agent-orchestration

What This Skill Does

Agent Orchestration is the flagship skill of the Hal Stack 🦞. It transforms your OpenClaw agent from a simple chatbot into a project manager. The skill provides a rigorous framework for spawning, tracking, and optimizing sub-agents. Instead of hoping for good results, it forces a contract-based approach to prompting, ensuring that every AI task follows the 5-Layer Architecture (Identity, Context, Task, Constraints, and Output). It tracks running processes, allows for iterative feedback loops, and converts raw intent into machine-executable instructions.

Installation

To add this skill to your workspace, run the following command in your terminal:

clawhub install openclaw/skills/skills/halthelobster/agent-orchestration

Ensure your OpenClaw environment is updated to the latest version to support concurrent agent tracking.

Use Cases

• Complex Project Management: Break down a massive technical migration into smaller, independent agent-driven tasks.
• Automated Research Pipelines: Spin up specialized sub-agents to gather data, synthesize findings, and write reports.
• Refining Prompt Quality: Use the skill's built-in analysis to identify why specific outputs are failing and iterate on the system prompt structure.
• Quality Assurance: Create 'Auditor' sub-agents that review the work of 'Creator' agents against a set of constraints.

Example Prompts

1. "Orchestrate a sub-agent to scrape the documentation for the latest React Router updates and summarize the breaking changes as a bulleted list."
2. "I need a research assistant to compare the pricing models of three different cloud providers. Track the process and log the status of each sub-agent."
3. "Evaluate my previous prompt for the documentation writer. Identify which of the 5-layer architecture components are missing and propose an improved version."

Tips & Limitations

• Be Specific: The quality of output is strictly tied to the precision of your contract. If your prompt lacks a defined 'Output' structure, expect the sub-agent to default to conversational summaries.
• Avoid Over-Spawning: Monitor your agent count. Launching too many concurrent agents can exceed context window limits or lead to confusing logging in the agent-orchestration dashboard.
• Resource Management: Remember that sub-agents consume the same token budget as the primary agent. Keep tasks focused to maintain efficiency and cost-effectiveness.