Wire a Complete Claude Code Workflow with project-starter

Most Claude Code setups are assembled piecemeal — one agent here, a hook there. claude-workflow-v2 packages the whole inner loop as a single plugin named project-starter: 7 specialized agents, 26 slash commands, 14 knowledge skills, and 14 automation hooks.

Why This Skill Matters

The plugin covers the daily loop end to end. Agents spawn automatically from your request — say "the login is broken" and the debugger agent investigates; ask for a review and the code-reviewer agent takes it. Skills such as designing-architecture and optimizing-performance supply domain knowledge when the task touches them. Hooks run without asking: a security scan blocks writes that look like leaked secrets, auto-format runs prettier or black after edits, and desktop notifications fire when input is needed. The README lists it as compatible with Claude Code, Cursor, Codex, and 35+ AI agents via skills.sh, and the repo has 1,415 stars on the SkillMap leaderboard.

Installation

Full plugin for Claude Code (v1.0.33 or later):

npx install-claude-workflow-v2@latest

Skills only, for any agent:

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

Note the difference: npx skills add installs the 14 skills only. Agents, commands, and hooks need the full plugin install. Hooks also require Python 3.

Verify after loading: run /plugin, tab to Installed — project-starter should be listed — and tab to Errors, which should be empty.

Real Workflow: Commit, Then Verify Before Shipping

1. Stage your changes and run the auto-commit command:

> /project-starter:commit

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

2. Before opening a PR, run multi-agent verification:

> /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!

Both transcripts come straight from the README's examples. A second workflow uses the output styles: /project-starter:architect forces architecture-before-code mode, /project-starter:rapid optimizes for shipping fast, and /project-starter:mentor explains the why as it works.

Tips

• The full command set splits into output styles (4), git workflow (7), verification (6), parallel review (2), planning and refactoring (3), and onboarding and knowledge (4) — /project-starter:commit-push-pr chains commit, push, and PR creation in one step.
• Copy templates/settings.local.json.template into your project's .claude/ to pre-allow common safe commands and cut permission prompts.
• Install the GitHub Action with claude /install-github-action, then tag @claude in PR comments to get suggestions and CLAUDE.md updates in review.
• Extend it with your own .md command and agent files — the README documents the frontmatter format for both.

When Not to Use This

The full plugin is Claude Code-only; on Cursor, Codex, or other agents you get the 14 skills, not the agents, commands, or hooks. And if your team already standardized on a different command set, layering 26 slash commands on top adds discovery cost before it adds value — start with the templates instead.

See the leaderboard for more skills.