⚠️ IMPORTANT NOTICE: Development focus has shifted to the MCP (Model Context Protocol) version of this workflow system. The MCP version provides enhanced features, real-time dashboard, and broader AI tool compatibility.🚀 View the new Spec Workflow MCP →
This Claude Code-specific version remains available for existing users but will receive limited updates.
Automated workflows for Claude Code with intelligent task execution.
Transform your development with structured workflows: Requirements → Design → Tasks → Implementation for new features, plus streamlined Report → Analyze → Fix → Verify for bug fixes.
- Install the workflow globally
`npm i -g @pimzino/claude-code-spec-workflow`
- Run the setup command in your project directory
claude-code-spec-workflow
- 📁 Complete .claude/ structure - All files and directories
- 📝 10 slash commands - 5 spec workflow + 5 bug fix workflow
- 🎯 Intelligent task execution - Automated implementation
- 🤖 4 specialized agents - Enhanced automation
- 📊 Real-time dashboard - Monitor progress visually
- 🔧 Auto-generated commands - One command per task
- 📋 Document templates - Professional spec documents
- ⚙️ Project steering - Persistent context and standards
- ⚡ Smart optimization - Intelligent context sharing and caching
Complete automation in one command:
/spec-create feature-name "Description"
What happens:
- Requirements → User stories + acceptance criteria
- Design → Technical architecture + diagrams
- Tasks → Atomic, agent-friendly breakdown
- Commands → Auto-generated task commands (optional)
Execute tasks:
# Manual control
/spec-execute 1 feature-name
/feature-name-task-1 # Auto-generated
/bug-create issue-name "Description" # Document the bug
/bug-analyze # Find root cause
/bug-fix # Implement solution
/bug-verify # Confirm resolution
/spec-steering-setup # Creates product.md, tech.md, structure.md
📊 Spec Workflow Commands
Command | Purpose |
---|---|
/spec-steering-setup |
Create project context documents |
/spec-create <name> |
Complete spec workflow |
/spec-execute <task-id> |
Manual task execution |
/<name>-task-<id> |
Auto-generated task commands |
/spec-status |
Show progress |
/spec-list |
List all specs |
🐛 Bug Fix Commands
Command | Purpose |
---|---|
/bug-create <name> |
Document bug with structured format |
/bug-analyze |
Investigate root cause |
/bug-fix |
Implement targeted solution |
/bug-verify |
Verify resolution |
/bug-status |
Show bug fix progress |
- Streamlined task implementation
- Context-aware execution with full specification context
- Agent-based implementation with spec-task-executor
4 AI agents for enhanced automation:
Core Workflow: spec-task-executor
, spec-requirements-validator
, spec-design-validator
, spec-task-validator
Note: Agents are optional - everything works with built-in fallbacks.
- Universal context sharing - Steering, specification, AND template documents optimized
- 60-80% token reduction - Eliminates redundant document fetching across all document types
- Triple optimization commands -
get-steering-context
,get-spec-context
, andget-template-context
- Smart document handling - Bug documents use direct reading (no redundancy), templates use bulk loading (high redundancy)
- Improved performance - Faster agent execution with cached context across all workflows
- Automatic fallback - Maintains reliability with individual
get-content
when optimization unavailable - Session-based caching - Intelligent file change detection and cache invalidation
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard
- Live progress tracking
- WebSocket updates
- Git integration
- Modern UI with Tailwind CSS
Share your dashboard securely with external stakeholders through temporary HTTPS URLs:
# Start dashboard with tunnel
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard --tunnel
# With password protection
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard --tunnel --tunnel-password mySecret123
# Choose specific provider
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard --tunnel --tunnel-provider cloudflare
Tunnel Features:
- 🔒 Secure HTTPS URLs - Share dashboard with managers, clients, or remote team members
- 👁️ Read-Only Access - External viewers cannot modify any project data
- 🔑 Optional Password - Protect access with password authentication
- 🌐 Multiple Providers - Automatic fallback between Cloudflare and ngrok
- 📊 Usage Analytics - Track who accessed your dashboard and when
- ⏰ Auto-Expiration - Tunnels close when you stop the dashboard
- 🚀 Zero Configuration - Works out of the box with built-in providers
# Setup in current directory
npx @pimzino/claude-code-spec-workflow
# Setup in specific directory
npx @pimzino/claude-code-spec-workflow --project /path/to/project
# Force overwrite existing files
npx @pimzino/claude-code-spec-workflow --force
# Skip confirmation prompts
npx @pimzino/claude-code-spec-workflow --yes
# Test the setup
npx @pimzino/claude-code-spec-workflow test
# Basic dashboard
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard
# Dashboard with tunnel (share externally)
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard --tunnel
# Full tunnel configuration
npx -p @pimzino/claude-code-spec-workflow claude-spec-dashboard \
--tunnel \
--tunnel-password mySecret123 \
--tunnel-provider cloudflare \
--port 3000 \
--open
Steering documents provide persistent project context that guides all spec development:
- Product vision and purpose
- Target users and their needs
- Key features and objectives
- Success metrics
- Technology stack and frameworks
- Development tools and practices
- Technical constraints and requirements
- Third-party integrations
- File organization patterns
- Naming conventions
- Import patterns
- Code organization principles
Run /spec-steering-setup
to create these documents. Claude will analyze your project and help you define these standards.
- Works out of the box with any project
- Auto-detects project type (Node.js, Python, Java, etc.)
- Validates Claude Code installation
- Beautiful CLI with progress indicators
- Confirmation prompts for safety
- Helpful error messages and guidance
- Complete workflow instructions in each command file
- Creates comprehensive directory structure
- Includes all necessary templates and configs
- Full TypeScript implementation with strict type checking
- Frontend converted to TypeScript for enhanced dashboard development
- 95%+ type coverage with no implicit any types
- Modern build pipeline with esbuild bundling and source maps
- Comprehensive error handling
- Follows npm best practices
- Persistent project context across all specs
- Automatic alignment with project standards
- Consistent code generation
- Reduced need for repetitive explanations
- Type-safe frontend code with comprehensive interfaces
- Real-time WebSocket communication with typed message handling
- Petite-vue integration with custom type definitions
- Build pipeline supporting development and production bundles
- Strict null checking and modern TypeScript patterns
- JSDoc documentation for all exported functions
your-project/
├── .claude/
│ ├── commands/ # 14 slash commands + auto-generated
│ ├── steering/ # product.md, tech.md, structure.md
│ ├── templates/ # Document templates
│ ├── specs/ # Generated specifications
│ ├── bugs/ # Bug fix workflows
│ └── agents/ # AI agents (enabled by default)
The package includes a built-in test command:
# Test setup in temporary directory
npx @pimzino/claude-code-spec-workflow test
- Node.js 16.0.0 or higher
- Claude Code installed and configured
- Any project directory
❓ Command not found after NPX
# Make sure you're using the correct package name
npx @pimzino/claude-code-spec-workflow
❓ Setup fails with permission errors
# Try with different directory permissions
npx @pimzino/claude-code-spec-workflow --project ~/my-project
❓ Claude Code not detected
# Install Claude Code first
npm install -g @anthropic-ai/claude-code
# Show verbose output
DEBUG=* npx @pimzino/claude-code-spec-workflow
# Check package version
npx @pimzino/claude-code-spec-workflow --version
cd my-awesome-project
npx @pimzino/claude-code-spec-workflow
claude
# Type: /spec-create user-dashboard "User profile management"
# Setup multiple projects
for dir in project1 project2 project3; do
npx @pimzino/claude-code-spec-workflow --project $dir --yes
done
The dashboard frontend is fully implemented in TypeScript for enhanced type safety and developer experience:
// Core dashboard types
interface Project {
path: string;
name: string;
level: number;
hasActiveSession: boolean;
specs: Spec[];
bugs: Bug[];
steeringStatus?: SteeringStatus;
}
// WebSocket message types with discriminated unions
type WebSocketMessage =
| { type: 'initial'; data: InitialData }
| { type: 'update'; data: UpdateData }
| { type: 'error'; data: ErrorData }
| { type: 'tunnel-status'; data: TunnelStatusData };
# TypeScript compilation and bundling
npm run build:frontend:dev # Build frontend for development
npm run build:frontend:prod # Build frontend for production (minified)
npm run watch:frontend # Watch mode with auto-rebuild
npm run typecheck:frontend # Type checking only (no output)
npm run typecheck # Check both backend and frontend types
# Development workflow
npm run dev:dashboard # Start dashboard with hot reload (frontend + backend)
npm run dev:dashboard:backend-only # Start only backend (for frontend debugging)
# Full build process
npm run build # Complete build: TypeScript + frontend + static files
npm run lint # Lint TypeScript files with auto-fix
npm run format # Format code with Prettier
- Strict TypeScript configuration with null checks
- Runtime type validation with type guards
- WebSocket message typing for real-time updates
- State management types for reactive UI components
- Error handling types with Result pattern
- Petite-vue integration with custom type definitions
// Import dashboard types
import type { Project, WebSocketMessage, AppState } from './dashboard.types';
// Type-safe project handling
function selectProject(project: Project): void {
console.log(`Selected: ${project.name} (${project.specs.length} specs)`);
// Safe property access with optional chaining
const steeringCount = project.steeringStatus?.totalDocs ?? 0;
if (steeringCount > 0) {
console.log(`Steering docs: ${steeringCount}`);
}
}
// WebSocket message handling with discriminated unions
function handleMessage(message: WebSocketMessage): void {
switch (message.type) {
case 'initial':
// TypeScript knows data is InitialData
console.log(`Loaded ${message.data.projects.length} projects`);
break;
case 'update':
// TypeScript knows data is UpdateData
console.log(`Updated project: ${message.data.projectPath}`);
break;
case 'error':
// TypeScript knows data is ErrorData
console.error(`Error: ${message.data.message}`);
break;
}
}
// Type guards for runtime validation
function isValidProject(obj: unknown): obj is Project {
return (
typeof obj === 'object' &&
obj !== null &&
'path' in obj &&
'name' in obj &&
'specs' in obj &&
Array.isArray((obj as Project).specs)
);
}
// Result type for error handling
type Result<T, E = Error> =
| { success: true; data: T }
| { success: false; error: E };
function parseProjectData(data: unknown): Result<Project> {
if (isValidProject(data)) {
return { success: true, data };
}
return { success: false, error: new Error('Invalid project data') };
}
- JSDoc documentation on all exported functions
- 95%+ type coverage maintained (no implicit any types)
- Modern TypeScript patterns (optional chaining, nullish coalescing)
- Type guards preferred over type assertions
- Interfaces for object shapes, union types for discriminated unions
- Result pattern for error handling
- Runtime validation with type guards for external data
- Full Documentation
- Tunnel Feature Guide - Comprehensive tunnel documentation
- Tunnel Examples - Ready-to-use tunnel scripts
- Claude Code Docs
- TypeScript API Reference - Generated from JSDoc comments
Contributions are welcome! Please see our Contributing Guide.
MIT License - see LICENSE for details.
See CHANGELOG.md for version history.
Scenario | Recommended Approach |
---|---|
New feature, well-defined | /spec-execute or individual task commands |
Complex/experimental feature | /spec-execute (manual control) |
Bug in existing code | Bug workflow (/bug-create → /bug-verify ) |
Learning the codebase | Manual execution with individual commands |
Production deployment | Full spec workflow with completion review |
# Install globally (recommended)
npm install -g @pimzino/claude-code-spec-workflow
# Verify installation
claude-code-spec-workflow --version
# Basic setup
claude-code-spec-workflow
# Advanced options
claude-code-spec-workflow --project /path --force --yes
During setup you choose:
- ✅ Enable agents? Enhanced automation vs simpler setup
- ✅ Project analysis Auto-detection of frameworks and patterns
Recommendation: Use Claude Opus 4 to generate the spec documentation '/spec-create', then use Claude Sonnet 4 for the implementation i.e. '/spec-execute' or '/{spec-name}-task-'.
Basic Workflow Example
# 1. Install globally (one time)
npm install -g @pimzino/claude-code-spec-workflow
# 2. Setup project (one time)
cd my-project
claude-code-spec-workflow
# 3. Create steering documents (recommended)
claude
/spec-steering-setup
# 4. Create feature spec
/spec-create user-authentication "Secure login system"
# 5. Execute tasks
/spec-execute 1 user-authentication
# 6. Monitor progress
/spec-status user-authentication
Bug Fix Example
/bug-create login-timeout "Users logged out too quickly"
/bug-analyze
/bug-fix
/bug-verify
The package includes optimized commands for efficient document loading across all document types:
Load all steering documents at once for context sharing:
claude-code-spec-workflow get-steering-context
Output: Formatted markdown with all steering documents (product.md, tech.md, structure.md)
Load all specification documents at once for context sharing:
claude-code-spec-workflow get-spec-context feature-name
Output: Formatted markdown with all spec documents (requirements.md, design.md, tasks.md)
Load templates by category for context sharing:
# Load all templates
claude-code-spec-workflow get-template-context
# Load specific template category
claude-code-spec-workflow get-template-context spec # Spec templates
claude-code-spec-workflow get-template-context bug # Bug templates
claude-code-spec-workflow get-template-context steering # Steering templates
Output: Formatted markdown with requested templates
- High-redundancy documents (steering, specs, templates): Use optimized bulk loading
- Low-redundancy documents (bug reports): Use direct file reading for simplicity
- Selective delegation: Main agents load full context, but pass only relevant portions to sub-agents
- Individual files: Continue using
get-content
for edge cases
- 60-80% token reduction compared to individual file loading
- Faster execution with cached context across all workflows
- Automatic fallback to individual
get-content
when needed - Session-based caching with intelligent file change detection
The system implements a sophisticated hierarchical context management strategy for maximum efficiency:
Main Agents (Commands like /spec-execute
, /spec-create
):
- Load ALL context once at the beginning using optimized commands
- Store complete context for task coordination and decision-making
- Distribute selective context to sub-agents without requiring reloads
Sub-Agents (Agents like spec-task-executor
):
- Priority 1: Use provided context from task instructions (steering, specification, task details)
- Priority 2: Fallback to loading context only if not provided above
- Never redundantly load context when it's already been provided by main agents
Context Distribution Pattern:
Main Agent loads: Steering + Full Spec + Task Details
↓ Delegates to Sub-Agent with:
├── Complete Steering Context
├── Selective Spec Context (Requirements + Design only)
├── Specific Task Details
└── Clear instruction: "Do NOT reload context"
This approach eliminates redundant loading while ensuring each agent has exactly the context it needs.
Common Issues
❓ "Command not found"
# Install globally first
npm install -g @pimzino/claude-code-spec-workflow
# Then use the command
claude-code-spec-workflow
❓ "Claude Code not detected"
npm install -g @anthropic-ai/claude-code
❓ "Permission errors"
claude-code-spec-workflow --project ~/my-project
- Node.js 16.0.0+
- Claude Code installed
- Any project directory
MIT License - LICENSE
Made with ❤️ by Pimzino
Special Thanks:
- @pimzino - Initial setup
- @boundless-oss - Steering documents
- @mquinnv - Dashboard feature
Powered by: Claude Code • Mermaid • TypeScript