swarm-advanced

Orchestrating AI agents across distributed systems involves initializing a swarm topology (mesh, hierarchical, star, or ring), spawning specialized agents, and coordinating tasks via MCP APIs like swarm_init, agent_spawn, and task_orchestrate. This Skill provides templates and patterns to manage parallel workflows, fault tolerance, and cross-session memory persistence across agent networks.

Mesh, hierarchical, star, and ring topologies each optimize different workflows. Mesh topologies parallelize information gathering across independent agents; hierarchical topologies delegate tasks through a command chain; star topologies route all work through a central coordinator; ring topologies pass tasks sequentially. Choose based on your coordination complexity and task dependencies.

Memory management in agent swarms uses MCP APIs like memory_usage and workflow_execute to persist state across sessions and coordinate knowledge between agents. This Skill implements fault-tolerant mechanisms that ensure agents retain learned patterns and cross-agent insights, enabling continuous operations without losing context between tasks.

Yes. This Skill applies workflow automation to distributed testing, research, and development by spawning task-specialized agents, orchestrating parallel and sequential execution, and monitoring results via swarm_status and task_orchestrate MCP APIs. Adaptive agent strategies dynamically adjust workload distribution based on task complexity.

Prerequisites include the claude-flow dependency and an MCP tooling environment with CLI command access. You need to define your desired topology, prepare task specifications, and understand your system's agent capacity. The Skill handles initialization, lifecycle management, and mesh orchestration once these are in place.

Swarm scale depends on memory management capacity, MCP API throughput, and task coordination overhead. Ring and mesh topologies scale differently; hierarchical topologies may bottleneck at the coordinator. Monitor swarm_status and memory_usage metrics to detect when agent count or task parallelism approaches your infrastructure limits.