A high-performance orchestration skill that spawns real AI sub-sessions in parallel to execute complex, multi-agent workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install parallel-agents
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 parallel-agents 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 Parallel Agents Skill?
The Parallel Agents Skill represents a significant evolution in AI orchestration, moving away from simulated templates to actual AI-powered sub-sessions. By utilizing the sessions_spawn tool, it creates isolated, fully functional AI instances that can think, reason, and utilize tools independently. This enables a robust ecosystem within Openclaw Skills where a single host can coordinate dozens of specialized agents simultaneously.
This skill is built specifically for the OpenClaw runtime, ensuring that each spawned agent has access to the same model capabilities and credentials as the host. Whether you are building a full-stack development team or a content factory, this skill provides the infrastructure for true parallel AI cognition and execution.
Parallel Agents Skill Use Cases
- Deploying a concurrent team of specialized developers and QA reviewers to architect and audit code in one pass.
- Generating large-scale content across multiple tones and platforms using a swarm of content writers.
- Bootstrapping new AI capabilities by using meta-agents to design, review, and refine new agent prompts.
- Conducting deep-dive research by spawning multiple agents to investigate different aspects of a topic simultaneously.
- Automating iterative quality improvement loops where reviewers and refiners collaborate to reach specific performance benchmarks.
How Parallel Agents Skill Works
- The orchestrator is invoked within an active OpenClaw session, gaining access to the sessions_spawn tool.
- User-defined tasks are mapped to specific agent types, such as frontend_developer or code_reviewer, from a registry of over 25 types.
- The skill utilizes a smart model hierarchy, attempting to use cost-effective models like Haiku before escalating to Kimi or Opus if tasks require more reasoning.
- Real AI sub-sessions are spawned in parallel, each running its own isolated logic and task description.
- The orchestrator monitors the status of these sessions, polling for completion and handling any necessary retries or escalations.
- Once finished, the orchestrator collects the JSON or Markdown output from the session history and compiles a comprehensive result set for the host.
Parallel Agents Skill Setup
The Parallel Agents Skill must be executed within an OpenClaw environment to access the necessary tool modules. It cannot function as a standalone Python script.
# Ensure the OpenClaw gateway is active
openclaw gateway start
# In your agent code, initialize the orchestrator
from ai_orchestrator import RealAIParallelOrchestrator, AgentTask
orch = RealAIParallelOrchestrator(max_concurrent=10)
Parallel Agents Skill Data Schema & Taxonomy
The skill manages data through structured objects to ensure consistency across the Openclaw Skills ecosystem.
| Object | Key | Type | Description |
|---|---|---|---|
| AgentTask | agent_type | String | The specialized agent profile to use for the task. |
| AgentTask | task_description | String | The core instructions and context for the spawned agent. |
| AgentResult | status | String | Current state of the agent: pending, running, completed, or failed. |
| AgentResult | output | Any | The generated payload, typically in JSON or Markdown format. |
| AgentResult | execution_time | Float | The total time taken for the sub-session to complete. |
Parallel Agents Skill Advanced Features
- Smart Model Hierarchy: Automatically optimizes costs by trying cheaper models first and escalating to high-reasoning models only when necessary.
- Autonomous Meta-Agent Hierarchy: Enables level 4 orchestration where agents design, review, and refine other agents without human intervention.
- Iterative Refinement Loops: Built-in workflow that cycles content through creators and reviewers until a target quality score (e.g., 8.5+) is achieved.
- Mass Concurrency Support: Proven capability to manage up to 20-50 simultaneous AI sessions depending on system resources.
- Production-Ready Auto-Retry: Includes helpers for spawning with retries and handling session timeouts to ensure high success rates in Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Lieutenant
An advanced security and trust verification layer designed to protect AI agents from prompt injection, jailbreaks, and malicious communications.
Pencil to Code
Transform Pencil .pen designs into production-ready React components and Tailwind CSS configurations automatically.
Pencil Renderer
A deterministic rendering engine that translates abstract aesthetic DNA codes into concrete Pencil .pen frames and visual screenshots.
PersonaNexus
A framework for defining structured AI agent identities using psychological models and compiling them into system prompts.
ClawStack
A high-speed, reputation-based Q&A platform where AI agents solve technical challenges and share knowledge autonomously.
UE5.7 Gamepiece Designer
A technical design architect for building multiplayer-ready Unreal Engine 5.7 systems, Blueprint recipes, and data structures.
Featured*
