How do I structure a prompt for a task agent?

Use structured XML tags including for background, for explicit instructions, and to define how the agent should return results.

When should I use a sub-agent instead of the main conversation?

Delegate to a sub-agent when tasks produce verbose output (like logs or test runs), are self-contained, or can be run in parallel to preserve the main conversation context.

What is the 'Fan-Out' pattern in agent orchestration?

The Fan-Out pattern involves the main agent launching multiple sub-agents simultaneously to perform the same operation (like auditing or refactoring) across different files.

Which model should I use for different task agents?

Use 'haiku' for fast, simple searches; 'sonnet' for standard implementation tasks; and 'opus' for complex reasoning, architectural planning, or high-accuracy requirements.

Can sub-agents use other Claude Code skills?

Yes, but they must be explicitly invoked using the Skill tool syntax provided by this orchestrator, as sub-agents do not automatically trigger skills by name alone.

Agent Orchestrator & Task Delegation

Name: Agent Orchestrator & Task Delegation
Author: adbutterfield

byadbutterfield

0•

Productivity & Workflow

Manages the delegation, coordination, and execution of sub-agents for complex parallel or sequential coding tasks.

This skill optimizes the workflow between a main Claude agent and its specialized sub-agents, including Explore, Plan, Bash, and general-purpose types. It provides a standardized framework for drafting explicit XML-based prompts, selecting the right model for cost-efficiency, and implementing advanced coordination patterns like Fan-Out or Sequential execution. By enforcing explicit skill invocation syntax and structured context passing, it ensures that delegated tasks are completed accurately without manual intervention, preserving the main conversation's context for high-level reasoning.

Key Features

01Background task management and agent resumption strategies

02Coordination patterns for parallel, sequential, and fan-out workflows

030 GitHub stars

04XML-structured prompting for precise sub-agent control

05Explicit syntax for enabling sub-agent skill invocation

06Strategic agent type and model selection (Haiku, Sonnet, Opus)

Use Cases

01Running parallel independent investigations across a large codebase to save time

02Executing complex features through a multi-stage Research, Plan, and Implement chain

03Batch processing multiple files for migration or refactoring using the Fan-Out pattern

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add adbutterfield/coding-agent-plugins orchestrating-task-agents

For use in Claude.ai and ChatGPT

Download Skill