Skip to content
/

cloudnative-co/claude-code-starter-kit

cloudnative-cc-starter · cloudnative-co/claude-code-starter-kit · ★ 116 · last commit 2026-05-19

Japanese-language wizard-driven Claude Code setup kit with 3 profiles, Anthropic plugin marketplace integration, and cross-platform (macOS/Linux/Windows) installation.

Best whenClaude Code setup should be as beginner-accessible as a GUI wizard, with a plugin marketplace that auto-installs official Anthropic plugins by profile.
vs seeds
spec-kit(commands + skills + profile-based installation) but wizard-driven shell installer instead of Python CLI, and installs g…
Primitive shape 40 total
Commands 19 Skills 12 Subagents 9
00

Summary

cloudnative-co/claude-code-starter-kit — Summary

cloudnative-co/claude-code-starter-kit is a Japanese-language interactive setup wizard for Claude Code, featuring a plugin marketplace system, 3 installation profiles (minimal/standard/full), and 19+ commands + 9 agents + 12 skills. The installer (install.sh / install.ps1) runs an interactive wizard that asks about OS, editor, and profile preference, then builds ~/.claude/CLAUDE.md, assembles settings.json via jq fragment merging, and deploys agents/rules/commands/skills/memory to ~/.claude/. The key innovation is a config/plugins.json pointing at two marketplace sources (anthropics/claude-plugins-official and anthropics/skills) with 10+ plugin names mapped to install profiles — standard and full profiles auto-install official Anthropic plugins. The framework was created by the CEO of CloudNative Inc., who is also the Ministry of Education's Chief Information Security Advisor.

differs_from_seeds: Closest to spec-kit (commands + skills + hooks + profile system) but implemented as a wizard-driven shell installer rather than a Python CLI. The plugin marketplace integration (pointing at anthropics/claude-plugins-official) is unique — no seed routes installations through the official Anthropic plugin registry. The /orchestrate command's sequential agent workflow (planner → tdd-guide → code-reviewer → security-reviewer) resembles claude-flow's sequential chain pattern but without an MCP server.

01

Overview

cloudnative-co/claude-code-starter-kit — Overview

Origin

Published by CloudNative Inc. (株式会社クラウドネイティブ), Japan. README author credited as "代表取締役社長・文部科学省最高情報セキュリティアドバイザーの齊藤愼仁" (President/CEO + Ministry of Education Chief Information Security Advisor). 116 GitHub stars, MIT license, Shell language. Last pushed 2026-05-19. 3 contributors.

Philosophy

From the README (Japanese, translated):

"A setup kit that lets even first-time Claude Code users set up their development environment with a single command. Explained gently enough for beginners to understand, even those new to programming and terminals."

The framework explicitly targets Japanese beginners unfamiliar with terminal usage — the README includes explanations of what "terminal" and "code editor" are, with screenshots and YouTube links.

The CLAUDE.md architecture section captures the execution flow philosophy:

install.sh
  → setup.sh (orchestrator)
      → wizard/wizard.sh (CLI parsing + interactive prompts)
      → lib/detect.sh (OS/WSL/MSYS detection)
      → lib/prerequisites.sh (dependency checks)
      → lib/deploy.sh
          → build_claude_md() — template engine assembles ~/.claude/CLAUDE.md
          → build_settings_file() — jq merges base JSON + permissions + hook fragments
      → deploy files to ~/.claude/{agents,rules,commands,skills,memory}/
      → plugin marketplace registration + install (multi-marketplace)

Plugin Marketplace Integration

The framework connects to:

  • anthropics/claude-plugins-official — official Anthropic plugins
  • anthropics/skills — Anthropic skills registry

Profiles determine which plugins are auto-installed:

  • standard: security-guidance, commit-commands, pr-review-toolkit, feature-dev, code-review, claude-md-management
  • full: all standard + typescript-lsp, gopls-lsp, pyright-lsp, tailwind-css-lsp, eslint-lsp, jest-runner, playwright-runner, python-debugger

Windows Support

Unlike most starters, this kit ships install.ps1 for Windows PowerShell with WSL2 setup automation.

02

Architecture

cloudnative-co/claude-code-starter-kit — Architecture

Distribution

Standalone GitHub repository with cross-platform install scripts (bash + PowerShell).

Install Methods

# Mac/Linux (interactive)
curl -fsSL https://raw.githubusercontent.com/cloudnative-co/claude-code-starter-kit/main/install.sh | bash

# Windows PowerShell
irm https://raw.githubusercontent.com/cloudnative-co/claude-code-starter-kit/main/install.ps1 | iex

# Non-interactive
bash setup.sh --non-interactive --profile=standard --language=en --editor=vscode

# Dry-run
bash setup.sh --dry-run

Directory Tree (source repo)

cloudnative-co/claude-code-starter-kit/
├── agents/              (9 agents)
├── commands/            (19 commands)
├── config/
│   ├── permissions.json
│   ├── plugins.json     (marketplace integration)
│   └── settings-base.json
├── docs/
├── features/
├── i18n/                (internationalization: ja, en)
├── lib/
│   ├── deploy.sh
│   ├── detect.sh        (OS/WSL/MSYS detection)
│   ├── fonts.sh
│   ├── ghostty.sh       (Ghostty terminal install)
│   └── prerequisites.sh
├── memory/
├── profiles/
│   ├── full.conf
│   ├── minimal.conf
│   └── standard.conf
├── rules/
├── skills/              (12 skills)
├── tests/
├── wizard/
│   └── wizard.sh        (interactive CLI prompts)
├── AGENTS.md
├── CLAUDE.md
├── install.ps1          (Windows)
├── install.sh           (Mac/Linux)
├── setup.sh             (orchestrator)
└── uninstall.sh

Installation Profiles

Profile Plugins Agents/Skills
minimal 0 Core only
standard 6 (security, commit, PR review, feature-dev, code-review, claude-md-mgmt) Standard set
full 10+ (standard + LSPs, test runners, debugger) Full set

Target AI Tools

Claude Code (primary). The installed files land in ~/.claude/ (global) rather than project-local .claude/.

Required Runtime

  • bash (macOS/Linux/WSL)
  • PowerShell (Windows — for install.ps1)
  • jq (for settings.json assembly)
  • shellcheck (for CI validation)
03

Components

cloudnative-co/claude-code-starter-kit — Components

Commands (19)

  • build-fix.md — diagnoses and fixes build failures
  • checkpoint.md — saves progress checkpoint
  • code-review.md — reviews code changes
  • e2e.md — runs end-to-end tests
  • eval.md — evaluates code/architecture quality
  • handover.md — generates handover documentation
  • learn.md — teaches concepts related to current code
  • orchestrate.md — sequential agent workflow (planner → tdd-guide → code-reviewer → security-reviewer)
  • plan.md — creates implementation plan
  • refactor-clean.md — safe refactoring with clean code principles
  • research.md — researches technical topics
  • spec-kit-init.md — initializes spec-kit style specs
  • tdd.md — test-driven development workflow
  • test-coverage.md — analyzes test coverage gaps
  • update-codemaps.md — updates code maps/documentation
  • update-docs.md — updates project documentation
  • update-kit-dry-run.md — previews kit update changes
  • update-kit.md — updates the installed kit
  • verify.md — verifies implementation correctness

Agents (9)

  • architect.md — architecture design specialist
  • build-error-resolver.md — diagnoses and resolves build errors
  • code-reviewer.md — code quality reviewer
  • doc-updater.md — documentation maintenance agent
  • e2e-runner.md — E2E test execution agent
  • planner.md — implementation planning agent
  • refactor-cleaner.md — safe refactoring agent
  • security-reviewer.md — security vulnerability reviewer
  • tdd-guide.md — TDD methodology guide agent

Skills (12)

  • backend-patterns/ — backend architecture patterns
  • clickhouse-io/ — ClickHouse database integration
  • coding-standards/ — coding standards enforcement
  • continuous-learning/ — continuous improvement tracking
  • eval-harness/ — evaluation framework
  • frontend-patterns/ — frontend architecture patterns
  • project-guidelines-example/ — project guideline templates
  • prompt-patterns/ — prompt engineering patterns
  • security-review/ — security review methodology
  • strategic-compact/ — strategic context management
  • tdd-workflow/ — TDD workflow skill
  • verification-loop/ — implementation verification

Plugin Marketplace (config/plugins.json)

Connects to two marketplaces:

  • claude-plugins-official (anthropics/claude-plugins-official)
  • anthropic-agent-skills (anthropics/skills)

Plugins installed by profile:

  • standard: security-guidance, commit-commands, pr-review-toolkit, feature-dev, code-review, claude-md-management
  • full: + typescript-lsp, gopls-lsp, pyright-lsp, tailwind-css-lsp, eslint-lsp, jest-runner, playwright-runner, python-debugger

Wizard (wizard/wizard.sh)

Interactive CLI with prompts:

  1. Profile selection (minimal/standard/full)
  2. Language (ja/en)
  3. Editor (vscode/cursor/none)
  4. Terminal preferences (Ghostty install on macOS if desired)
  5. Font installation (cross-platform)
05

Prompts

cloudnative-co/claude-code-starter-kit — Prompts

Verbatim Excerpt 1: /orchestrate command

# Orchestrate Command

Sequential agent workflow for complex tasks.

## Usage

`/orchestrate [workflow-type] [task-description]`

## Workflow Types

### feature
Full feature implementation workflow:

planner -> tdd-guide -> code-reviewer -> security-reviewer


### bugfix
Bug investigation and fix workflow:

explorer -> tdd-guide -> code-reviewer


## Execution Pattern

For each agent in the workflow:

1. **Invoke agent** with context from previous agent
2. **Collect output** as structured handoff document
3. **Pass to next agent** in chain
4. **Aggregate results** into final report

## Handoff Document Format

Between agents, create handoff document:

```markdown
## HANDOFF: [previous-agent] -> [next-agent]

### Context
[Summary of what was done]

### Findings
[Key discoveries or decisions]

### Files Modified
[List of files touched]

### Open Questions
[Unresolved issues for next agent]

**Prompting technique**: Sequential agent chaining with structured handoff documents. Each agent receives the output of the previous agent as a formatted HANDOFF document. The workflow type is a dispatch key that selects the agent chain.

---

## Verbatim Excerpt 2: `tdd.md` command structure

```markdown
# TDD Command

Test-Driven Development workflow with mandatory test-first approach.

## Workflow

1. **Write failing test** - Define expected behavior
2. **Verify test fails** - Confirm test is actually testing something
3. **Implement** - Write minimal code to pass
4. **Refactor** - Clean up while keeping tests green

## Gate

Before writing implementation code:
- Tests must exist
- Tests must fail for the right reason
- Test intention documented

Never skip to implementation without a failing test.

Prompting technique: Ordered instruction list with explicit gate. "Never skip" rule is a behavioral constraint. The "fail for the right reason" verification is TDD sophistication beyond simple test-first.

09

Uniqueness

cloudnative-co/claude-code-starter-kit — Uniqueness

differs_from_seeds

Closest to spec-kit (commands + skills + profile system) but implemented as wizard-driven shell installer rather than Python CLI, and installs globally to ~/.claude/ rather than per-project. The plugin marketplace integration via config/plugins.json pointing at anthropics/claude-plugins-official and anthropics/skills is unique — no seed routes installations through the official Anthropic plugin registry, making this the only framework that auto-installs official Anthropic plugins based on chosen profile. The /orchestrate sequential agent chain with HANDOFF documents resembles claude-flow's chaining but is command-driven (not MCP). Japanese-language documentation and Windows PowerShell support are unique in the batch. The global ~/.claude/ installation scope (vs project-local) is a distinct architectural choice.

Positioning

"One-command Claude Code setup for beginners" targeting Japanese developers and non-developer users. The Ministry of Education security advisor authorship suggests enterprise/education institutional use cases.

Observable Failure Modes

  1. Global installation side effects: Installing to ~/.claude/ means ALL Claude Code sessions on the machine are affected. A wrong installation can break unrelated projects.
  2. Plugin marketplace availability: If anthropics/claude-plugins-official changes structure or becomes unavailable, plugin installation fails silently.
  3. jq dependency: build_settings_file() requires jq. Missing jq on minimal macOS/Windows systems causes install failure.
  4. Ghostty macOS-only: lib/ghostty.sh only works on macOS; Linux/Windows users see this step skipped.
  5. Japanese-first content: English users need the --language=en flag or the wizard language prompt.

Cross-References

  • Plugin marketplace references: anthropics/claude-plugins-official, anthropics/skills
  • Spec-kit init command (spec-kit-init.md) suggests awareness of spec-kit pattern
04

Workflow

cloudnative-co/claude-code-starter-kit — Workflow

Setup Phases

Phase Action Artifact
1. Run installer `curl ... bashorirm ...
2. Wizard interaction Profile, language, editor selection Configuration choices recorded
3. OS detection lib/detect.sh detects macOS/Linux/WSL/MSYS Correct deploy path selected
4. Prerequisites check lib/prerequisites.sh verifies dependencies Missing deps reported
5. CLAUDE.md assembly build_claude_md() assembles from template + i18n ~/.claude/CLAUDE.md
6. Settings assembly jq merges base JSON + permissions + hook fragments ~/.claude/settings.json
7. Deploy Agents/rules/commands/skills/memory → ~/.claude/ Installed components
8. Plugin install marketplace plugins installed per profile Official Anthropic plugins
9. Optional: Ghostty Terminal install (macOS only) Ghostty terminal configured
10. Optional: Fonts Cross-platform font install Dev fonts installed

Orchestrate Command Workflow

The /orchestrate [workflow-type] [task] command implements sequential agent chains:

Workflow Agent Chain
feature planner → tdd-guide → code-reviewer → security-reviewer
bugfix explorer → tdd-guide → code-reviewer
refactor architect → code-reviewer → tdd-guide
security security-reviewer → code-reviewer → architect

Between each agent: handoff document created with Context, Findings, Files Modified, Open Questions.

Update Workflow

/update-kit         # Live update from upstream
/update-kit-dry-run # Preview changes first

write_manifest() tracks deployed files for uninstall. _write_snapshot() saves deployed state for future comparison.

Approval Gates

No formal gates. The wizard interaction is the primary approval step (profile selection). The --non-interactive flag bypasses all prompts.

06

Memory Context

cloudnative-co/claude-code-starter-kit — Memory & Context

Memory Type

File-based. Installs to ~/.claude/memory/ (global user memory).

Memory System

The installer deploys files to ~/.claude/ globally, meaning the installed agents/rules/commands/skills are available across ALL projects (not project-local). This is a global installation pattern, distinct from most starters that operate project-locally.

CLAUDE.md Assembly

build_claude_md() in lib/deploy.sh assembles ~/.claude/CLAUDE.md from:

  1. Base template
  2. i18n overlay (ja/en)
  3. Profile-specific additions

The assembled CLAUDE.md is global — applies to all Claude Code sessions by default.

State Files

  • ~/.claude/CLAUDE.md — global assembled context
  • ~/.claude/settings.json — global assembled settings
  • ~/.claude/memory/ — global memory files
  • Manifest file (for uninstall tracking)
  • Snapshot file (for update comparison)

Compaction Handling

Not explicitly addressed. strategic-compact skill name suggests compaction strategy support but content not deeply analyzed.

Cross-Session Handoff

Yes — via ~/.claude/CLAUDE.md (global, persists across all sessions and projects).

Persistence Scope

Global (unusual) — most other starters are project-local. This means all Claude Code sessions across all projects benefit from the installed configuration.

Update Mechanism

/update-kit command + /update-kit-dry-run for preview. _write_snapshot() tracks installed state for comparison. This is the most complete update management system in the batch.

07

Orchestration

cloudnative-co/claude-code-starter-kit — Orchestration

Multi-Agent Support

Yes — 9 named agents with /orchestrate sequential chaining.

Orchestration Pattern

sequential — the /orchestrate command explicitly chains agents in predefined sequences with handoff documents between each step.

Multi-Model Usage

No — single model (Claude Code). No multi-model routing.

Isolation Mechanism

None (in-place editing).

Execution Mode

Interactive-loop for most commands. The /orchestrate command implements a multi-step agent chain.

Hooks

Not explicitly documented in the analyzed content. The build_settings_file() assembles settings.json from "hook fragments" suggesting some hooks are included, but specific hook events were not enumerated in the available content.

Consensus Mechanism

None.

Crash Recovery

None explicit.

Update Management

The /update-kit + /update-kit-dry-run command pair is the most sophisticated update mechanism in the batch — it tracks what was installed (via manifest + snapshot) and can update in place without full reinstall.

Windows Cross-Platform

install.ps1 handles WSL2 setup on Windows, then delegates to setup.sh. This cross-platform support is unique among the batch's starters.

08

Ui Cli Surface

cloudnative-co/claude-code-starter-kit — UI / CLI Surface

Dedicated CLI Binary

No — setup.sh is the primary executable, but it's an install script not a binary.

Wizard Interface

wizard/wizard.sh — interactive terminal wizard with prompts:

  • Profile selection (minimal/standard/full)
  • Language selection (ja/en)
  • Editor selection (vscode/cursor/none)
  • Terminal preferences

The wizard is the richest interactive setup experience in the batch.

Local Web Dashboard

No.

IDE Integration

VSCode and Cursor explicitly supported. The wizard detects editor choice and configures accordingly.

Windows Support

install.ps1 — PowerShell installer that sets up WSL2 and runs setup.sh. Unique in the batch.

Terminal Support

lib/ghostty.sh — optional Ghostty terminal installation (macOS only). Cross-platform font installation (lib/fonts.sh) including Windows Terminal auto-config.

ShellCheck CI

.github/workflows/shellcheck.yml — ShellCheck runs automatically on PRs (severity: warning). bash tests/run-unit-tests.sh and bash tests/run-scenarios.sh also run in CI.

Observability

  • Manifest file tracks installed components (for uninstall)
  • Snapshot file tracks installed state (for updates)
  • /update-kit-dry-run shows what would change before updating

Internationalization

i18n/ directory with Japanese (ja) and English (en) content. The wizard language selection determines which i18n overlay is applied to the assembled CLAUDE.md.

Related frameworks

same archetype · same primary tool · same memory type

Claude-Flow / Ruflo ★ 55k

Eliminates single-agent context limits and sequential bottlenecks by orchestrating fault-tolerant swarms of specialized AI agents…

Hermes Agent (NousResearch) ★ 168k

Self-improving personal AI agent with closed learning loop, 7 terminal backends, and messaging gateway — not tied to any AI…

OpenCode ★ 165k

Terminal-first AI coding agent with multi-model routing, native desktop app, and a typed .opencode/ configuration system for…

OpenHands ★ 75k

Open-source AI software development platform (open-source Devin alternative) with Docker sandbox isolation, 77.6% SWE-bench…

DeerFlow ★ 70k

Long-horizon superagent that researches, codes, and creates by orchestrating parallel sub-agents with isolated contexts in Docker…

oh-my-openagent (omo) ★ 60k

Multi-provider AI agent orchestration for OpenCode: escape vendor lock-in by routing Sisyphus (Claude/Kimi/GLM) and Hephaestus…