project-starter

A universal Claude Code workflow plugin with specialized agents, skills, hooks, and mode commands for any software project. Compatible with skills.sh — works with Claude Code, Cursor, Codex, and 35+ AI agents.

Quick Start

Option 1: skills.sh (Recommended — Any Agent)

npx skills add CloudAI-X/claude-workflow-v2

Installs skills to Claude Code, Cursor, Codex, Windsurf, Cline, and 35+ other AI agents automatically.

Option 2: npx (Claude Code — Full Plugin)

npx install-claude-workflow-v2@latest

Copies agents, commands, skills and hook scripts into .claude/ and registers the hooks in .claude/settings.json (existing files and settings are kept). Two differences from a plugin install: commands are not namespaced (/commit instead of /project-starter:commit), and permissionMode: acceptEdits on the debugger, docs-writer, refactorer and test-architect agents takes effect (Claude Code ignores that field for plugin agents).

Option 3: CLI (Per-Session)

# Clone the plugin
git clone https://github.com/CloudAI-X/claude-workflow-v2.git

# Run Claude Code with the plugin
claude --plugin-dir ./claude-workflow-v2

Option 4: Agent SDK

import { query } from "@anthropic-ai/claude-agent-sdk";

for await (const message of query({
  prompt: "Hello",
  options: {
    plugins: [{ type: "local", path: "./claude-workflow-v2" }],
  },
})) {
  // Plugin commands, agents, and skills are now available
}

Option 5: Install Permanently

# Add this repository as a marketplace, then install the plugin from it
claude plugin marketplace add CloudAI-X/claude-workflow-v2
claude plugin install project-starter@claude-workflow

Verify Installation

After loading the plugin, verify it's working:

> /plugin

Tab to Installed - you should see project-starter listed. Tab to Errors - should be empty (no errors).

These commands become available:

/project-starter:architect    # Architecture-first mode
/project-starter:rapid        # Ship fast mode
/project-starter:commit       # Auto-generate commit message
/project-starter:verify-changes  # Multi-agent verification

What's Included

Usage Examples

Commands in Action

Auto-commit your changes:

> /project-starter:commit

Looking at staged changes...
✓ Created commit: feat(auth): add JWT refresh token endpoint

Full git workflow:

> /project-starter:commit-push-pr

✓ Committed: feat: add user dashboard
✓ Pushed to origin/feature/dashboard
✓ Created PR #42: https://github.com/you/repo/pull/42

Verify before shipping:

> /project-starter:verify-changes

Spawning verification agents...
├─ build-validator: ✓ Build passes
├─ test-runner: ✓ 42 tests pass
├─ lint-checker: ⚠ 2 warnings (non-blocking)
└─ security-scanner: ✓ No vulnerabilities

Ready to ship!

Agents in Action

Agents spawn automatically based on your request:

You say: "The login is broken, users get 401 errors"

[debugger agent activated]
→ Checking auth middleware... found issue
→ Token validation uses wrong secret in production
→ Fix: Update AUTH_SECRET in .env.production

You say: "Review my changes"

[code-reviewer agent activated]
→ Analyzing 3 files changed...
✓ Logic is correct
⚠ Missing null check on line 42
⚠ Consider adding rate limiting to this endpoint

You say: "Add authentication to the API"

[orchestrator agent activated]
→ Breaking down into subtasks:
  1. Design auth schema (applying the designing-architecture skill)
  2. Implement JWT middleware
  3. Add login/register endpoints
  4. Write tests (spawning test-architect)
  5. Update API docs (spawning docs-writer)

Skills in Action

Skills provide domain knowledge automatically:

You ask: "How should I structure the payment service?"

[designing-architecture skill applied]
→ Recommending hexagonal architecture
→ Payment providers as adapters
→ Core domain isolated from infrastructure

You ask: "Make this endpoint faster"

[optimizing-performance skill applied]
→ Adding database indexes
→ Implementing response caching
→ Using pagination for large results

Hooks in Action

Hooks run automatically on events:

Security block (pre-edit):

🚫 BLOCKED - Security issue detected in src/config.ts:
  - Potential API key detected → Move to .env file and use environment variables (e.g., process.env.API_KEY)

Auto-format (post-edit):

The edited file is formatted silently with the project's own formatter; the hook prints nothing.

Desktop notifications:

🔔 "Claude needs input" - when waiting for your response
🔔 "Task complete" - when finished

Commands Reference

All commands use the format /project-starter:<command>.

Mode Commands (Output Styles)

Git Workflow (Inner-Loop)

Verification

Parallel

Planning & Refactoring

Onboarding & Knowledge

Agents

Agents are specialized subagents that Claude spawns automatically based on your task.