swarm-commander

name	swarm-commander
description	Orchestrate 10-50 expert agents in parallel-style execution with intelligent task decomposition, file-based coordination, and unified synthesis

Swarm Commander

Transform from single-agent interaction to commanding an expert army. You become the CEO reviewing synthesized outputs while specialized agents execute in orchestrated swarms.

Overview

Swarm Commander is the Tier 1.5 implementation of agent swarm orchestration—achieving 80-95% of true parallelism benefits using sophisticated prompt engineering, file-based coordination, and the Agent-as-Tool pattern. Built on three foundational methodologies:

Source Expert	Pattern Borrowed
Boris	Batch processing, work orders, multi-instance orchestration
Lance & Yichao	Agent-as-Tool isolation, file system coordination, context engineering
Mark Kashef	Intelligent agent selection, tension design, deliberation protocols

Core Insight: True parallel execution isn't required when you architect minimal-context work orders and file-based aggregation. The speed gains come from eliminating redundant context loading, not from simultaneous API calls.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    YOUR REQUEST                                  │
└───────────────────────────┬─────────────────────────────────────┘
                            ▼
┌─────────────────────────────────────────────────────────────────┐
│         PHASE 1: SWARM PLANNING                                  │
│  • Decompose task into parallelizable work units                │
│  • Select optimal agents from 44+ expert registry               │
│  • Map dependency graph (parallel vs sequential)                │
│  • Generate work orders with rigid output schemas               │
│  → Output: execution_plan.md + work_orders/                     │
└───────────────────────────┬─────────────────────────────────────┘
                            ▼
┌─────────────────────────────────────────────────────────────────┐
│         PHASE 2: SWARM EXECUTION                                 │
│  • Each agent receives ONLY their work order (isolated context) │
│  • Agent embodies expert persona + methodology                  │
│  • Output written to structured file (not conversation)         │
│  • Parallel-compatible agents batched together                  │
│  → Output: agent_outputs/[agent_name].md                        │
└───────────────────────────┬─────────────────────────────────────┘
                            ▼
┌─────────────────────────────────────────────────────────────────┐
│         PHASE 3: SWARM SYNTHESIS                                 │
│  • Fresh context reads ONLY file outputs                        │
│  • Identifies agreements, conflicts, and minority positions     │
│  • Applies quality verification from source experts             │
│  • Produces unified deliverable with provenance                 │
│  → Output: final_synthesis.md                                   │
└─────────────────────────────────────────────────────────────────┘

Deployable Prompts

Prompt	Use When
Swarm Planning	Decompose any task into a multi-agent execution plan
Agent Constellation	Select optimal experts based on task requirements
Batch Execution	Execute work orders with Agent-as-Tool isolation
Swarm Synthesis	Aggregate outputs into unified deliverable

Key Principles

1. Work Orders > Conversations: Agents receive minimal context + schema, not full history
2. Files > Messages: Coordination happens through file system, not context passing
3. Batching > Looping: Conceptually parallel agents execute in optimized batches
4. Schemas > Prose: Rigid output formats enable fast synthesis
5. Provenance > Anonymity: Track which expert said what for accountability

Swarm Sizes

Size	Agents	Best For
Squad	3-5	Focused tasks, quick turnaround
Team	6-12	Comprehensive analysis, multi-perspective
Platoon	13-25	Large-scale research, complex projects
Army	26-50	Enterprise initiatives, exhaustive coverage

Quick Start

1. Describe your objective to Swarm Commander
2. Review the proposed agent constellation
3. Approve the execution plan
4. Receive synthesized output with full provenance
5. Iterate on specific agent outputs if needed

Example Use Cases

• Competitive Analysis: 10 research agents analyze different competitors in parallel
• Content Campaign: Writers, SEO, design, psychology experts contribute simultaneously
• Product Launch: Market research, pricing, positioning, copy experts coordinate
• Strategic Decision: Multiple domain experts deliberate with structured synthesis

Resources

• Genius Patterns - Core patterns from Boris, Lance/Yichao, Kashef
• Expert Registry - Reference to available 44+ agents

Execution Prompts (structure-pure v2)

4 deterministic practitioner prompts — each carries an Output Contract, Output Skeleton, and Quality Gate. When a deliverable matches one, Read it and honor its contract instead of improvising the output shape.

• Agent Constellation — skills/swarm-commander/references/prompts-v2/agent-constellation.md
• Batch Execution — skills/swarm-commander/references/prompts-v2/batch-execution.md
• Swarm Planning — skills/swarm-commander/references/prompts-v2/swarm-planning.md
• Swarm Synthesis — skills/swarm-commander/references/prompts-v2/swarm-synthesis.md