orchestrate

djgrant's avatarfrom djgrant

When you are running any orchestration operation, load this skill first to understand the core methodology using sub-agents and work packages.

⭐2stars🔀0forks📁View on GitHub🕐Updated Jan 8, 2026⬇️1downloads

When & Why to Use This Skill

This Claude skill provides a robust methodology for task orchestration and multi-agent management. It establishes a standardized framework for defining success criteria, delegating work packages to specialized sub-agents, and implementing advanced strategies like map-reduce to solve complex problems efficiently while maintaining high-level oversight and quality control.

Use Cases

  • Complex Project Decomposition: Breaking down large-scale, multifaceted requirements into structured work packages that can be handled by specialized sub-agents.
  • Multi-Agent Workflow Management: Implementing sophisticated orchestration strategies (e.g., map-reduce) to coordinate multiple AI instances toward a unified goal.
  • Quality Assurance & Validation: Establishing iterative feedback loops where sub-agents validate hypotheses and test results before final reporting.
  • Strategic Delegation: Communicating a high-level vision to sub-agents and managing their output without the need for the primary agent to perform manual implementation or coding.
nameorchestrate
descriptionWhen orchestrating tasks, load this core skill first to understand the general methodology.

Commission and track work packages. Delegate to sub-agents. Report results.

::workflow::

DO

  1. Understand the requirements
  2. Define a success criteria
  3. Create a master work package
  4. Communicate the user's vision clearly to sub-agents
  5. Trust sub-agents to solve problems, rather than over-prescribing tasks
  6. Use sub-agents to create scoped work packages
  7. Choose an orchestration strategy (map-reduce or ad-hoc)
  8. Use the appropriate sub-agent for each task
  9. Request sub-agents load skills as appropriate
  10. Ensure work has been tested and hypothesis are validaed
  11. Iterate until success or diminishing returns (up to 5 times)
  12. Report findings
  13. Report failure truthfully END

DO NOT

  • write code or edit files
  • implement solutions yourself END