Join the Skool - https://www.skool.com/iss-ai-automation-school-6342/about

A simple yet powerful orchestration system for Claude Code that uses specialized agents to manage complex projects from start to finish, with mandatory human oversight and visual testing.

This is a custom Claude Code orchestration system that transforms how you build software projects. Claude Code itself acts as the orchestrator with its 200k context window, managing the big picture while delegating individual tasks to specialized subagents:

• 🧠 Claude (You) - The orchestrator with 200k context managing todos and the big picture
• ✍️ Coder Subagent - Implements one todo at a time in its own clean context
• 👁️ Tester Subagent - Verifies implementations using Playwright in its own context
• 🆘 Stuck Subagent - Human escalation point when ANY problem occurs

• No Fallbacks: When ANY agent hits a problem, you get asked - no assumptions, no workarounds
• Visual Testing: Playwright MCP integration for screenshot-based verification
• Todo Tracking: Always see exactly where your project stands
• Simple Flow: Claude creates todos → delegates to coder → tester verifies → repeat
• Human Control: The stuck agent ensures you're always in the loop

1. Claude Code CLI installed (get it here)
2. Node.js (for Playwright MCP)

# Clone this repository
git clone https://github.com/IncomeStreamSurfer/claude-code-agents-wizard-v2.git
cd claude-code-agents-wizard-v2

# Start Claude Code in this directory
claude

That's it! The agents are automatically loaded from the .claude/ directory.

When you want to build something, just tell Claude your requirements:

You: "Build a todo app with React and TypeScript"

Claude will automatically:

1. Create a detailed todo list using TodoWrite
2. Delegate the first todo to the coder subagent
3. The coder implements in its own clean context window
4. Delegate verification to the tester subagent (Playwright screenshots)
5. If ANY problem occurs, the stuck subagent asks you what to do
6. Mark todo complete and move to the next one
7. Repeat until project complete

USER: "Build X"
    ↓
CLAUDE: Creates detailed todos with TodoWrite
    ↓
CLAUDE: Invokes coder subagent for todo #1
    ↓
CODER (own context): Implements feature
    ↓
    ├─→ Problem? → Invokes STUCK → You decide → Continue
    ↓
CODER: Reports completion
    ↓
CLAUDE: Invokes tester subagent
    ↓
TESTER (own context): Playwright screenshots & verification
    ↓
    ├─→ Test fails? → Invokes STUCK → You decide → Continue
    ↓
TESTER: Reports success
    ↓
CLAUDE: Marks todo complete, moves to next
    ↓
Repeat until all todos done ✅

Your 200k Context Window

• Creates and maintains comprehensive todo lists
• Sees the complete project from A-Z
• Delegates individual todos to specialized subagents
• Tracks overall progress across all tasks
• Maintains project state and context

How it works: Claude IS the orchestrator - it uses its 200k context to manage everything

Fresh Context Per Task

• Gets invoked with ONE specific todo item
• Works in its own clean context window
• Writes clean, functional code
• Never uses fallbacks - invokes stuck agent immediately
• Reports completion back to Claude

When it's used: Claude delegates each coding todo to this subagent

Fresh Context Per Verification

• Gets invoked after each coder completion
• Works in its own clean context window
• Uses Playwright MCP to see rendered output
• Takes screenshots to verify layouts
• Tests interactions (clicks, forms, navigation)
• Never marks failing tests as passing
• Reports pass/fail back to Claude

When it's used: Claude delegates testing after every implementation

Fresh Context Per Problem

• Gets invoked when coder or tester hits a problem
• Works in its own clean context window
• ONLY subagent that can ask you questions
• Presents clear options for you to choose
• Blocks progress until you respond
• Returns your decision to the calling agent
• Ensures no blind fallbacks or workarounds

When it's used: Whenever ANY subagent encounters ANY problem

This is the key differentiator:

Traditional AI: Hits error → tries workaround → might fail silently This system: Hits error → asks you → you decide → proceeds correctly

Every agent is hardwired to invoke the stuck agent rather than use fallbacks. You stay in control.

You: "Build a landing page with a contact form"

Claude creates todos:
  [ ] Set up HTML structure
  [ ] Create hero section
  [ ] Add contact form with validation
  [ ] Style with CSS
  [ ] Test form submission

Claude invokes coder(todo #1: "Set up HTML structure")

Coder (own context): Creates index.html
Coder: Reports completion to Claude

Claude invokes tester("Verify HTML structure loads")

Tester (own context): Uses Playwright to navigate
Tester: Takes screenshot
Tester: Verifies HTML structure visible
Tester: Reports success to Claude

Claude: Marks todo #1 complete ✓

Claude invokes coder(todo #2: "Create hero section")

Coder (own context): Implements hero section
Coder: ERROR - image file not found
Coder: Invokes stuck subagent

Stuck (own context): Asks YOU:
  "Hero image 'hero.jpg' not found. How to proceed?"
  Options:
  - Use placeholder image
  - Download from Unsplash
  - Skip image for now

You choose: "Download from Unsplash"

Stuck: Returns your decision to coder
Coder: Proceeds with Unsplash download
Coder: Reports completion to Claude

... and so on until all todos done

.
├── .claude/
│   ├── CLAUDE.md              # Orchestration instructions for main Claude
│   └── agents/
│       ├── coder.md          # Coder subagent definition
│       ├── tester.md         # Tester subagent definition
│       └── stuck.md          # Stuck subagent definition
├── .mcp.json                  # Playwright MCP configuration
├── .gitignore
└── README.md

• SEO Grove - AI-powered SEO automation platform
• ISS AI Automation School - Join our community to learn AI automation
• Income Stream Surfers YouTube - Tutorials, breakdowns, and AI automation content

Have questions or want to share what you built?

• Join the ISS AI Automation School community
• Subscribe to Income Stream Surfers on YouTube
• Check out SEO Grove for automated SEO solutions

This is an open system! Feel free to:

• Add new specialized agents
• Improve existing agent prompts
• Share your agent configurations
• Submit PRs with enhancements

This system leverages Claude Code's subagent system:

1. CLAUDE.md instructs main Claude to be the orchestrator
2. Subagents are defined in .claude/agents/*.md files
3. Each subagent gets its own fresh context window
4. Main Claude maintains the 200k context with todos and project state
5. Playwright MCP is configured in .mcp.json for visual testing

The magic happens because:

• Claude (200k context) = Maintains big picture, manages todos
• Coder (fresh context) = Implements one task at a time
• Tester (fresh context) = Verifies one implementation at a time
• Stuck (fresh context) = Handles one problem at a time with human input
• Each subagent has specific tools and hardwired escalation rules

1. Trust Claude - Let it create and manage the todo list
2. Review screenshots - The tester provides visual proof of every implementation
3. Make decisions when asked - The stuck agent needs your guidance
4. Don't interrupt the flow - Let subagents complete their work
5. Check the todo list - Always visible, tracks real progress

• Use /agents command to see all available subagents
• Claude maintains the todo list in its 200k context - check anytime
• Screenshots from tester are saved and can be reviewed
• Each subagent has specific tools - check their .md files
• Subagents get fresh contexts - no context pollution!

MIT - Use it, modify it, share it!

Built by Income Stream Surfer

Powered by Claude Code's agent system and Playwright MCP.

Ready to build something amazing? Just run claude in this directory and tell it what you want to create! 🚀