dream-team

Dream Team

A Claude Code plugin for planning and executing development projects across three execution modes.

Overview

Dream Team provides structured planning skills that generate spec files, and a universal build skill that reads the spec and executes it using the appropriate strategy. It ships with eight specialized agents for common development roles.

Prerequisites

For all features to work — especially the Team execution mode — you need to enable experimental agent teams in Claude Code:

export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

See the Agent Teams documentation for details.

A terminal multiplexer like tmux or iTerm2 is recommended, as agent teams spawn multiple Claude instances that benefit from separate panes/windows for monitoring.

Optional: Playwright MCP

If your specs set playwright: true, builder and tester agents will use Playwright to verify UI changes visually (navigate pages, take screenshots, interact with elements, check for console errors). Install the Playwright MCP server to enable this:

claude mcp add playwright npx @playwright/mcp@latest

Without it, agents will skip visual verification and note it in their reports.

Frontend Design Direction

For frontend/UI work, the planning skill asks about aesthetic direction (Minimal, Editorial, Brutalist, Retro-Futuristic, and others) along with color palette, dark/light mode, and typography preferences. When the spec sets frontend-design: true, the build skill injects comprehensive design guidelines into builder prompts — covering anti-generic rules, typography, animation timing, interaction patterns, accessibility, and component library recommendations. This prevents the default "AI look" of generic cards and purple gradients.

Installation

Register the repository as a local marketplace, then install:

claude plugin marketplace add Ratler/dream-team-marketplace
claude plugin install dream-team@dream-team-marketplace

Usage

1. Brainstorm your idea:

   /dream-team:plan Build a REST API for user management
   

2. Write the spec (after brainstorming completes):

   /dream-team:spec-delegated
   

3. Execute:

   /dream-team:build specs/2026-02-08-user-management.md
   

Features At a Glance

Area	Dream Team
Philosophy	Turn Claude into a project lead
Skills	6 pipeline-connected skills (plan, spec x3, build, debug)
Agents	8 named roles with per-agent model and tool policies
Execution modes	Sequential, Delegated, Team
Planning	Formal specs with YAML frontmatter
Task tracking	TaskCreate/TaskUpdate with dependency graph
Git workflow	Built into execution (branch, commit, review gates)
Code review	Configurable review policy with retry loops
Team support	Parallel Claude instances via experimental agent teams
Playwright	First-class integration (opt-in per spec)
Frontend design	Aesthetic direction, anti-generic rules, guidelines
Validation	Structural spec checks, dep graph, preflight hooks
Build resume	Interrupted builds resume from last completed task

How It Works

Dream Team separates the workflow into three distinct phases: brainstorm, spec, build. Each phase has its own command and produces a clear handoff to the next.

1. Plan (brainstorm)

/dream-team:plan starts an interactive conversation. It explores the codebase, asks clarifying questions one at a time, proposes 2-3 approaches with trade-offs, and walks through the task breakdown section by section. No files are created; the output is a shared understanding between you and Claude. At the end, it recommends which execution mode fits best.

2. Spec (write)

Once brainstorming is complete, run the spec command for your chosen execution mode. The command you pick determines how the build phase will run:

• /dream-team:spec-sequential — produces a spec for single-session execution
• /dream-team:spec-delegated — produces a spec with agent assignments for orchestrated execution
• /dream-team:spec-team — produces a spec with team configuration for parallel Claude instances