Tasker System

Micro-task development workflow for Claude Code with specialized AI agents. Optimized for startup velocity with focused, single-task iterations.

Installation

Option 1: NPX

npx claude-tasker init

Option 2: Manual Setup (Current)

1. Navigate to your project directory and copy the required files:

   # Copy configuration files from claude-tasker repository
   cp /path/to/claude-tasker/.claude .
   cp /path/to/claude-tasker/.claude-settings.json .
   cp /path/to/claude-tasker/CLAUDE.md .

2. Initialize the docs structure:

   mkdir -p docs/features
   echo "# Global Business Ideas" > docs/backlog.md

3. Verify setup by checking files are in place:

   ls -la .claude* CLAUDE.md docs/

Quick Start

Once installed, use these commands in Claude Code:

/tasker-start [feature-name]    # Create new feature with business analysis
/tasker-continue [feature-name] # Resume workflow with intelligent next-step detection
/tasker-status [feature-name]   # Show current progress and backlog management
/tasker-implement [feature-name] # Execute development task with strict TDD
/tasker-close [feature-name]    # Mark feature complete, cleanup files

Required Files

The Tasker system requires these files in your project root:

• .claude - Claude Code hooks and workflow commands
• .claude-settings.json - Tool configuration and TDD enforcement hooks
• CLAUDE.md - Complete development guidelines and workflow documentation
• docs/backlog.md - Global business ideas backlog
• docs/features/ - Feature-specific workflows and backlogs

Intelligent Workflow

Philosophy: Each agent focuses on ONE task only, everything else goes to backlog for later prioritization.

Smart Agent Skipping: Agents automatically detect when they're not needed and skip themselves:

• Business Analyst: Skips technical fixes, refactoring, internal tooling
• Product Manager: Skips pure backend work, technical changes with no UX impact
• Solution Architect: Skips simple bug fixes, low-risk changes, styling/config
• QA Engineer: Skips refactoring, docs, config changes with no user-facing impact
• Security Analyst: Skips styling, docs, changes with no data/auth/external interactions

Agents Process:

1. Business Analyst → Business need (2-3 lines) + ideas to global backlog
2. Product Manager → UX details (2-3 lines) + feature ideas to backlog
3. Solution Architect → Risk assessment (can REJECT and redirect to safer alternatives)
4. Tech Lead → Single implementation task (1-2 days max) + Frontend/Backend focus
5. Full-Stack Developer → TDD implementation (one side per task)
6. QA Engineer → Quick validation with ✅ QA APPROVED marker
7. Security Analyst → Fast security check with ✅ SECURITY APPROVED marker

File Structure

docs/
├── backlog.md                           # Global business ideas across features
└── features/
    ├── YY-MM-DD-[feature-name]/         # Active features (folders)
    │   ├── work.md                      # Current single task (all agents append)
    │   └── backlog.md                   # Feature-specific backlog by section
    └── YY-MM-DD-[completed].md          # Done features (files)

Status Logic: Folders = active work, Files = completed features

Append-Only Workflow

All agents append to single work.md with standardized approval markers:

**Business Need**: [2-3 sentences about problem and solution]
**Current Task**: [The one thing we'll build]  
**Success Metric**: [How we know this task worked]

**User Experience**: [How users will interact with this task]
**Priority Rationale**: [Why this task first vs alternatives]

**✅ TECHNICAL APPROVED**
**Implementation Notes**: [Key technical approach or constraints]

**Development Task**: [Single focused implementation story]
**Focus**: [Frontend | Backend] - which side to implement
**Files to Change**: 
- Create: [specific-file-path]
- Modify: [specific-file-path:line-numbers]

**✅ QA APPROVED**
**Acceptance Criteria Verified:**
- [✓] [Criteria 1] - Works as expected

**✅ SECURITY APPROVED**  
**Security Check:**
- [✓] Input validation present
- [✓] No obvious injection vulnerabilities

Two-Tier Backlog System

Global Backlog (docs/backlog.md): Cross-feature concepts and new feature ideas

Feature Backlog (backlog.md): Sectioned by role (Product Features, Technical Concerns, Development Tasks)

Decision Gates and User Control

• Solution Architect can REJECT tasks → redirects to safer alternatives
• QA Engineer can REJECT implementation → back to Developer for fixes
• Security Analyst can BLOCK deployment → back to Developer for security fixes
• User chooses next direction after each task completion

TDD Enforcement

TEST-DRIVEN DEVELOPMENT IS NON-NEGOTIABLE

• Red→Green→Refactor cycle enforced
• Every single line of production code written in response to failing test
• No exceptions, no shortcuts

Key Features

• Micro-tasks: 1-2 day implementation cycles maximum
• Single focus: One agent, one task at a time
• Quality gates: Standardized approval markers for QA and Security
• Two-tier backlog: Global ideas + feature-specific backlogs
• Strict TDD: Non-negotiable test-first development
• Intelligent skipping: Agents skip themselves when not needed

See CLAUDE.md for comprehensive development guidelines and technology stack details.