Parallel Agents

What This Skill Does

The Parallel Agents skill by jdalbright elevates OpenClaw to a multi-agent orchestration platform. Unlike legacy versions that simulated concurrency via templates or local scripts, this skill utilizes the native sessions_spawn tool to create independent, live OpenClaw agent sessions. It enables users to offload complex, multi-stage workflows to a hierarchy of specialized AI agents, ranging from individual task-focused workers to meta-agents that manage, review, and refine complex outputs autonomously. It supports up to 20 concurrent sessions and implements an intelligent model hierarchy (Haiku, Kimi, Opus) to balance performance and cost.

Installation

To integrate this skill, run the following command in your terminal within the OpenClaw environment: clawhub install openclaw/skills/skills/jdalbright/parallel-agents

Ensure your OpenClaw gateway is active and the sessions module is authorized for the environment, as this skill relies on the system's ability to fork new agent sessions dynamically.

Use Cases

• Large-Scale Content Generation: Spin up five research agents to gather data on a topic, then a 'Meta Agent' to synthesize the findings into a single report.
• Automated Quality Assurance: Utilize the iterative refinement loop where a 'Creator' agent drafts code or text, a 'Reviewer' flags issues, and a 'Refiner' polishes the final output.
• Complex Task Delegation: Break down massive user requests into parallelized sub-tasks, such as concurrently searching for market trends, analyzing competitor pricing, and drafting a strategic memo.
• Multi-Model Testing: Execute the same prompt across different model tiers (Haiku vs. Opus) simultaneously to compare output quality for specific, specialized domains.

Example Prompts

1. "Spawn a team of 3 agents to research the latest advancements in quantum computing, have a fourth agent summarize the results, and save the final report to a text file."
2. "Use the iterative refinement loop: create a technical document about server security, have a reviewer critique it for vulnerability gaps, and then refine the draft accordingly."
3. "Launch an Orchestrator agent to plan a marketing strategy for my new product and manage the 3 specialized agents needed to execute the campaign steps."

Tips & Limitations

• Orchestration Context: Always invoke this skill from within an active OpenClaw session. Attempting to run it as a standalone script via exec will fail because it requires access to the internal tools registry.
• Resource Management: While the system supports 20 concurrent agents, monitor your API limits, as each spawned agent consumes tokens based on the model tier selected.
• Clean-up: Use the cleanup parameter in your sessions_spawn calls to ensure inactive sessions are deleted, keeping your environment clean and memory usage stable.