agent-orchestration

What This Skill Does

Agent Orchestration is the mission-critical component of the Hal Stack designed to move you from passive prompting to active system design. It provides a structured framework for spawning, managing, and optimizing sub-agents. By implementing the 5-Layer Architecture (Identity, Context, Task, Constraints, and Output), this skill forces you to define agent behavior as a formal contract, drastically reducing hallucinations and generic output. It acts as an orchestrator, enabling you to track the execution paths of sub-agents and iteratively refine their logic based on real-world outcomes.

Installation

To integrate Agent Orchestration into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/clawdnw/agent-orchestration

Use Cases

• Automated Research Pipelines: Spin up specialized researcher agents that follow rigorous citation protocols.
• Content Engineering: Orchestrate a hierarchy of agents where one agent drafts content, a second critiques it based on SEO requirements, and a third finalizes the output.
• Code Review & Refactoring: Deploy sub-agents to analyze specific codebase modules, adhering to strictly defined style guides and security constraints.
• System Monitoring: Manage agents that periodically verify the health of your services and escalate issues based on pre-defined severity levels.

Example Prompts

1. "Orchestrate a sub-agent to analyze the last 50 GitHub issues in my repo and summarize the top three technical blockers as a JSON file."
2. "I need a research agent created with the persona of a senior cybersecurity auditor. Task it with reviewing my current security policy and finding compliance gaps relative to SOC2."
3. "Track the current sub-agents. I want to see the task, current status, and reasoning logs for the agent tasked with landing page optimization."

Tips & Limitations

• Tip: Always enforce the 'Output' layer in your prompt contracts. If you don't explicitly define a schema, the agent will default to narrative text, which is harder to parse programmatically.
• Tip: Treat the Identity layer as a constraint. A 'Senior Lead' persona will consistently produce higher-quality logic than a vague 'Assistant' role.
• Limitation: Recursive agent spawning can consume significant compute resources. Always implement a 'max-depth' constraint in your agent-to-agent instructions to prevent infinite loops and runaway costs.