Turning Claude Code into a multi-agent dev team

What it does

This is a workflow plugin for Claude Code and similar agents. It bundles slash commands, specialized subagent personas, domain knowledge prompts, and automation hooks into a single package. The goal is to stop repeating yourself in chat and let the assistant switch into architect, debugger, or security-auditor mode automatically based on what you’re doing.

The interesting bit

The project treats prompt curation as infrastructure. Rather than hoping the model picks the right persona, it hard-codes seven specialist subagents that auto-trigger on keywords like “review” or “broken”, plus guardrail hooks that block secrets or protected files before they reach code. It is essentially an ops manual for AI assistants, packaged as a plugin.

Key highlights

• 26 slash commands covering git workflow, verification, planning, and onboarding
• 7 auto-triggering subagents (orchestrator, debugger, code-reviewer, etc.) activated by task context
• 14 domain skills (architecture, performance, security patterns) loaded as on-demand context
• 14 hooks for pre-commit checks, secret blocking, auto-formatting, and branch protection
• Compatible with Claude Code, Cursor, Codex, Windsurf, Cline, and 35+ other agents via skills.sh

Caveats

• The README describes behaviors but never shows the underlying plugin code or prompt files, so it’s unclear how much is novel logic versus curated system prompts
• The project uses the name “project-starter” internally (commands, title) while living in the “claude-workflow-v2” repository, suggesting naming drift
• Several installation paths reference a marketplace or npx package that are marked “when available,” implying they may not be live yet

Verdict

Worth exploring if you spend your day in Claude Code or Cursor and want pre-canned workflows to cut down on repetitive prompting. Skip it if you’re looking for a standalone agent runtime—this is a configuration and rules layer, not an engine.