This skill should be used when the user asks about "orchestration patterns", "plan-then-execute", "hierarchical decomposition", "blackboard pattern", "event sourcing pattern", "which pattern to use", "parallel execution strategies", or needs to select an orchestration approach for complex multi-agent tasks. Provides comprehensive guidance on 4 orchestration patterns for coordinating multiple agents.
This skill inherits all available tools. When active, it can use any tool Claude has access to.
Establish the appropriate multi-agent coordination strategy by selecting from four proven orchestration patterns based on task characteristics and requirements.
Choose the orchestration pattern based on task complexity and coordination needs:
| Pattern | Best For | Parallelism | State Management |
|---|---|---|---|
| Plan-then-Execute | Well-defined tasks with clear steps | Level-based | Centralized |
| Hierarchical Decomposition | Complex objectives requiring breakdown | Bottom-up aggregation | Tree-structured |
| Blackboard | Collaborative problem-solving | Concurrent contributions | Shared knowledge space |
| Event Sourcing | Audit trails and replay capability | Event-driven | Append-only log |
The most common pattern for structured, multi-phase work. Generate a comprehensive plan, validate it, then execute systematically.
Phase 1: EXPLORE (2+ agents) → Analysis, research, context gathering
Phase 2: PLAN (1-2 agents) → Strategy, architecture, task breakdown
Phase 3: CODE (2-4 agents) → Implementation, parallel development
Phase 4: TEST (2-3 agents) → Unit, integration, E2E testing
Phase 5: FIX (1-2 agents) → Bug fixes, refinements
Phase 6: DOCUMENT (1-2 agents) → Documentation, knowledge transfer
Recursively break down complex objectives into atomic, executable tasks with clear ownership.
Shared knowledge space where multiple agents contribute specialized knowledge collaboratively.
Blackboard State:
├── problem_description: string
├── hypotheses: Hypothesis[]
├── partial_solutions: Solution[]
├── constraints: Constraint[]
└── contributions: Contribution[]
Event-driven task coordination with complete audit trail and replay capability.
interface OrchestrationEvent {
id: string;
type: 'AgentSpawned' | 'PhaseTransition' | 'Checkpoint' | 'Error' | 'Recovery';
timestamp: number;
agentId?: string;
payload: any;
}
Combine patterns for complex scenarios:
Use Plan-then-Execute at the top level with Hierarchical Decomposition for implementation phases.
Shared knowledge space with complete audit trail of contributions.
Map agents to appropriate layers based on pattern:
| Layer | P-t-E Role | Hierarchical Role | Blackboard Role |
|---|---|---|---|
| Strategic | Plan generation | Root decomposition | Problem framing |
| Tactical | DAG scheduling | Level coordination | Contribution selection |
| Operational | Task execution | Leaf implementation | Knowledge contribution |
| Quality | Validation gates | Aggregation verification | Solution validation |
references/pattern-details.md - Detailed implementation guides for each patternreferences/pattern-selection.md - Decision tree for pattern selectionexamples/plan-execute-dag.json - Sample DAG for P-t-E patternexamples/hierarchical-tree.json - Sample decomposition tree