A sophisticated task orchestration framework that transforms complex requirements into coordinated actions across a swarm of specialized AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-swarm-ex
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-swarm-ex using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agent Swarm Orchestrator?
Agent Swarm is a high-level orchestration skill designed for Openclaw Skills users who need to manage multi-agent workflows. It allows a lead agent to act as a commander, delegating sub-tasks to a specialized team of autonomous agents—including project managers, researchers, and coders—to solve complex problems that a single model cannot handle efficiently. By leveraging task decomposition and parallel execution, it significantly improves speed and output quality.
This skill is particularly effective for large-scale projects requiring cross-functional collaboration, such as technical research reports, software module refactoring, or automated data analysis. It optimizes operational costs by assigning specific models to specific roles, ensuring that expensive high-reasoning models are only used for the most difficult coordination tasks while efficient mini-models handle routine data fetching or communication.
Agent Swarm Orchestrator Use Cases
- Complex technical research requiring deep web searching and structured synthesis.
- Software engineering projects needing modular decomposition, coding, and peer review.
- Data analysis workflows involving cleaning, statistical processing, and visualization.
- Content creation pipelines that include planning, drafting, and editorial review.
- Automated daily monitoring of technical trends via GitHub or web scrapers.
How Agent Swarm Orchestrator Works
- The lead agent analyzes the incoming request to determine its complexity and whether it requires swarm orchestration.
- Complex tasks are decomposed into independent sub-tasks with defined inputs, outputs, and dependencies.
- Specialized agents (e.g., researcher, coder, or writer) are selected based on the nature of each sub-task.
- Tasks are dispatched via the sessions_spawn command, allowing for parallel or serial execution depending on the workflow.
- The lead agent collects results from all spawned agents, deduplicates information, and performs a final synthesis.
- A comprehensive execution report is generated, including token usage, cost analysis, and efficiency statistics.
Agent Swarm Orchestrator Setup
To get started with this skill within the Openclaw Skills ecosystem, initialize your workspace using the provided scripts:
# Initialize the agent workspace and directory structure
python3 scripts/init_agents.py --base-path /workspace/agents
# List available agents in the swarm
python3 scripts/agent_manager.py list
# (Optional) Create a custom agent from a template
python3 scripts/agent_manager.py add my_custom_agent --template researcher --name "Expert Searcher"
Agent Swarm Orchestrator Data Schema & Taxonomy
The skill maintains a structured workspace for agent personas and persistent memory. Data is organized as follows:
| Path | Description |
|---|---|
/workspace/agents/ |
Root directory for all swarm agents defined in Openclaw Skills. |
.../<agent_id>/SOUL.md |
Core persona and behavioral definitions for the specific agent. |
.../<agent_id>/AGENTS.md |
Specific operating procedures and task execution standards. |
.../<agent_id>/memory/ |
Directory containing experience.md and experience.json for long-term learning. |
/workspace/research/ |
Default output directory for research-focused sub-tasks. |
Agent Swarm Orchestrator Advanced Features
- Asynchronous Task Spawning: Initiate multiple agent sessions simultaneously for massive throughput.
- Cost-Aware Routing: Automatically assign different LLM models (e.g., Claude Opus vs GPT-4 Mini) based on agent role difficulty to save resources.
- Autonomous Experience Logging: Agents summarize their own learnings after tasks to improve future performance automatically.
- Statistics Reporting: Detailed breakdowns of token consumption, time saved via parallelism, and actual cost savings.
- Dynamic Persona Management: CLI tools to add, remove, or update agent identities and capabilities on the fly.
SKILL.md
Loading
Related Openclaw Skills
Polymarket Pro
A professional-grade CLI integration for trading, researching, and managing Polymarket prediction market positions with granular control.
mcporter
A comprehensive CLI utility for managing, authenticating, and interacting with Model Context Protocol (MCP) servers and tools.
Social Work Case Record Generator
Converts LINE chats and oral descriptions into structured social work case records with professional intervention plans.
Calorie Lookup (Spoonacular + USDA)
A high-performance nutrition lookup skill for Openclaw Skills that provides calorie estimation and macro-nutrient data using multiple API sources.
Kid Tutor
An AI-powered Socratic tutor for children aged 6-12 focusing on math and science through interactive guidance.
Agent Swarm
A sophisticated orchestration framework that coordinates a team of specialized AI agents to execute complex, multi-stage tasks through parallel and serial workflows.
Featured*
