Where does the skill store its execution plans?

Task plans and execution states are stored as Markdown files within a .swarm/ directory in your project's root, serving as both the definition and current state of the work.

When should I use the /swarm command?

Invoke /swarm when a task is too large for a single context window or requires multiple interdependent steps that can be safely parallelized to save time.

Can I use different AI models for different sub-tasks?

Yes, the skill supports various worker types including Haiku for fast/cheap tasks, Opus for standard implementation, and Ultrathink for complex architectural logic.

What are the different agent roles within a swarm?

The system uses four specialized roles: Architects for plan generation, Workers for implementation, Auditors for output validation, and Fixers for resolving minor issues found during audits.

How many agents can run at once using the Swarm skill?

The Swarm Execution skill can manage up to 10 parallel worker agents, each running in its own separate context window to maximize efficiency.

Swarm Execution

Name: Swarm Execution
Author: amoscicki

byamoscicki

生産性とワークフロー

Orchestrates parallel multi-agent workflows to decompose and execute complex coding tasks through autonomous dependency-mapped waves.

概要

Swarm Execution is a specialized orchestration skill that transforms complex, multi-step tasks into manageable, dependency-mapped workflows. By leveraging a hierarchy of specialized agents—Architects for planning, Workers for implementation, and Auditors for quality control—it enables parallel execution of sub-tasks without overloading a single context window. This skill is ideal for large-scale refactoring, migrations, or any scenario requiring simultaneous updates across multiple files while maintaining high standards through automated auditing and fixing loops.

主な機能

Incremental plan management stored in .swarm/ for tracking execution state and checkpoints
Dependency-mapped 'wave' execution to ensure tasks are completed in the correct order
Role-based workflow featuring specialized Architect, Worker, Auditor, and Fixer agents
Parallel agent execution with up to 10 concurrent context windows
0 GitHub stars
Automated quality assurance via dedicated Auditor agents and auto-fixing capabilities

ユースケース

Large-scale codebase refactoring or complex migration projects
Rapid boilerplate generation and configuration across multiple directories
Complex feature implementations requiring synchronized changes in several modules

概要

主な機能

Incremental plan management stored in .swarm/ for tracking execution state and checkpoints
Dependency-mapped 'wave' execution to ensure tasks are completed in the correct order
Role-based workflow featuring specialized Architect, Worker, Auditor, and Fixer agents
Parallel agent execution with up to 10 concurrent context windows
0 GitHub stars
Automated quality assurance via dedicated Auditor agents and auto-fixing capabilities

ユースケース

Large-scale codebase refactoring or complex migration projects
Rapid boilerplate generation and configuration across multiple directories
Complex feature implementations requiring synchronized changes in several modules