Do agents need to maintain state between tasks?

No, agents are designed to be stateless workers. All critical state is maintained via Git branches and a minimal local state file, allowing tasks to be reassigned to any available agent if one goes offline.

What happens if an agent fails during a task execution?

Failures are isolated. If a task fails, only its dependent tasks are blocked. The orchestrator allows you to retry the specific failed task or skip it to unblock the rest of the queue, while independent tasks continue to run.

How does Swarm Orchestration handle task dependencies?

The skill builds a Directed Acyclic Graph (DAG) from your manifest. Tasks with no dependencies run immediately in parallel, while tasks with dependencies wait until their parent tasks have successfully produced Pull Requests.

When should I use Swarm Orchestration versus a standard /autotask?

Use Swarm Orchestration when you have three or more independent tasks, a full sprint's worth of work, or a large refactoring project. Use /autotask for single, isolated code changes or exploratory work.

Swarm Orchestration

Name: Swarm Orchestration
Author: Light-Brands

byLight-Brands

0•

Productivity & Workflow

Coordinates and scales parallel AI development tasks across distributed agents using declarative manifests and dependency graphs.

This skill empowers developers to scale their Claude Code workflows by distributing multiple independent or dependent tasks across a fleet of stateless agents. It manages the complex logic of work manifests, DAG-based dependency resolution, and failure recovery, ensuring that large-scale refactorings or sprint items are processed in parallel with high reliability. By utilizing Git branches as the source of truth, it maintains clean execution states and enables seamless integration with CI/CD pipelines or local multi-agent environments to drastically reduce development cycle times.

Key Features

01Declarative YAML manifests for defining multi-task development workflows

02Parallel task distribution across multiple stateless agents or virtual machines

03Git-centric state management using branches as the primary source of truth

04Automated DAG-based dependency resolution for intelligent execution ordering

050 GitHub stars

06Robust failure isolation and recovery with per-task retry and skip logic

Use Cases

01Executing high-volume sprint tasks across parallel agents to accelerate delivery

02Managing large-scale codebase refactorings split across multiple modules or microservices

03Automating complex CI/CD workflows that require parallel AI-driven code updates

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add light-brands/ntx swarm-orchestration

For use in Claude.ai and ChatGPT

Download Skill