Role Orchestrator

Role Orchestrator - Multi-Agent Coordination

Self-contained orchestration that works in ANY user project after specweave init.

Purpose

Coordinate multiple specialized agents for complex, multi-step tasks through intelligent decomposition and role assignment.

Architecture: Hierarchical Orchestrator-Worker Pattern

User Request → Orchestrator → PM → Architect → Tech Lead → Implement → QA → Deploy

When to Activate

Activates for requests requiring 3+ agents or full product development:

Agent Roles

Strategic Layer

PM Agent (pm-agent)

• Product strategy, user stories, prioritization
• When: Starting new products/features

Architect Agent (architect-agent)

• System design, ADRs, technology stack
• When: Designing systems or major features

Execution Layer

Tech Lead Agent (tech-lead-agent)

• Technical planning, code review, quality standards
• When: Complex technical decisions

Backend Agents

• nodejs-backend, python-backend, dotnet-backend
• When: Server-side implementation

Frontend Agent (frontend-agent)

• React/Next.js, UI/UX implementation
• When: Building user interfaces

Quality & Operations Layer

QA Lead Agent (qa-lead-agent)

• Test strategy, quality assurance, coverage
• When: Defining testing approach

Security Agent (security-agent)

• Security architecture, threat modeling, compliance
• When: Security-critical features

DevOps Agent (devops-agent)

• Infrastructure, CI/CD, deployment, monitoring
• When: Operations tasks

CRITICAL: Safe Orchestration Pattern

Rule: Orchestrator creates structure, guides user to invoke agents in MAIN context (NOT nested spawning).

Phase 0: Create Increment Structure FIRST

Before invoking ANY agents, create increment folder:

// 1. Parse user request
const projectName = extractProjectName(userRequest);
// "event management" → "event-management"

// 2. Get next number
const nextNumber = getNextIncrementNumber();
// e.g., 0001, 0002, 0003

// 3. Create structure
const incrementPath = `.specweave/increments/${nextNumber}-${projectName}/`;
mkdir -p ${incrementPath}
mkdir -p ${incrementPath}logs/
mkdir -p ${incrementPath}scripts/
mkdir -p ${incrementPath}reports/

// 4. Create placeholder files (ORDER MATTERS!)
// metadata.json MUST be created FIRST (metadata-json-guard.sh blocks spec.md otherwise)
write ${incrementPath}metadata.json (MANDATORY - CREATE FIRST!)
write ${incrementPath}spec.md (basic template)
write ${incrementPath}plan.md (basic template)
write ${incrementPath}tasks.md (basic template)

metadata.json template (CREATE FIRST!):

{
  "id": "0001-project-name",
  "status": "planned",
  "type": "feature",
  "priority": "P1",
  "created": "2025-11-24T12:00:00Z",
  "lastActivity": "2025-11-24T12:00:00Z"
}

spec.md template (create AFTER metadata.json):

---
increment: 0001-project-name
title: "Project Name"
type: feature
priority: P1
status: planned
created: 2025-11-24
---

# Project Name

## Overview
(To be filled by PM Agent)

## User Stories
(To be filled by PM Agent)

Phase 1: Guide User Through Agent Workflow

Output this workflow to user:

✅ Increment structure created: .specweave/increments/0001-project-name/

🎯 Complete workflow (run these commands in MAIN conversation):

STEP 1: Product Strategy & Requirements
Tell Claude: "Complete the spec for increment 0001-project-name"
(PM agent will activate automatically)

STEP 2: Architecture & Design
Tell Claude: "Design architecture for increment 0001-project-name"
(Architect agent will create ADRs and system design)

STEP 3: Technical Planning
Tell Claude: "Create technical plan for increment 0001-project-name"
(Tech Lead agent will create implementation approach)

STEP 4: Implementation Tasks
Tell Claude: "Create tasks for increment 0001-project-name"
(Test-aware planner will generate tasks with tests)

STEP 5: Security Review (if needed)
Tell Claude: "Review security for increment 0001-project-name"
(Security agent will perform threat modeling)

STEP 6: Implementation
Tell Claude: "Implement increment 0001-project-name"
(Backend/Frontend agents will implement code)

STEP 7: Quality Assurance
Tell Claude: "Run QA for increment 0001-project-name"
(QA agent will verify tests and coverage)

STEP 8: Deployment Planning
Tell Claude: "Plan deployment for increment 0001-project-name"
(DevOps agent will create infrastructure)

⚠️  Run these sequentially in MAIN conversation to prevent context explosion!

DO NOT spawn all agents from this skill using Task() tool!

Orchestration Patterns

Pattern 1: Sequential (Default)

When: Dependencies between steps

PM → Architect → Tech Lead → Backend → Frontend → QA → DevOps

User workflow:

1. Create increment structure
2. Guide user to invoke agents one by one
3. Each agent completes before next starts
4. User tracks progress

Pattern 2: Parallel Execution

When: Independent work streams

PM + Architect (parallel)
    ↓
Backend + Frontend (parallel)
    ↓
QA + DevOps (parallel)

User workflow:

1. Create increment structure
2. Identify parallel opportunities
3. Guide user: "You can run these in parallel: [list]"
4. User invokes agents concurrently

Pattern 3: Adaptive (Context-Aware)

When: Requirements discovered during execution

PM → Architect → [Discover need] → Security → Tech Lead → ...

User workflow:

1. Start with basic plan
2. Agent discovers additional needs
3. Inject new agents mid-workflow
4. Adjust plan dynamically

Quality Gates (Checkpoints)

Gate 1: After PM (Requirements Complete)

Check:

• User stories defined with AC-IDs
• Success criteria measurable
• Dependencies identified
• Out of scope defined

Decision: Proceed to architecture OR refine requirements

Gate 2: After Architect (Design Complete)

Check:

• System design documented
• ADRs created (≥3)
• Technology stack chosen
• Data model defined

Decision: Proceed to implementation OR redesign

Gate 3: After Implementation (Code Complete)

Check:

• All P1 tasks completed
• Tests passing (≥80% coverage)
• Code reviewed
• Documentation updated

Decision: Proceed to deployment OR fix issues

Gate 4: Before Deployment (Production Ready)

Check:

• Security review passed
• Performance acceptable
• Monitoring configured
• Rollback plan exists

Decision: Deploy OR address blockers

Error Handling

Agent Failure

If agent fails:
1. Capture error details
2. Retry with backoff (max 3 attempts)
3. Try fallback agent if available
4. If critical: pause and notify user
5. NEVER fail silently

Graceful Degradation

If PM fails → Use basic template
If Architect fails → Tech Lead creates simple design
If Backend fails → Create manual implementation guide
If DevOps fails → Create manual deployment steps

User Intervention

If orchestrator stuck:
1. Show current state
2. Present options to user
3. Let user decide next step
4. Resume from checkpoint

Progress Tracking

Real-Time Updates

Show user current phase and progress:

🚀 Building SaaS for Event Management

Progress: 3/8 phases complete (37.5%)

✅ Phase 1: Requirements (PM) - COMPLETE
✅ Phase 2: Architecture (Architect) - COMPLETE
✅ Phase 3: Technical Planning (Tech Lead) - COMPLETE
🔄 Phase 4: Implementation (Backend + Frontend) - IN PROGRESS
⏳ Phase 5: QA - PENDING
⏳ Phase 6: Security Review - PENDING
⏳ Phase 7: DevOps - PENDING
⏳ Phase 8: Deployment - PENDING

Current: Backend implementing user authentication (T-015/45)

Example Workflows

Example 1: Build SaaS Product

User request: "Build a SaaS for event management"

Orchestrator response:

🎯 Creating increment for Event Management SaaS

Created: .specweave/increments/0001-event-management/

📋 8-Phase Workflow:

Phase 1: Requirements & Strategy
Tell Claude: "Complete spec for increment 0001"
→ PM will define user stories, features, success criteria

Phase 2: System Architecture
Tell Claude: "Design architecture for increment 0001"
→ Architect will create system design and ADRs

Phase 3: Technical Planning
Tell Claude: "Create technical plan for increment 0001"
→ Tech Lead will plan implementation approach

Phase 4: Test Strategy
Tell Claude: "Create test strategy for increment 0001"
→ QA Lead will define testing approach

Phase 5: Backend Implementation
Tell Claude: "Implement backend for increment 0001"
→ Backend agent will build APIs and services

Phase 6: Frontend Implementation
Tell Claude: "Implement frontend for increment 0001"
→ Frontend agent will build UI

Phase 7: Security Review
Tell Claude: "Review security for increment 0001"
→ Security agent will validate security

Phase 8: Deployment Setup
Tell Claude: "Setup deployment for increment 0001"
→ DevOps agent will create infrastructure

Start with Phase 1!

Example 2: Critical Hotfix

User request: "Fix critical security vulnerability CVE-2024-1234"

Orchestrator response:

🚨 CRITICAL HOTFIX WORKFLOW

Created: .specweave/increments/0016-security-fix-cve-2024-1234/
Type: hotfix (expedited workflow)

⚡ 4-Phase Fast Track:

Phase 1: Security Assessment (5 min)
Tell Claude: "Assess security vulnerability for increment 0016"

Phase 2: Fix Implementation (15 min)
Tell Claude: "Implement security fix for increment 0016"

Phase 3: Validation (10 min)
Tell Claude: "Validate security fix for increment 0016"

Phase 4: Emergency Deploy (5 min)
Tell Claude: "Deploy hotfix for increment 0016"

Total estimated time: 35 minutes

Best Practices

✅ DO:

• Create increment structure FIRST
• Guide user through sequential workflow
• Use quality gates between phases
• Track progress and show user updates
• Handle errors gracefully with retry
• Allow user to skip optional phases

❌ DON'T:

• Spawn all agents at once (context explosion)
• Skip increment creation
• Proceed without quality gate checks
• Fail silently on agent errors
• Assume agents are infallible
• Block user from manual intervention

Integration with SpecWeave Commands

After orchestration complete:

# Check status
/sw:status

# Sync to external tools
/sw:sync-progress 0001

# Validate quality
/sw:qa 0001

# Close increment
/sw:done 0001

This skill is self-contained and works in ANY user project after specweave init.

Project-Specific Learnings

Before starting work, check for project-specific learnings:

# Check if skill memory exists for this skill
cat .specweave/skill-memories/role-orchestrator.md 2>/dev/null || echo "No project learnings yet"

Project learnings are automatically captured by the reflection system when corrections or patterns are identified during development. These learnings help you understand project-specific conventions and past decisions.