Delegate

What This Skill Does

The Delegate skill empowers your OpenClaw agent to act as a sophisticated task manager, efficiently distributing workloads across a hierarchy of sub-agents. By leveraging a structured cost-benefit analysis, the skill ensures that expensive, high-capability models are reserved for truly complex reasoning tasks, while lightweight models handle routine operations. This orchestration layer not only optimizes your token expenditure but also enhances output reliability through integrated error recovery, automated verification protocols, and clear output constraints.

Installation

You can integrate this skill into your environment by running the following command in your terminal: clawhub install openclaw/skills/skills/ivangdavila/delegate

Use Cases

• Complex Project Management: Breaking down architectural design documents into sub-tasks that are delegated to specialized model tiers based on difficulty.
• Data Pipelines: Automating the extraction, cleaning, and formatting of large datasets by routing initial discovery to small models and final synthesis to medium or large models.
• Code Quality Assurance: Delegating the writing of unit tests to a medium-tier sub-agent while keeping the primary agent focused on high-level feature integration.
• Content Production: Managing a team of agents where one drafts content, another performs SEO optimization, and a final agent verifies factual accuracy.

Example Prompts

1. "Delegate the task of summarizing the attached whitepaper to a small-tier agent. Once finished, have a medium-tier agent write a technical critique of the methodology."
2. "I need to refactor this entire codebase. Please delegate the individual component tests to sub-agents and escalate to the 'Large' tier only if the tests fail twice."
3. "Research the latest trends in renewable energy. Delegate the search and classification to a small-tier model, and use the results to prepare a summary report for me."

Tips & Limitations

To maximize the efficiency of the Delegate skill, always adhere to the core rule: if the overhead of spawning a sub-agent outweighs the cost of the task itself, perform it manually. The skill is best suited for asynchronous, well-defined tasks. It is not recommended to delegate tasks that require continuous conversation or nuance from your current session's memory, as sub-agents lack the broader context of the primary interaction. Always perform a spot-check on delegated output—while the agent is designed to verify, human oversight remains the final line of defense for critical operations. When encountering errors, utilize the built-in escalation path: Small to Medium to Large to ensure task success before manually intervening.