What is the benefit of spawning subagents?

Subagents provide context isolation, preventing your main conversation from becoming cluttered. They also allow for parallelization of tasks and cost optimization by using smaller models for simpler objectives.

Is there a limit to how many subagents can be spawned?

While you can spawn up to 5 parallel subagents for independent tasks, it is recommended to avoid spawning subagents for one-liner tasks to prevent unnecessary coordination overhead.

What information do subagents receive?

Subagents receive the specific prompt provided during spawning, global agent rules, and access to the same tools as the parent, but they do not see the parent's full conversation history.

How does model selection work for subagents?

The system follows a hierarchy where trivial or simple tasks use Haiku, medium-to-complex tasks use Sonnet, and only the orchestrator or very complex tasks use Opus. Agents can self-downgrade to save costs.

Can subagents perform git operations?

Yes, there is a specialized git-agent subagent type designed specifically for handling git operations using the Haiku model for maximum efficiency.

Agent Swarm Spawner

Name: Agent Swarm Spawner
Author: c-daly

byc-daly

0•

生産性とワークフロー

Orchestrates specialized subagents to handle complex workflows through parallelization, context isolation, and cost-efficient model selection.

The Spawn skill empowers Claude Code to operate as an orchestrator, delegating specific development tasks to a 'swarm' of specialized subagents such as Explorers, Architects, and Reviewers. By isolating subagent work within their own contexts, the skill prevents the main conversation from becoming bloated, while its intelligent model-selection logic automatically downgrades simpler tasks to cheaper models like Haiku to optimize performance and cost. This skill is ideal for complex, multi-step engineering tasks that require simultaneous operations or deep focus on specific code modules.

主な機能

01Recursive spawning capabilities for large-scale task decomposition

02Context isolation to prevent main thread token overflow

03Parallel spawning for simultaneous execution of independent tasks

040 GitHub stars

05Specialized agent roles for exploration, research, architecture, and implementation

06Dynamic model selection to optimize for cost and task complexity

ユースケース

01Automating multi-file code reviews and security audits without cluttering the main workspace

02Searching and indexing large codebases using parallel Explorer agents

03Delegating routine code implementations to low-cost models while maintaining high-level oversight

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add c-daly/agent-swarm spawn

For use in Claude.ai and ChatGPT

Download Skill