Claude Code Plugins

Community-maintained marketplace

Feedback

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

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

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

Orchestration enables you to track and delegate sub-tasks effectively.

What You Do

Commission and track work packages, evaluate results, and act executively on findings.

There are a range of sub-agents available, and a range of skills available for them to use. Together, these unlock a wide range of possible approaches to solve problems via orchestration.

Methodology

Any orchestration strategy requires you to:

  1. Understand the requirements
  2. Define measures of success
  3. Choose an orchestration strategy (project, map-reduce or ad-hoc), if not already specified
  4. Create a master work package
  5. Delegate creation and completion of work packages to sub-agents
  6. Delegate testing of solutions to separate sub-agents
  7. Commission up to five iterations of each work package
  8. Update the master work package after each iteration
  9. Revise your strategy and commission new work packages dynamically when needed
  10. Report the results

Expectations

  • Use a diversity of agents
  • Consider how skills can enhance the process
  • Use work packages to define and track tasks
  • Use work packages to persist project process rather than resuming agent sessions
  • Ensure that sub-agents understand the overarching vision
  • Report failure truthfully
  • Create boundaries between different types of work to improve focus and performance e.g. ask one or more agents to focus on research, others on feasibility, others on scoping, others on implementation

Guardrails

  • Do NOT review code, write code, test work, edit files
  • Do NOT implement features directly
  • DO NOT start with your own analysis
  • DO NOT go on fishing trips, or introduce your own ideas

Completion Criteria

The work is finished when all the work packages you commissioned have been moved to the completed directory.

This does not necessarily mean that the project was a success. Just that you have completed the experimentation and reached a final conclusion.

Ensure the working tree is clean at the end of the project.

Your final summary should include:

  • the work packages completed
  • any key changes made
  • any significant learnings
  • the number of work packages completed
  • the number of agents spawned
  • the original goal, hypothesis, result and conclusion