Project Orchestration Skill

Purpose: Orchestrate multi-agent workflows for feature development

What I Do

Coordinate multi-agent planning and execution workflows:

• Pass minimal context between planning agents
• Track planning agent outputs
• Manage 8-stage feature delivery workflow
• Maintain planning session state

Three Workflows

1. Context Handoff (Lightweight)

Pass minimal context between agents (RECOMMENDED for automation)

• Commands: create, get-context, add-output, show
• Use when: Orchestrating planning agents
• See: workflows/context-handoff.md

2. Session Context (Interactive)

Human-readable planning narrative (for debugging/review)

• Commands: session-create, session-load, session-summary
• Use when: Debugging or reviewing planning sessions
• See: workflows/context-handoff.md

3. Multi-Stage Delivery (8 Stages)

Full feature delivery workflow

• Commands: stage-init, stage-status, stage-complete, stage-rollback
• Use when: Running complete 8-stage feature workflow
• See: workflows/8-stage-delivery.md

Quick Start

Context Handoff Workflow

# Create context index
bash .opencode/skill/project-orchestration/router.sh create auth-system

# Get minimal context for specific agent
bash .opencode/skill/project-orchestration/router.sh get-context auth-system StoryMapper

# Track agent output
bash .opencode/skill/project-orchestration/router.sh add-output auth-system ArchitectureAnalyzer .tmp/architecture/auth-system/contexts.json

# Show full index
bash .opencode/skill/project-orchestration/router.sh show auth-system

8-Stage Workflow

# Initialize workflow
bash .opencode/skill/project-orchestration/router.sh stage-init auth-system

# Check status
bash .opencode/skill/project-orchestration/router.sh stage-status auth-system

# Mark stage complete
bash .opencode/skill/project-orchestration/router.sh stage-complete auth-system 1

# Rollback if needed
bash .opencode/skill/project-orchestration/router.sh stage-rollback auth-system 1

Command Reference

Context Management

Command	Description
create <feature>	Create context index for feature
get-context <feature> <agentType>	Get minimal context for specific agent
add-output <feature> <agent> <path>	Track agent output
show <feature>	Show full context index

Session Management

Command	Description
session-create <feature> <request>	Create session context
session-load <sessionId>	Load session context
session-summary <sessionId>	Get session summary

Stage Management

Command	Description
stage-init <feature>	Initialize 8-stage workflow
stage-status <feature>	Show stage progress
stage-complete <feature> <stage>	Mark stage complete
stage-rollback <feature> <stage>	Rollback stage
stage-validate <feature> <stage>	Validate stage
stage-abort <feature>	Abort workflow

Architecture

.opencode/skill/project-orchestration/
├── SKILL.md                          # This file
├── router.sh                         # CLI router
├── scripts/
│   ├── context-index.ts              # Lightweight context handoff
│   ├── session-context-manager.ts    # Session context management
│   └── stage-cli.ts                  # 8-stage workflow management
└── workflows/
    ├── context-handoff.md            # Detailed context handoff guide
    ├── 8-stage-delivery.md           # Detailed 8-stage workflow guide
    └── planning-agents.md            # Planning agents integration guide

File Locations

Scripts

• Context Index: .opencode/skill/project-orchestration/scripts/context-index.ts
• Session Context: .opencode/skill/project-orchestration/scripts/session-context-manager.ts
• Stage CLI: .opencode/skill/project-orchestration/scripts/stage-cli.ts

Runtime Files

• Context Index: .tmp/context-index/{feature}.json
• Sessions: .tmp/sessions/{session-id}/context.md
• Stage Tracking: .tmp/sessions/{session-id}/stage-tracking.json

Workflow Guides

• Context Handoff: .opencode/skill/project-orchestration/workflows/context-handoff.md
• 8-Stage Delivery: .opencode/skill/project-orchestration/workflows/8-stage-delivery.md
• Planning Agents: .opencode/skill/project-orchestration/workflows/planning-agents.md

Context Handoff Pattern

Lightweight Context Index (RECOMMENDED)

Use when: Multi-agent pipelines, performance-critical workflows, automated orchestration

How it works:

• Orchestrator maintains lightweight index (.tmp/context-index/{feature}.json)
• Each agent gets ONLY the files they need
• 83% reduction in context per agent

Example:

import { createContextIndex, getContextForAgent, addAgentOutput } from './scripts/context-index';

// Create index
createContextIndex('auth-system', { request: 'Build JWT auth' });

// Get minimal context for StoryMapper
const context = getContextForAgent('auth-system', 'StoryMapper');
// Returns: { contextFiles: [...], agentOutputs: { ArchitectureAnalyzer: "..." } }

// After agent completes, update index
addAgentOutput('auth-system', 'StoryMapper', '.tmp/story-maps/auth-system/map.json', 
  { verticalSlice: 'user-login' });

Benefits:

• ✅ Minimal context per agent (only what they need)
• ✅ Fast (small JSON files)
• ✅ Scalable (index stays small)
• ✅ Clear dependencies (explicit file paths)

Session Context (Interactive)

Use when: Interactive sessions, human review, debugging, narrative tracking

How it works:

• Creates markdown file (.tmp/sessions/{session-id}/context.md)
• Human-readable narrative of entire workflow
• All agents read/update same file

Example:

import { createSession, updateSession, markStageComplete } from './scripts/session-context-manager';

// Create session
const sessionId = await createSession('auth-system', {
  request: 'Build JWT authentication',
  contextFiles: ['.opencode/context/core/standards/code-quality.md'],
  exitCriteria: ['All endpoints implemented']
});

// Update after agent completes
await updateSession(sessionId, {
  architectureContext: { boundedContext: 'authentication' }
});

await markStageComplete(sessionId, 'Stage 1: Architecture Decomposition', [
  '.tmp/architecture/auth-system/contexts.json'
]);

Benefits:

• ✅ Human-readable (markdown format)
• ✅ Complete narrative (see entire workflow)
• ✅ Good for debugging (trace decisions)
• ✅ Audit trail (who did what when)

Comparison

Feature	Context Index	Session Context
Format	JSON	Markdown
Size	Small (~200 lines)	Large (2000+ lines)
Audience	Agents only	Humans + Agents
Context per agent	Minimal (2-5 files)	Everything (all files)
Performance	Fast	Slower
Use case	Automated pipelines	Interactive sessions
Human readable	No	Yes
Debugging	Harder	Easier

Recommendation

For most multi-agent workflows: Use Context Index (lightweight, fast, minimal context per agent)

For interactive development: Use Session Context (human-readable, complete narrative)

You can use both: Context Index for automation, Session Context for human review

8-Stage Delivery Workflow

The Multi-Stage Orchestration Workflow provides a systematic approach to complex feature development:

Stages

1. Architecture Decomposition - Define system boundaries and components
2. Story Mapping - Map user journeys and create stories
3. Prioritization - Sequence work by value and dependencies
4. Enhanced Task Breakdown - Create atomic, executable tasks
5. Contract Definition - Define interfaces before implementation
6. Parallel Execution - Execute independent work simultaneously
7. Integration & Validation - Integrate and validate components
8. Release & Learning - Deploy and capture insights

Stage Management

# Initialize workflow
bash .opencode/skill/project-orchestration/router.sh stage-init auth-system

# Check status
bash .opencode/skill/project-orchestration/router.sh stage-status auth-system

# Validate stage readiness
bash .opencode/skill/project-orchestration/router.sh stage-validate auth-system 1

# Mark stage complete
bash .opencode/skill/project-orchestration/router.sh stage-complete auth-system 1

# Rollback if needed
bash .opencode/skill/project-orchestration/router.sh stage-rollback auth-system 1

# Abort workflow
bash .opencode/skill/project-orchestration/router.sh stage-abort auth-system

Stage Tracking

Stage tracking files are stored in .tmp/sessions/{timestamp}-{feature}/stage-tracking.json

Each stage tracks:

• Status (pending, in_progress, completed, failed)
• Start and completion timestamps
• Outputs produced
• Validation results
• Error messages (if failed)

Planning Agents Integration

The orchestration skill integrates with planning agents to enrich task metadata:

ArchitectureAnalyzer

Provides bounded_context and module fields based on domain analysis.

StoryMapper

Provides vertical_slice field for feature slice architecture.

PrioritizationEngine

Provides rice_score, wsjf_score, and release_slice fields for prioritization.

ContractManager

Provides contracts array for API/interface dependencies.

ADRManager

Provides related_adrs array for architectural decision references.

See workflows/planning-agents.md for detailed integration guide.

Best Practices

Context Handoff

• Use Context Index for automated pipelines
• Use Session Context for human review
• Update index immediately after agent completes
• Keep metadata small and useful

Stage Management

• Validate stage prerequisites before execution
• Mark stages complete only when all criteria met
• Document rollback reasons
• Track stage outputs for traceability

Planning Integration

• Call ArchitectureAnalyzer for complex domain logic
• Use StoryMapper for user journey mapping
• Apply PrioritizationEngine for MVP identification
• Define contracts with ContractManager for parallel dev
• Document decisions with ADRManager

See Also

• Task Management: .opencode/skill/task-management/SKILL.md
• Planning Agents Guide: .opencode/docs/agents/planning-agents-guide.md
• Multi-Stage Orchestration: .opencode/context/core/workflows/multi-stage-orchestration.md