Dream Team v3.0

Effective Claude workflows based on official Anthropic patterns and real-world feedback.

Philosophy

"A single-threaded linear agent maintains continuous context. This simple approach will get you very far." - Cognition (Devin team)

Less ceremony, more results. Claude works best when it can:

• Build context itself
• Ask questions only when genuinely needed
• Work incrementally
• Self-review before presenting

---

Quick Start

Installation

# Option 1: Plugin (Recommended)
claude plugins add https://github.com/ashchupliak/dream-team

# Option 2: Manual
git clone https://github.com/ashchupliak/dream-team.git
cd dream-team
./install.sh

Usage

# For 90% of tasks - use /solo
/solo Fix the login bug
/solo Add rate limiting to the API
/solo Why are the tests failing?

# For complex features needing multiple perspectives
/team Add OAuth with Google and GitHub

---

Commands

/solo - Primary Workflow (Recommended)

The main command for most tasks. Claude will:

1. Build context automatically - Read CLAUDE.md, explore codebase, find patterns
2. Assess confidence - Determine if it has enough info to proceed
3. Ask only when needed - No forced questions, no ceremonies
4. Work incrementally - Small steps that compile and work
5. Self-review - Check own work before presenting

/solo Fix the login bug
/solo Add rate limiting to the API
/solo Why are the tests failing?
/solo Review my changes in the auth module
/solo Refactor the user service

/team - Multi-Agent Workflow

For genuinely complex features where you want:

• Multiple architecture options presented
• Parallel exploration of different aspects
• Explicit user checkpoints

Based on Anthropic's feature-dev plugin.

7-phase workflow: Discovery → Exploration → Questions → Architecture → Implementation → Review → Summary

/team Add OAuth authentication with Google and GitHub
/team Implement real-time notifications system

Legacy Commands

Kept for comparison. Use /solo instead for most tasks.

---

When To Use What

Rule of thumb: Start with /solo. Use /team only when you genuinely need multiple perspectives.

---

Skills

Specialized knowledge Claude can invoke on demand:

Just mention the domain in your request.

---

Agents

Available for /team workflows:

---

Project Setup

For best results, add a CLAUDE.md to your project root:

# Project Name

## Commands
- Build: npm run build
- Test: npm test
- Lint: npm run lint

## Tech Stack
- Next.js 14, TypeScript, Prisma

## Conventions
- Use server components by default
- API routes in /app/api/
- Database queries through Prisma only

## Current Focus
- Working on user authentication

This helps Claude understand your specific project.

---

Research Behind This

Based on:

• Anthropic: Claude Code Best Practices
• Anthropic: Multi-Agent Research System
• Cognition: Don't Build Multi-Agents
• [Anthropic: feature-dev Plugin](https://github.com/anthropics/cla

...