Claude Code Task Completion System

A multi-agent orchestration system that automates feature implementation with production-ready quality standards.

We've restructured the command system for better clarity and alignment with newer Claude models (Opus 4.6, Sonnet 4.6, Haiku 4.5):

✨ What's New:

• Strengthened Orchestrator Pattern: All commands now explicitly enforce the "orchestrator only" pattern with imperative language ("IMMEDIATELY call Task tool"), ensuring better delegation to specialized agents
• Cleaner Command Names:
  • task_complete.md — Comprehensive feature implementation (unchanged)
  • task_easy.md — Simple problem solving with reasoning (unchanged)
  • task_tool_finalize.md — Error fixing & task completion (formerly task_finish)
  • commit.md — Git commit conventions (new)
  • Removed task_code.md to reduce redundancy

🔧 Technical Improvements:

• Removed dangling variable declarations in command files
• More direct, imperative instructions to prevent model over-confidence
• Better blocking gates between sequential agent calls
• Increased error-fixing cycles from 5 to 20 for task_tool_finalize

Why This Matters: Newer Claude models are more confident and capable, which means they sometimes skip delegation to specialized agents. The updated language patterns ensure the orchestrator stays focused on coordination while agents handle implementation.

From the creator:

I've spent whole weekends coding huge tasks with AI. Starting with Chats (ChatGPT, Claude, DeepSeek, Gemini), to Cursor, to Claude Code. I ran often into the limitations of AI coding agents and I'm always trying to optimize my prompts.

With the introduction of Sub Agents and MCP servers, I finally found a workflow for complex tasks to have them implemented with a high quality standard that suits my projects. I also tested Ralph but i was not happy with its results.

I',m happy to see that people starred my repo, which was a personal motivation to open what works on one project on others and collaborate in improving coding workflows as a community.

I'm also working towards making this work with Codex! My subscriptions are Claude on the 100 USD plan and ChatGPT on the 20 USD plan and I often run this command parallel to develop 3-4 features at once. The limits are fine as a heavy user but still needs to be watched more than single code tabs. Let's vibe

curl -fsSL https://raw.githubusercontent.com/codeoutin/claude-code-agency/main/install.sh | bash

Or download install.sh and run: ./install.sh

The fastest way to set up the task completion system in your project:

• Claude Code CLI installed and configured
• Anthropic account (Pro/Max plan recommended for complex tasks)
• Git for version control

1. Navigate to your project directory

   cd /path/to/your/project

2. Initialize with Claude Code and install task completion system

   # Initialize Claude Code in your project
   claude
   /init
   
   # Clone the task completion system
   git clone https://github.com/codeoutin/claude-code-agency.git temp-claude-agency
   
   # Copy the task completion components
   cp temp-claude-agency/examples/settings.json .claude/settings.json
   cp temp-claude-agency/examples/settings.local.json .claude/settings.local.json
   cp -r temp-claude-agency/agents .claude/
   cp -r temp-claude-agency/commands .claude/
   cp -r temp-claude-agency/examples .
   cp -r temp-claude-agency/docs .
   
   # Clean up
   rm -rf temp-claude-agency

3. Customize Your Setup

   a. Set up project templates (AI-assisted)

   # Copy the templates
   cp examples/PROJECT_STATUS.md .
   cp examples/PROJECT_DESCRIPTION.md .
   
   # Use Claude Code to customize them for your project
   claude "Please customize PROJECT_STATUS.md and PROJECT_DESCRIPTION.md for my project. Replace all placeholder content with details specific to my codebase and project goals."

   b. Configure agent settings

   # Customize test credentials
   claude "Update .claude/agents/tc-frontend-tester.md with my email: your-email@example.com"
   
   # Update technology stack info
   claude "Update .claude/agents/tc-codex-critic.md with my tech stack details based on my package.json and codebase"

   c. Set up local permissions (optional)

   cp examples/settings.local.json .claude/settings.local.json
   # Edit for any project-specific permissions you need

4. Verify Setup

   # Test the task completion commands
   claude-code /task_easy "Test the task completion system setup"

your-project/
├── .claude/
│   ├── settings.json          # Core permissions and MCP servers
│   ├── settings.local.json    # Project-specific permissions (optional)
│   ├── agents/                # Task completion agents
│   │   ├── tc-context-gatherer.md
│   │   ├── tc-implementation-agent.md
│   │   ├── tc-quality-reviewer.md
│   │   ├── tc-frontend-tester.md
│   │   ├── tc-codex-critic.md
│   │   └── tc-task-planner.md
│   └── commands/              # Command definitions
│       ├── task_complete.md
│       ├── task_tool_finalize.md
│       ├── task_easy.md
│       └── commit.md
├── CLAUDE.md                  # Your project documentation
└── claude-tasks/              # Generated task directories (created automatically)

Commands not found:

• Verify files exist: task_complete.md, task_tool_finalize.md, task_easy.md, commit.md in .claude/commands/
• Check that Claude Code CLI is properly installed
• Try restarting your terminal

MCP servers not working:

• Review docs/MCP_SERVERS.md for MCP server setup
• Check that required MCP servers are installed and configured
• Verify .claude/settings.json has correct MCP server references

Agents not accessible:

• Ensure agent files are in .claude/agents/ directory
• Verify agent files have correct frontmatter (name, description, tools, color)
• Check that agent names match references in command files

Transform feature requests like "Add user authentication" into:

• ✅ Complete implementation with error handling
• ✅ TypeScript/lint error validation
• ✅ Automated testing and quality checks
• ✅ Production-ready code documentation

/task_complete — The Full Pipeline

1. Context Gatherer - Analyzes your codebase patterns
2. Task Planner - Creates detailed implementation roadmap
3. Implementation Agent - Writes code with strict validation (5-cycle loop)
4. Quality Reviewer - Verifies all quality gates pass (5-cycle loop)
5. Frontend Tester - Validates UI/UX with automated tests
6. Code Critic - External unbiased quality assessment

/task_tool_finalize — Error Fixing Only

• Focuses on remaining lint/TypeScript errors after /task_complete
• Systematic error resolution (up to 20 cycles)
• Verification that no regressions occurred

/task_easy — Quick Problem Solving

• Uses sequential thinking MCP for deeper reasoning
• Single cycle, focused implementation
• Best for debugging and optimizations

# Full feature implementation (6-agent pipeline)
claude /task_complete "Add user authentication with JWT"

# Fix errors & finalize incomplete tasks
claude /task_tool_finalize "path/to/task/directory"

# Quick problem solving
claude /task_easy "Debug the authentication redirect issue"

# Commit with conventional format
claude /commit

• Maintains consistency across complex architectures
• Prevents breaking existing functionality

• Specialized validation for TS errors
• Component pattern consistency

• Mission-critical functionality that MUST work
• Code that will be reviewed by senior developers

• Simple one-file changes
• Quick prototypes/experiments
• Non-TypeScript projects (limited validation)

• Node.js 18+
• Claude Code CLI (install guide)
• TypeScript project with lint/build commands
• 4GB+ RAM (for Claude Code)
• Anthropic account (Pro/Max plan recommended)

After installation, edit these files for your project:

1. CLAUDE.md - Project overview and development commands
2. .claude/agents/tc-frontend-tester.md - Test credentials
3. .claude/agents/tc-codex-critic.md - Technology stack

Each task creates a timestamped directory with complete audit trail:

claude-tasks/2025-08-27-authentication/
├── context.md              # Codebase analysis
├── plan.md                 # Implementation roadmap  
├── implementation-1.md     # Development log
├── review-1.md            # Quality validation
├── test-results.md        # UI/UX testing
├── final-critique.md      # External assessment
└── screenshots/           # Visual proof

For detailed setup and customization options, see:

• Setup Guide - Comprehensive configuration
• MCP Servers - Required integrations
• Examples - Project-specific configurations

"MCP server not found"

# Verify MCP servers installed
claude --check-mcp

"Dev server won't start"

# Check your package.json scripts match CLAUDE.md configuration
npm run dev

"Validation failed"

# Run quality checks manually
npm run lint
npx tsc --noEmit
npm run build

Before: Ask Claude to "Add user authentication"

• Gets 50% complete then claims it's done
• Breaks existing TypeScript compilation
• No error handling for edge cases
• Missing integration with existing routing

After: Same request with this system

• ✅ Complete JWT auth with refresh tokens
• ✅ Integrates with existing user management
• ✅ Comprehensive error handling & validation
• ✅ TypeScript errors fixed, lint warnings resolved
• ✅ Mobile-responsive login/signup forms tested
• ✅ Ready for production deployment

Found on GitHub/Reddit? This is actively maintained and I welcome feedback, improvements, and real-world use cases.

• Issues & Ideas: GitHub Issues
• Questions?: Start a discussion or reach out directly
• Using this? Share what you're building — would love to hear how it evolves

Created by: Patrick Steger | Version: 2.0 | February 2026