kind of project description works

Any format — Claude extracts what it needs:

"A CLI tool that converts Markdown to HTML. TypeScript, Node.js. The output should support custom CSS themes via a flag."

"A REST API for a task manager. Python, FastAPI, PostgreSQL. Users can create tasks, assign them, set deadlines. Auth via JWT."

"Resume as Code — a developer-first resume builder. JSON data + pluggable templates + Git versioning. Generates HTML, DOCX, PDF from a single source of truth. Tailoring for specific companies via override files."

Or paste an entire product spec — the longer the description, the more context Claude has to configure your project accurately.

Why this structure

Claude decides the stack dynamically. Technology best practices change faster than template files. Rather than shipping pre-baked presets that age, Claude uses its current knowledge to choose dependencies, config, and tooling for your specific project.

Guides teach Claude how to configure, not what to use. The guides/ directory describes what makes a good CLAUDE.md, how to write effective skills, and when to use agent_docs/. Claude applies these principles to your project — the output is always project-specific.

Hooks enforce what instructions can only request. Auto-format on edit and typecheck before done are deterministic behaviors — they run regardless of what Claude decides in a given turn.

agent_docs/ keeps CLAUDE.md short. Details that would bloat the session context live in separate files loaded on demand. CLAUDE.md stays under 150 lines; Claude reads the detail files when a topic becomes relevant.

Contributing

See CONTRIBUTING.md to suggest improvements to guides, templates, or examples.

License

MIT — see LICENSE.