When is it better to work directly instead of spawning an agent?

You should work directly for simple single tasks (like adding a line of code) or tasks with sequential dependencies where one step relies on the file changes of a previous step to avoid race conditions.

Can I run multiple agents at the same time?

Yes, by making multiple Task tool calls in a single message, you can trigger parallel execution. You can also use the run_in_background flag to continue working while agents process tasks.

How can I save tokens using agent patterns?

Use the Agent Summary Pattern: let an agent perform an extensive search and return a one-line summary. This prevents the main conversation from becoming bloated with exploration logs, saving up to 90% in tokens.

What is the primary benefit of using sub-agents in Claude Code?

The main benefits are parallel execution, which dramatically speeds up broad tasks like codebase searches, and token efficiency, as agents only load the specific context required for their narrow task.

Agent Orchestration Patterns

Name: Agent Orchestration Patterns
Author: willsigmon

bywillsigmon

•

Productivity & Workflow

Optimizes AI sub-agent workflows to maximize parallel execution, specialization, and token efficiency in Claude Code.

About

Agent Orchestration Patterns provides a strategic framework for managing sub-agents within the Claude Code environment. It helps users navigate the complexities of agentic workflows by defining when to spawn autonomous agents for parallel processing versus working directly in the main conversation. By implementing specialized agent types—such as Explore, Worker, and Specialist agents—and avoiding common anti-patterns like sequential dependencies or over-spawning, this skill ensures high-speed execution and significant token savings through narrow context loading and intelligent handoff patterns.

Key Features

Strategic spawning logic to balance parallel speed with token overhead
Background task management for long-running asynchronous operations
Token-saving summary and handoff techniques to reduce context costs
Specialized agent roles for security, performance, style, and debugging
Parallel execution patterns for simultaneous multi-module analysis
6 GitHub stars

Use Cases

Scaling bug searches by spawning simultaneous agents across different codebase modules
Conducting multi-disciplinary PR reviews with specialized security and performance agents
Optimizing large-scale refactors by separating analysis, transformation, and validation phases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add willsigmon/sigstack agent-patterns

For use in Claude.ai and ChatGPT

Download Skill

GitHub